prompt
stringlengths 131
11.8k
| completion
stringlengths 7
173
| api
stringlengths 11
48
|
|---|---|---|
get_ipython().run_line_magic('load_ext', 'autoreload')
get_ipython().run_line_magic('autoreload', '2')
import dspy
from dspy.evaluate import Evaluate
from dspy.datasets.hotpotqa import HotPotQA
from dspy.teleprompt import BootstrapFewShotWithRandomSearch, BootstrapFinetune
ports = [7140, 7141, 7142, 7143, 7144, 7145]
llamaChat = dspy.HFClientTGI(model="meta-llama/Llama-2-13b-chat-hf", port=ports, max_tokens=150)
colbertv2 = dspy.ColBERTv2(url='http://20.102.90.50:2017/wiki17_abstracts')
dspy.settings.configure(rm=colbertv2, lm=llamaChat)
dataset = HotPotQA(train_seed=1, train_size=200, eval_seed=2023, dev_size=1000, test_size=0)
trainset = [x.with_inputs('question') for x in dataset.train]
devset = [x.with_inputs('question') for x in dataset.dev]
testset = [x.with_inputs('question') for x in dataset.test]
len(trainset), len(devset), len(testset)
trainset[0]
from dsp.utils.utils import deduplicate
class BasicMH(dspy.Module):
def __init__(self, passages_per_hop=3):
super().__init__()
self.retrieve = dspy.Retrieve(k=passages_per_hop)
self.generate_query = [dspy.ChainOfThought("context, question -> search_query") for _ in range(2)]
self.generate_answer = dspy.ChainOfThought("context, question -> answer")
def forward(self, question):
context = []
for hop in range(2):
search_query = self.generate_query[hop](context=context, question=question).search_query
passages = self.retrieve(search_query).passages
context = deduplicate(context + passages)
return self.generate_answer(context=context, question=question).copy(context=context)
RECOMPILE_INTO_LLAMA_FROM_SCRATCH = False
NUM_THREADS = 24
metric_EM = dspy.evaluate.answer_exact_match
if RECOMPILE_INTO_LLAMA_FROM_SCRATCH:
tp = BootstrapFewShotWithRandomSearch(metric=metric_EM, max_bootstrapped_demos=2, num_threads=NUM_THREADS)
basicmh_bs = tp.compile(BasicMH(), trainset=trainset[:50], valset=trainset[50:200])
ensemble = [prog for *_, prog in basicmh_bs.candidate_programs[:4]]
for idx, prog in enumerate(ensemble):
pass
if not RECOMPILE_INTO_LLAMA_FROM_SCRATCH:
ensemble = []
for idx in range(4):
prog = BasicMH()
prog.load(f'multihop_llama213b_{idx}.json')
ensemble.append(prog)
llama_program = ensemble[0]
evaluate_hotpot = Evaluate(devset=devset[:1000], metric=metric_EM, num_threads=NUM_THREADS, display_progress=True, display_table=0)
evaluate_hotpot(llama_program)
llama_program(question="How many storeys are in the castle that David Gregory inherited?")
llamaChat.inspect_history(n=3)
unlabeled_train = HotPotQA(train_seed=1, train_size=3000, eval_seed=2023, dev_size=0, test_size=0).train
unlabeled_train = [ | dspy.Example(question=x.question) | dspy.Example |
get_ipython().system('pip install clarifai')
get_ipython().system('pip install dspy-ai')
import dspy
from dspy.retrieve.clarifai_rm import ClarifaiRM
MODEL_URL = "https://clarifai.com/meta/Llama-2/models/llama2-70b-chat"
PAT = "CLARIFAI_PAT"
USER_ID = "YOUR_ID"
APP_ID = "YOUR_APP"
from langchain.text_splitter import CharacterTextSplitter
from langchain.document_loaders import TextLoader
from langchain.vectorstores import Clarifai as clarifaivectorstore
loader = TextLoader("YOUR_TEXT_FILE") #replace with your file path
documents = loader.load()
text_splitter = CharacterTextSplitter(chunk_size=1024, chunk_overlap=200)
docs = text_splitter.split_documents(documents)
clarifai_vector_db = clarifaivectorstore.from_documents(
user_id=USER_ID,
app_id=APP_ID,
documents=docs,
pat=PAT
)
llm=dspy.Clarifai(model=MODEL_URL, api_key=PAT, n=2, inference_params={"max_tokens":100,'temperature':0.6})
retriever_model=ClarifaiRM(clarifai_user_id=USER_ID, clarfiai_app_id=APP_ID, clarifai_pat=PAT, k=2)
dspy.settings.configure(lm=llm, rm=retriever_model)
sentence = "disney again ransacks its archives for a quick-buck sequel ." # example from the SST-2 dataset.
classify = dspy.Predict('sentence -> sentiment')
print(classify(sentence=sentence).sentiment)
retrieve = dspy.Retrieve()
topK_passages = retrieve("can I test my vehicle engine in pit?").passages
print(topK_passages)
class GenerateAnswer(dspy.Signature):
"""Think and Answer questions based on the context provided."""
context = dspy.InputField(desc="may contain relevant facts about user query")
question = dspy.InputField(desc="User query")
answer = dspy.OutputField(desc="Answer in one or two lines")
class RAG(dspy.Module):
def __init__(self):
super().__init__()
self.retrieve = dspy.Retrieve()
self.generate_answer = dspy.ChainOfThought(GenerateAnswer)
def forward(self, question):
context = self.retrieve(question).passages
prediction = self.generate_answer(context=context, question=question)
return dspy.Prediction(context=context, answer=prediction.answer)
my_question = "can I test my vehicle engine in pit before inspection?"
Rag_obj= RAG()
predict_response_llama70b=Rag_obj(my_question)
print(f"Question: {my_question}")
print(f"Predicted Answer: {predict_response_llama70b.answer}")
print(f"Retrieved Contexts (truncated): {[c[:200] + '...' for c in predict_response_llama70b.context]}")
mistral_lm = dspy.Clarifai(model="https://clarifai.com/mistralai/completion/models/mistral-7B-Instruct", api_key=PAT, n=2, inference_params={'temperature':0.6})
dspy.settings.configure(lm=mistral_lm, rm=retriever_model)
my_question = "can I test my vehicle engine in pit before inspection?"
Rag_obj= RAG()
predict_response_mistral=Rag_obj(my_question)
print(f"Question: {my_question}")
print(f"Predicted Answer: {predict_response_mistral.answer}")
print(f"Retrieved Contexts (truncated): {[c[:200] + '...' for c in predict_response_mistral.context]}")
gemini_lm = dspy.Clarifai(model="https://clarifai.com/gcp/generate/models/gemini-pro", api_key=PAT, n=2)
| dspy.settings.configure(lm=gemini_lm, rm=retriever_model) | dspy.settings.configure |
get_ipython().run_line_magic('load_ext', 'autoreload')
get_ipython().run_line_magic('autoreload', '2')
import sys
import os
try: # When on google Colab, let's clone the notebook so we download the cache.
import google.colab
repo_path = 'dspy'
get_ipython().system('git -C $repo_path pull origin || git clone https://github.com/stanfordnlp/dspy $repo_path')
except:
repo_path = '.'
if repo_path not in sys.path:
sys.path.append(repo_path)
os.environ["DSP_NOTEBOOK_CACHEDIR"] = os.path.join(repo_path, 'cache')
import dspy
get_ipython().run_line_magic('load_ext', 'autoreload')
get_ipython().run_line_magic('autoreload', '2')
import sys; sys.path.append('/future/u/okhattab/repos/public/stanfordnlp/dspy')
import dspy
from dspy.evaluate import Evaluate
from dspy.datasets.hotpotqa import HotPotQA
from dspy.teleprompt import BootstrapFewShot, BootstrapFewShotWithRandomSearch, BootstrapFinetune
llama = dspy.HFClientTGI(model="meta-llama/Llama-2-13b-chat-hf", port=[7140, 7141, 7142, 7143], max_tokens=150)
colbertv2 = dspy.ColBERTv2(url='http://20.102.90.50:2017/wiki17_abstracts')
dspy.settings.configure(rm=colbertv2, lm=llama)
train = [('Who was the director of the 2009 movie featuring Peter Outerbridge as William Easton?', 'Kevin Greutert'),
('The heir to the Du Pont family fortune sponsored what wrestling team?', 'Foxcatcher'),
('In what year was the star of To Hell and Back born?', '1925'),
('Which award did the first book of Gary Zukav receive?', 'U.S. National Book Award'),
('What documentary about the Gilgo Beach Killer debuted on A&E?', 'The Killing Season'),
('Which author is English: John Braine or Studs Terkel?', 'John Braine'),
('Who produced the album that included a re-recording of "Lithium"?', 'Butch Vig')]
train = [dspy.Example(question=question, answer=answer).with_inputs('question') for question, answer in train]
dev = [('Who has a broader scope of profession: E. L. Doctorow or Julia Peterkin?', 'E. L. Doctorow'),
('Right Back At It Again contains lyrics co-written by the singer born in what city?', 'Gainesville, Florida'),
('What year was the party of the winner of the 1971 San Francisco mayoral election founded?', '1828'),
('Anthony Dirrell is the brother of which super middleweight title holder?', 'Andre Dirrell'),
('The sports nutrition business established by Oliver Cookson is based in which county in the UK?', 'Cheshire'),
('Find the birth date of the actor who played roles in First Wives Club and Searching for the Elephant.', 'February 13, 1980'),
('Kyle Moran was born in the town on what river?', 'Castletown River'),
("The actress who played the niece in the Priest film was born in what city, country?", 'Surrey, England'),
('Name the movie in which the daughter of Noel Harrison plays Violet Trefusis.', 'Portrait of a Marriage'),
('What year was the father of the Princes in the Tower born?', '1442'),
('What river is near the Crichton Collegiate Church?', 'the River Tyne'),
('Who purchased the team Michael Schumacher raced for in the 1995 Monaco Grand Prix in 2000?', 'Renault'),
('André Zucca was a French photographer who worked with a German propaganda magazine published by what Nazi organization?', 'the Wehrmacht')]
dev = [dspy.Example(question=question, answer=answer).with_inputs('question') for question, answer in dev]
predict = dspy.Predict('question -> answer')
predict(question="What is the capital of Germany?")
class CoT(dspy.Module): # let's define a new module
def __init__(self):
super().__init__()
self.generate_answer = dspy.ChainOfThought('question -> answer')
def forward(self, question):
return self.generate_answer(question=question) # here we use the module
metric_EM = dspy.evaluate.answer_exact_match
teleprompter = BootstrapFewShot(metric=metric_EM, max_bootstrapped_demos=2)
cot_compiled = teleprompter.compile(CoT(), trainset=train)
cot_compiled("What is the capital of Germany?")
llama.inspect_history(n=1)
NUM_THREADS = 32
evaluate_hotpot = | Evaluate(devset=dev, metric=metric_EM, num_threads=NUM_THREADS, display_progress=True, display_table=15) | dspy.evaluate.Evaluate |
import dspy
from dsp.utils import deduplicate
from dspy.datasets import HotPotQA
from dspy.predict.retry import Retry
from dspy.teleprompt import BootstrapFewShot, BootstrapFewShotWithRandomSearch
from dspy.evaluate.evaluate import Evaluate
from dspy.primitives.assertions import assert_transform_module, backtrack_handler
import os
import openai
openai.api_key = os.getenv('OPENAI_API_KEY')
colbertv2_wiki17_abstracts = dspy.ColBERTv2(url='http://20.102.90.50:2017/wiki17_abstracts')
dspy.settings.configure(rm=colbertv2_wiki17_abstracts)
turbo = dspy.OpenAI(model='gpt-3.5-turbo', max_tokens=500)
| dspy.settings.configure(lm=turbo, trace=[], temperature=0.7) | dspy.settings.configure |
get_ipython().system('git clone https://huggingface.co/arnavs11/DSPy_TweetGen_Cache')
get_ipython().run_line_magic('cd', 'DSPy_TweetGen_Cache/')
get_ipython().system('git checkout master')
get_ipython().run_line_magic('cd', '..')
import os
repo_clone_path = '/content/DSPy_TweetGen_Cache'
if not os.access('/content', os.W_OK):
repo_clone_path = os.path.join(os.getcwd(), 'DSPy_TweetGen_Cache')
os.environ["DSP_NOTEBOOK_CACHEDIR"] = repo_clone_path
get_ipython().run_line_magic('load_ext', 'autoreload')
get_ipython().run_line_magic('autoreload', '2')
import sys
import os
import regex as re
import json
try: # When on google Colab, let's clone the notebook so we download the cache.
import google.colab
repo_path = 'dspy'
get_ipython().system('git -C $repo_path pull origin || git clone https://github.com/stanfordnlp/dspy $repo_path')
except:
repo_path = '.'
if repo_path not in sys.path:
sys.path.append(repo_path)
import pkg_resources # Install the package if it's not installed
if not "dspy-ai" in {pkg.key for pkg in pkg_resources.working_set}:
get_ipython().system('pip install -U pip')
get_ipython().system('pip install dspy-ai')
get_ipython().system('pip install openai~=0.28.1')
get_ipython().system('pip install -e $repo_path')
import dspy
from dspy.predict import Retry
from dspy.datasets import HotPotQA
from dspy.teleprompt import BootstrapFewShotWithRandomSearch
from dsp.utils import deduplicate
from dspy.evaluate.evaluate import Evaluate
from dspy.primitives.assertions import assert_transform_module, backtrack_handler
colbertv2_wiki17_abstracts = dspy.ColBERTv2(url='http://20.102.90.50:2017/wiki17_abstracts')
dspy.settings.configure(rm=colbertv2_wiki17_abstracts)
turbo = dspy.OpenAI(model='gpt-3.5-turbo', max_tokens=500)
dspy.settings.configure(lm=turbo, trace=[], temperature=0.7)
dataset = HotPotQA(train_seed=1, train_size=300, eval_seed=2023, dev_size=300, test_size=0, keep_details=True)
trainset = [x.with_inputs('question', 'answer') for x in dataset.train]
devset = [x.with_inputs('question', 'answer') for x in dataset.dev]
class GenerateSearchQuery(dspy.Signature):
"""Write a simple search query that will help answer a complex question."""
context = dspy.InputField(desc="may contain relevant facts")
question = dspy.InputField()
query = dspy.OutputField()
class GenerateTweet(dspy.Signature):
"""Generate an engaging tweet that effectively answers a question staying faithful to the context, is less than 280 characters, and has no hashtags."""
question = dspy.InputField()
context = dspy.InputField(desc="may contain relevant facts")
tweet = dspy.OutputField()
class Tweeter(dspy.Module):
def __init__(self):
super().__init__()
self.generate_tweet = dspy.ChainOfThought(GenerateTweet)
def forward(self, question, answer):
context = []
max_hops=2
passages_per_hop=3
generate_query = [dspy.ChainOfThought(GenerateSearchQuery) for _ in range(max_hops)]
retrieve = dspy.Retrieve(k=passages_per_hop)
for hop in range(max_hops):
query = generate_query[hop](context=context, question=question).query
passages = retrieve(query).passages
context = deduplicate(context + passages)
generated_tweet = self.generate_tweet(question=question, context=context).tweet
return dspy.Prediction(generated_tweet=generated_tweet, context=context)
tweeter = Tweeter()
def has_no_hashtags(text):
return len(re.findall(r"#\w+", text)) == 0
def is_within_length_limit(text, length_limit=280):
return len(text) <= length_limit
def is_assessment_yes(assessment_answer):
"""Check if the first word of the assessment answer is 'yes'."""
return assessment_answer.split()[0].lower() == 'yes'
def has_correct_answer(text, answer):
return answer in text
class AssessTweet(dspy.Signature):
"""Assess the quality of a tweet along the specified dimension."""
context = dspy.InputField(desc='ignore if N/A')
assessed_text = dspy.InputField()
assessment_question = dspy.InputField()
assessment_answer = dspy.OutputField(desc="Yes or No")
def no_hashtags_metric(gold, pred, trace=None):
tweet = pred.generated_tweet
no_hashtags = has_no_hashtags(tweet)
score = no_hashtags
return score
def is_correct_metric(gold, pred, trace=None):
answer, tweet = gold.answer, pred.generated_tweet
correct = has_correct_answer(tweet, answer)
score = correct
return score
def within_length_metric(gold, pred, trace=None):
tweet = pred.generated_tweet
within_length_limit = is_within_length_limit(tweet, 280)
score = within_length_limit
return score
def engaging_metric(gold, pred, trace=None):
tweet = pred.generated_tweet
engaging = "Does the assessed text make for a self-contained, engaging tweet? Say no if it is not engaging."
engaging = dspy.Predict(AssessTweet)(context='N/A', assessed_text=tweet, assessment_question=engaging)
engaging = engaging.assessment_answer.split()[0].lower() == 'yes'
score = engaging
return score
def faithful_metric(gold, pred, trace=None):
context, tweet = pred.context, pred.generated_tweet
faithful = "Is the assessed text grounded in the context? Say no if it includes significant facts not in the context."
faithful = dspy.Predict(AssessTweet)(context=context, assessed_text=tweet, assessment_question=faithful)
faithful = faithful.assessment_answer.split()[0].lower() == 'yes'
score = faithful
return score
def overall_metric(gold, pred, trace=None):
answer, context, tweet = gold.answer, pred.context, pred.generated_tweet
no_hashtags = has_no_hashtags(tweet)
within_length_limit = is_within_length_limit(tweet, 280)
correct = has_correct_answer(tweet, answer)
engaging = "Does the assessed text make for a self-contained, engaging tweet? Say no if it is not engaging."
faithful = "Is the assessed text grounded in the context? Say no if it includes significant facts not in the context."
faithful = dspy.Predict(AssessTweet)(context=context, assessed_text=tweet, assessment_question=faithful)
engaging = dspy.Predict(AssessTweet)(context='N/A', assessed_text=tweet, assessment_question=engaging)
engaging, faithful = [m.assessment_answer.split()[0].lower() == 'yes' for m in [engaging, faithful]]
score = (correct + engaging + faithful + no_hashtags + within_length_limit) if correct and within_length_limit else 0
return score / 5.0
metrics = [no_hashtags_metric, is_correct_metric, within_length_metric, engaging_metric, faithful_metric, overall_metric]
for metric in metrics:
evaluate = Evaluate(metric=metric, devset=devset, num_threads=1, display_progress=True, display_table=5)
evaluate(tweeter)
example = devset[10]
tweet = tweeter(question=example.question, answer = example.answer)
print(f'Generated Tweet: ', tweet.generated_tweet)
tweet.context
for metric in metrics:
evaluate = Evaluate(metric=metric, devset=devset[10:11], num_threads=1, display_progress=True, display_table=5)
evaluate(tweeter)
class TweeterWithAssertions(dspy.Module):
def __init__(self):
super().__init__()
self.generate_tweet = dspy.ChainOfThought(GenerateTweet)
def forward(self, question, answer):
context = []
max_hops=2
passages_per_hop=3
generate_query = [dspy.ChainOfThought(GenerateSearchQuery) for _ in range(max_hops)]
retrieve = dspy.Retrieve(k=passages_per_hop)
for hop in range(max_hops):
query = generate_query[hop](context=context, question=question).query
passages = retrieve(query).passages
context = deduplicate(context + passages)
generated_tweet = self.generate_tweet(question=question, context=context).tweet
dspy.Suggest(has_no_hashtags(generated_tweet), f"Please revise the tweet to remove hashtag phrases following it.", target_module=GenerateTweet)
dspy.Suggest(is_within_length_limit(generated_tweet, 280), f"Please ensure the tweet is within {280} characters.", target_module=GenerateTweet)
dspy.Suggest(has_correct_answer(generated_tweet, answer), "The tweet does not include the correct answer to the question. Please revise accordingly.", target_module=GenerateTweet)
engaging_question = "Does the assessed text make for a self-contained, engaging tweet? Say no if it is not engaging."
engaging_assessment = | dspy.Predict(AssessTweet) | dspy.Predict |
get_ipython().system('git clone https://huggingface.co/arnavs11/DSPy_TweetGen_Cache')
get_ipython().run_line_magic('cd', 'DSPy_TweetGen_Cache/')
get_ipython().system('git checkout master')
get_ipython().run_line_magic('cd', '..')
import os
repo_clone_path = '/content/DSPy_TweetGen_Cache'
if not os.access('/content', os.W_OK):
repo_clone_path = os.path.join(os.getcwd(), 'DSPy_TweetGen_Cache')
os.environ["DSP_NOTEBOOK_CACHEDIR"] = repo_clone_path
get_ipython().run_line_magic('load_ext', 'autoreload')
get_ipython().run_line_magic('autoreload', '2')
import sys
import os
import regex as re
import json
try: # When on google Colab, let's clone the notebook so we download the cache.
import google.colab
repo_path = 'dspy'
get_ipython().system('git -C $repo_path pull origin || git clone https://github.com/stanfordnlp/dspy $repo_path')
except:
repo_path = '.'
if repo_path not in sys.path:
sys.path.append(repo_path)
import pkg_resources # Install the package if it's not installed
if not "dspy-ai" in {pkg.key for pkg in pkg_resources.working_set}:
get_ipython().system('pip install -U pip')
get_ipython().system('pip install dspy-ai')
get_ipython().system('pip install openai~=0.28.1')
get_ipython().system('pip install -e $repo_path')
import dspy
from dspy.predict import Retry
from dspy.datasets import HotPotQA
from dspy.teleprompt import BootstrapFewShotWithRandomSearch
from dsp.utils import deduplicate
from dspy.evaluate.evaluate import Evaluate
from dspy.primitives.assertions import assert_transform_module, backtrack_handler
colbertv2_wiki17_abstracts = dspy.ColBERTv2(url='http://20.102.90.50:2017/wiki17_abstracts')
dspy.settings.configure(rm=colbertv2_wiki17_abstracts)
turbo = dspy.OpenAI(model='gpt-3.5-turbo', max_tokens=500)
dspy.settings.configure(lm=turbo, trace=[], temperature=0.7)
dataset = HotPotQA(train_seed=1, train_size=300, eval_seed=2023, dev_size=300, test_size=0, keep_details=True)
trainset = [x.with_inputs('question', 'answer') for x in dataset.train]
devset = [x.with_inputs('question', 'answer') for x in dataset.dev]
class GenerateSearchQuery(dspy.Signature):
"""Write a simple search query that will help answer a complex question."""
context = dspy.InputField(desc="may contain relevant facts")
question = dspy.InputField()
query = dspy.OutputField()
class GenerateTweet(dspy.Signature):
"""Generate an engaging tweet that effectively answers a question staying faithful to the context, is less than 280 characters, and has no hashtags."""
question = dspy.InputField()
context = dspy.InputField(desc="may contain relevant facts")
tweet = dspy.OutputField()
class Tweeter(dspy.Module):
def __init__(self):
super().__init__()
self.generate_tweet = dspy.ChainOfThought(GenerateTweet)
def forward(self, question, answer):
context = []
max_hops=2
passages_per_hop=3
generate_query = [dspy.ChainOfThought(GenerateSearchQuery) for _ in range(max_hops)]
retrieve = dspy.Retrieve(k=passages_per_hop)
for hop in range(max_hops):
query = generate_query[hop](context=context, question=question).query
passages = retrieve(query).passages
context = deduplicate(context + passages)
generated_tweet = self.generate_tweet(question=question, context=context).tweet
return dspy.Prediction(generated_tweet=generated_tweet, context=context)
tweeter = Tweeter()
def has_no_hashtags(text):
return len(re.findall(r"#\w+", text)) == 0
def is_within_length_limit(text, length_limit=280):
return len(text) <= length_limit
def is_assessment_yes(assessment_answer):
"""Check if the first word of the assessment answer is 'yes'."""
return assessment_answer.split()[0].lower() == 'yes'
def has_correct_answer(text, answer):
return answer in text
class AssessTweet(dspy.Signature):
"""Assess the quality of a tweet along the specified dimension."""
context = dspy.InputField(desc='ignore if N/A')
assessed_text = dspy.InputField()
assessment_question = dspy.InputField()
assessment_answer = dspy.OutputField(desc="Yes or No")
def no_hashtags_metric(gold, pred, trace=None):
tweet = pred.generated_tweet
no_hashtags = has_no_hashtags(tweet)
score = no_hashtags
return score
def is_correct_metric(gold, pred, trace=None):
answer, tweet = gold.answer, pred.generated_tweet
correct = has_correct_answer(tweet, answer)
score = correct
return score
def within_length_metric(gold, pred, trace=None):
tweet = pred.generated_tweet
within_length_limit = is_within_length_limit(tweet, 280)
score = within_length_limit
return score
def engaging_metric(gold, pred, trace=None):
tweet = pred.generated_tweet
engaging = "Does the assessed text make for a self-contained, engaging tweet? Say no if it is not engaging."
engaging = dspy.Predict(AssessTweet)(context='N/A', assessed_text=tweet, assessment_question=engaging)
engaging = engaging.assessment_answer.split()[0].lower() == 'yes'
score = engaging
return score
def faithful_metric(gold, pred, trace=None):
context, tweet = pred.context, pred.generated_tweet
faithful = "Is the assessed text grounded in the context? Say no if it includes significant facts not in the context."
faithful = dspy.Predict(AssessTweet)(context=context, assessed_text=tweet, assessment_question=faithful)
faithful = faithful.assessment_answer.split()[0].lower() == 'yes'
score = faithful
return score
def overall_metric(gold, pred, trace=None):
answer, context, tweet = gold.answer, pred.context, pred.generated_tweet
no_hashtags = has_no_hashtags(tweet)
within_length_limit = is_within_length_limit(tweet, 280)
correct = has_correct_answer(tweet, answer)
engaging = "Does the assessed text make for a self-contained, engaging tweet? Say no if it is not engaging."
faithful = "Is the assessed text grounded in the context? Say no if it includes significant facts not in the context."
faithful = | dspy.Predict(AssessTweet) | dspy.Predict |
get_ipython().system('pip install clarifai')
get_ipython().system('pip install dspy-ai')
import dspy
from dspy.retrieve.clarifai_rm import ClarifaiRM
MODEL_URL = "https://clarifai.com/meta/Llama-2/models/llama2-70b-chat"
PAT = "CLARIFAI_PAT"
USER_ID = "YOUR_ID"
APP_ID = "YOUR_APP"
from langchain.text_splitter import CharacterTextSplitter
from langchain.document_loaders import TextLoader
from langchain.vectorstores import Clarifai as clarifaivectorstore
loader = TextLoader("YOUR_TEXT_FILE") #replace with your file path
documents = loader.load()
text_splitter = CharacterTextSplitter(chunk_size=1024, chunk_overlap=200)
docs = text_splitter.split_documents(documents)
clarifai_vector_db = clarifaivectorstore.from_documents(
user_id=USER_ID,
app_id=APP_ID,
documents=docs,
pat=PAT
)
llm=dspy.Clarifai(model=MODEL_URL, api_key=PAT, n=2, inference_params={"max_tokens":100,'temperature':0.6})
retriever_model=ClarifaiRM(clarifai_user_id=USER_ID, clarfiai_app_id=APP_ID, clarifai_pat=PAT, k=2)
dspy.settings.configure(lm=llm, rm=retriever_model)
sentence = "disney again ransacks its archives for a quick-buck sequel ." # example from the SST-2 dataset.
classify = dspy.Predict('sentence -> sentiment')
print(classify(sentence=sentence).sentiment)
retrieve = dspy.Retrieve()
topK_passages = retrieve("can I test my vehicle engine in pit?").passages
print(topK_passages)
class GenerateAnswer(dspy.Signature):
"""Think and Answer questions based on the context provided."""
context = dspy.InputField(desc="may contain relevant facts about user query")
question = | dspy.InputField(desc="User query") | dspy.InputField |
get_ipython().system('git clone https://huggingface.co/arnavs11/DSPy_TweetGen_Cache')
get_ipython().run_line_magic('cd', 'DSPy_TweetGen_Cache/')
get_ipython().system('git checkout master')
get_ipython().run_line_magic('cd', '..')
import os
repo_clone_path = '/content/DSPy_TweetGen_Cache'
if not os.access('/content', os.W_OK):
repo_clone_path = os.path.join(os.getcwd(), 'DSPy_TweetGen_Cache')
os.environ["DSP_NOTEBOOK_CACHEDIR"] = repo_clone_path
get_ipython().run_line_magic('load_ext', 'autoreload')
get_ipython().run_line_magic('autoreload', '2')
import sys
import os
import regex as re
import json
try: # When on google Colab, let's clone the notebook so we download the cache.
import google.colab
repo_path = 'dspy'
get_ipython().system('git -C $repo_path pull origin || git clone https://github.com/stanfordnlp/dspy $repo_path')
except:
repo_path = '.'
if repo_path not in sys.path:
sys.path.append(repo_path)
import pkg_resources # Install the package if it's not installed
if not "dspy-ai" in {pkg.key for pkg in pkg_resources.working_set}:
get_ipython().system('pip install -U pip')
get_ipython().system('pip install dspy-ai')
get_ipython().system('pip install openai~=0.28.1')
get_ipython().system('pip install -e $repo_path')
import dspy
from dspy.predict import Retry
from dspy.datasets import HotPotQA
from dspy.teleprompt import BootstrapFewShotWithRandomSearch
from dsp.utils import deduplicate
from dspy.evaluate.evaluate import Evaluate
from dspy.primitives.assertions import assert_transform_module, backtrack_handler
colbertv2_wiki17_abstracts = dspy.ColBERTv2(url='http://20.102.90.50:2017/wiki17_abstracts')
dspy.settings.configure(rm=colbertv2_wiki17_abstracts)
turbo = | dspy.OpenAI(model='gpt-3.5-turbo', max_tokens=500) | dspy.OpenAI |
get_ipython().run_line_magic('load_ext', 'autoreload')
get_ipython().run_line_magic('autoreload', '2')
import sys
import pkg_resources
try: # When on Colab, let's install pyserini, Pytorch, and Faiss
import google.colab
repo_path = 'dspy'
get_ipython().system('git -C $repo_path pull origin || git clone https://github.com/stanfordnlp/dspy $repo_path')
get_ipython().run_line_magic('cd', '$repo_path')
get_ipython().system('pip install -e .')
if not "pyserini" in {pkg.key for pkg in pkg_resources.working_set}:
get_ipython().system('pip install pyserini')
if not "torch" in {pkg.key for pkg in pkg_resources.working_set}:
get_ipython().system('pip install torch')
if not "faiss-cpu" in {pkg.key for pkg in pkg_resources.working_set}:
get_ipython().system('pip install faiss-cpu')
except:
repo_path = '.'
if not "dspy-ai" in {pkg.key for pkg in pkg_resources.working_set}:
get_ipython().system('pip install -U pip')
get_ipython().system('pip install dspy-ai')
if repo_path not in sys.path:
sys.path.append(repo_path)
import dspy
pys_ret_prebuilt = dspy.Pyserini(index='beir-v1.0.0-nfcorpus.contriever-msmarco', query_encoder='facebook/contriever-msmarco', id_field='_id', text_fields=['title', 'text'])
dspy.settings.configure(rm=pys_ret_prebuilt)
example_question = "How Curry Can Kill Cancer Cells"
retrieve = dspy.Retrieve(k=3)
topK_passages = retrieve(example_question).passages
print(f"Top {retrieve.k} passages for question: {example_question} \n", '-' * 30, '\n')
for idx, passage in enumerate(topK_passages):
print(f'{idx+1}]', passage, '\n')
get_ipython().system('wget https://public.ukp.informatik.tu-darmstadt.de/thakur/BEIR/datasets/nfcorpus.zip -P collections')
get_ipython().system('unzip collections/nfcorpus.zip -d collections')
get_ipython().system('python -m pyserini.encode input --corpus collections/nfcorpus/corpus.jsonl --fields title text output --embeddings indexes/faiss.nfcorpus.contriever-msmarco --to-faiss encoder --encoder facebook/contriever-msmarco --device cuda:0 --pooling mean --fields title text')
from datasets import load_dataset
dataset = load_dataset(path='json', data_files='collections/nfcorpus/corpus.jsonl', split='train')
pys_ret_local = | dspy.Pyserini(index='indexes/faiss.nfcorpus.contriever-msmarco', query_encoder='facebook/contriever-msmarco', dataset=dataset, id_field='_id', text_fields=['title', 'text']) | dspy.Pyserini |
import dspy
from dspy.evaluate import Evaluate
from dspy.datasets.gsm8k import GSM8K, gsm8k_metric
from dspy.teleprompt import BootstrapFewShotWithRandomSearch
gms8k = GSM8K()
turbo = dspy.OpenAI(model='gpt-3.5-turbo-instruct', max_tokens=250)
trainset, devset = gms8k.train, gms8k.dev
dspy.settings.configure(lm=turbo)
NUM_THREADS = 4
evaluate = | Evaluate(devset=devset[:], metric=gsm8k_metric, num_threads=NUM_THREADS, display_progress=True, display_table=0) | dspy.evaluate.Evaluate |
get_ipython().system('git clone https://huggingface.co/arnavs11/DSPy_TweetGen_Cache')
get_ipython().run_line_magic('cd', 'DSPy_TweetGen_Cache/')
get_ipython().system('git checkout master')
get_ipython().run_line_magic('cd', '..')
import os
repo_clone_path = '/content/DSPy_TweetGen_Cache'
if not os.access('/content', os.W_OK):
repo_clone_path = os.path.join(os.getcwd(), 'DSPy_TweetGen_Cache')
os.environ["DSP_NOTEBOOK_CACHEDIR"] = repo_clone_path
get_ipython().run_line_magic('load_ext', 'autoreload')
get_ipython().run_line_magic('autoreload', '2')
import sys
import os
import regex as re
import json
try: # When on google Colab, let's clone the notebook so we download the cache.
import google.colab
repo_path = 'dspy'
get_ipython().system('git -C $repo_path pull origin || git clone https://github.com/stanfordnlp/dspy $repo_path')
except:
repo_path = '.'
if repo_path not in sys.path:
sys.path.append(repo_path)
import pkg_resources # Install the package if it's not installed
if not "dspy-ai" in {pkg.key for pkg in pkg_resources.working_set}:
get_ipython().system('pip install -U pip')
get_ipython().system('pip install dspy-ai')
get_ipython().system('pip install openai~=0.28.1')
get_ipython().system('pip install -e $repo_path')
import dspy
from dspy.predict import Retry
from dspy.datasets import HotPotQA
from dspy.teleprompt import BootstrapFewShotWithRandomSearch
from dsp.utils import deduplicate
from dspy.evaluate.evaluate import Evaluate
from dspy.primitives.assertions import assert_transform_module, backtrack_handler
colbertv2_wiki17_abstracts = dspy.ColBERTv2(url='http://20.102.90.50:2017/wiki17_abstracts')
dspy.settings.configure(rm=colbertv2_wiki17_abstracts)
turbo = dspy.OpenAI(model='gpt-3.5-turbo', max_tokens=500)
dspy.settings.configure(lm=turbo, trace=[], temperature=0.7)
dataset = HotPotQA(train_seed=1, train_size=300, eval_seed=2023, dev_size=300, test_size=0, keep_details=True)
trainset = [x.with_inputs('question', 'answer') for x in dataset.train]
devset = [x.with_inputs('question', 'answer') for x in dataset.dev]
class GenerateSearchQuery(dspy.Signature):
"""Write a simple search query that will help answer a complex question."""
context = dspy.InputField(desc="may contain relevant facts")
question = dspy.InputField()
query = dspy.OutputField()
class GenerateTweet(dspy.Signature):
"""Generate an engaging tweet that effectively answers a question staying faithful to the context, is less than 280 characters, and has no hashtags."""
question = dspy.InputField()
context = dspy.InputField(desc="may contain relevant facts")
tweet = dspy.OutputField()
class Tweeter(dspy.Module):
def __init__(self):
super().__init__()
self.generate_tweet = dspy.ChainOfThought(GenerateTweet)
def forward(self, question, answer):
context = []
max_hops=2
passages_per_hop=3
generate_query = [dspy.ChainOfThought(GenerateSearchQuery) for _ in range(max_hops)]
retrieve = dspy.Retrieve(k=passages_per_hop)
for hop in range(max_hops):
query = generate_query[hop](context=context, question=question).query
passages = retrieve(query).passages
context = deduplicate(context + passages)
generated_tweet = self.generate_tweet(question=question, context=context).tweet
return dspy.Prediction(generated_tweet=generated_tweet, context=context)
tweeter = Tweeter()
def has_no_hashtags(text):
return len(re.findall(r"#\w+", text)) == 0
def is_within_length_limit(text, length_limit=280):
return len(text) <= length_limit
def is_assessment_yes(assessment_answer):
"""Check if the first word of the assessment answer is 'yes'."""
return assessment_answer.split()[0].lower() == 'yes'
def has_correct_answer(text, answer):
return answer in text
class AssessTweet(dspy.Signature):
"""Assess the quality of a tweet along the specified dimension."""
context = dspy.InputField(desc='ignore if N/A')
assessed_text = dspy.InputField()
assessment_question = dspy.InputField()
assessment_answer = dspy.OutputField(desc="Yes or No")
def no_hashtags_metric(gold, pred, trace=None):
tweet = pred.generated_tweet
no_hashtags = has_no_hashtags(tweet)
score = no_hashtags
return score
def is_correct_metric(gold, pred, trace=None):
answer, tweet = gold.answer, pred.generated_tweet
correct = has_correct_answer(tweet, answer)
score = correct
return score
def within_length_metric(gold, pred, trace=None):
tweet = pred.generated_tweet
within_length_limit = is_within_length_limit(tweet, 280)
score = within_length_limit
return score
def engaging_metric(gold, pred, trace=None):
tweet = pred.generated_tweet
engaging = "Does the assessed text make for a self-contained, engaging tweet? Say no if it is not engaging."
engaging = dspy.Predict(AssessTweet)(context='N/A', assessed_text=tweet, assessment_question=engaging)
engaging = engaging.assessment_answer.split()[0].lower() == 'yes'
score = engaging
return score
def faithful_metric(gold, pred, trace=None):
context, tweet = pred.context, pred.generated_tweet
faithful = "Is the assessed text grounded in the context? Say no if it includes significant facts not in the context."
faithful = dspy.Predict(AssessTweet)(context=context, assessed_text=tweet, assessment_question=faithful)
faithful = faithful.assessment_answer.split()[0].lower() == 'yes'
score = faithful
return score
def overall_metric(gold, pred, trace=None):
answer, context, tweet = gold.answer, pred.context, pred.generated_tweet
no_hashtags = has_no_hashtags(tweet)
within_length_limit = is_within_length_limit(tweet, 280)
correct = has_correct_answer(tweet, answer)
engaging = "Does the assessed text make for a self-contained, engaging tweet? Say no if it is not engaging."
faithful = "Is the assessed text grounded in the context? Say no if it includes significant facts not in the context."
faithful = dspy.Predict(AssessTweet)(context=context, assessed_text=tweet, assessment_question=faithful)
engaging = dspy.Predict(AssessTweet)(context='N/A', assessed_text=tweet, assessment_question=engaging)
engaging, faithful = [m.assessment_answer.split()[0].lower() == 'yes' for m in [engaging, faithful]]
score = (correct + engaging + faithful + no_hashtags + within_length_limit) if correct and within_length_limit else 0
return score / 5.0
metrics = [no_hashtags_metric, is_correct_metric, within_length_metric, engaging_metric, faithful_metric, overall_metric]
for metric in metrics:
evaluate = Evaluate(metric=metric, devset=devset, num_threads=1, display_progress=True, display_table=5)
evaluate(tweeter)
example = devset[10]
tweet = tweeter(question=example.question, answer = example.answer)
print(f'Generated Tweet: ', tweet.generated_tweet)
tweet.context
for metric in metrics:
evaluate = Evaluate(metric=metric, devset=devset[10:11], num_threads=1, display_progress=True, display_table=5)
evaluate(tweeter)
class TweeterWithAssertions(dspy.Module):
def __init__(self):
super().__init__()
self.generate_tweet = dspy.ChainOfThought(GenerateTweet)
def forward(self, question, answer):
context = []
max_hops=2
passages_per_hop=3
generate_query = [dspy.ChainOfThought(GenerateSearchQuery) for _ in range(max_hops)]
retrieve = dspy.Retrieve(k=passages_per_hop)
for hop in range(max_hops):
query = generate_query[hop](context=context, question=question).query
passages = retrieve(query).passages
context = deduplicate(context + passages)
generated_tweet = self.generate_tweet(question=question, context=context).tweet
dspy.Suggest(has_no_hashtags(generated_tweet), f"Please revise the tweet to remove hashtag phrases following it.", target_module=GenerateTweet)
dspy.Suggest(is_within_length_limit(generated_tweet, 280), f"Please ensure the tweet is within {280} characters.", target_module=GenerateTweet)
dspy.Suggest(has_correct_answer(generated_tweet, answer), "The tweet does not include the correct answer to the question. Please revise accordingly.", target_module=GenerateTweet)
engaging_question = "Does the assessed text make for a self-contained, engaging tweet? Say no if it is not engaging."
engaging_assessment = dspy.Predict(AssessTweet)(context=context, assessed_text=generated_tweet, assessment_question=engaging_question)
dspy.Suggest(is_assessment_yes(engaging_assessment.assessment_answer), "The text is not engaging enough. Please revise to make it more captivating.", target_module=GenerateTweet)
faithful_question = "Is the assessed text grounded in the context? Say no if it includes significant facts not in the context."
faithful_assessment = | dspy.Predict(AssessTweet) | dspy.Predict |
import dspy
from dsp.utils import deduplicate
from dspy.datasets import HotPotQA
from dspy.predict.retry import Retry
from dspy.teleprompt import BootstrapFewShot, BootstrapFewShotWithRandomSearch
from dspy.evaluate.evaluate import Evaluate
from dspy.primitives.assertions import assert_transform_module, backtrack_handler
import os
import openai
openai.api_key = os.getenv('OPENAI_API_KEY')
colbertv2_wiki17_abstracts = dspy.ColBERTv2(url='http://20.102.90.50:2017/wiki17_abstracts')
dspy.settings.configure(rm=colbertv2_wiki17_abstracts)
turbo = dspy.OpenAI(model='gpt-3.5-turbo', max_tokens=500)
dspy.settings.configure(lm=turbo, trace=[], temperature=0.7)
dataset = | HotPotQA(train_seed=1, train_size=300, eval_seed=2023, dev_size=300, test_size=0) | dspy.datasets.HotPotQA |
import dspy
from dspy.evaluate.evaluate import Evaluate
from dspy.teleprompt import BootstrapFewShotWithRandomSearch
colbertv2 = dspy.ColBERTv2(url='http://20.102.90.50:2017/wiki17_abstracts')
| dspy.configure(rm=colbertv2) | dspy.configure |
get_ipython().system('git clone https://huggingface.co/arnavs11/DSPy_TweetGen_Cache')
get_ipython().run_line_magic('cd', 'DSPy_TweetGen_Cache/')
get_ipython().system('git checkout master')
get_ipython().run_line_magic('cd', '..')
import os
repo_clone_path = '/content/DSPy_TweetGen_Cache'
if not os.access('/content', os.W_OK):
repo_clone_path = os.path.join(os.getcwd(), 'DSPy_TweetGen_Cache')
os.environ["DSP_NOTEBOOK_CACHEDIR"] = repo_clone_path
get_ipython().run_line_magic('load_ext', 'autoreload')
get_ipython().run_line_magic('autoreload', '2')
import sys
import os
import regex as re
import json
try: # When on google Colab, let's clone the notebook so we download the cache.
import google.colab
repo_path = 'dspy'
get_ipython().system('git -C $repo_path pull origin || git clone https://github.com/stanfordnlp/dspy $repo_path')
except:
repo_path = '.'
if repo_path not in sys.path:
sys.path.append(repo_path)
import pkg_resources # Install the package if it's not installed
if not "dspy-ai" in {pkg.key for pkg in pkg_resources.working_set}:
get_ipython().system('pip install -U pip')
get_ipython().system('pip install dspy-ai')
get_ipython().system('pip install openai~=0.28.1')
get_ipython().system('pip install -e $repo_path')
import dspy
from dspy.predict import Retry
from dspy.datasets import HotPotQA
from dspy.teleprompt import BootstrapFewShotWithRandomSearch
from dsp.utils import deduplicate
from dspy.evaluate.evaluate import Evaluate
from dspy.primitives.assertions import assert_transform_module, backtrack_handler
colbertv2_wiki17_abstracts = dspy.ColBERTv2(url='http://20.102.90.50:2017/wiki17_abstracts')
dspy.settings.configure(rm=colbertv2_wiki17_abstracts)
turbo = dspy.OpenAI(model='gpt-3.5-turbo', max_tokens=500)
dspy.settings.configure(lm=turbo, trace=[], temperature=0.7)
dataset = HotPotQA(train_seed=1, train_size=300, eval_seed=2023, dev_size=300, test_size=0, keep_details=True)
trainset = [x.with_inputs('question', 'answer') for x in dataset.train]
devset = [x.with_inputs('question', 'answer') for x in dataset.dev]
class GenerateSearchQuery(dspy.Signature):
"""Write a simple search query that will help answer a complex question."""
context = dspy.InputField(desc="may contain relevant facts")
question = dspy.InputField()
query = dspy.OutputField()
class GenerateTweet(dspy.Signature):
"""Generate an engaging tweet that effectively answers a question staying faithful to the context, is less than 280 characters, and has no hashtags."""
question = dspy.InputField()
context = dspy.InputField(desc="may contain relevant facts")
tweet = dspy.OutputField()
class Tweeter(dspy.Module):
def __init__(self):
super().__init__()
self.generate_tweet = dspy.ChainOfThought(GenerateTweet)
def forward(self, question, answer):
context = []
max_hops=2
passages_per_hop=3
generate_query = [dspy.ChainOfThought(GenerateSearchQuery) for _ in range(max_hops)]
retrieve = dspy.Retrieve(k=passages_per_hop)
for hop in range(max_hops):
query = generate_query[hop](context=context, question=question).query
passages = retrieve(query).passages
context = deduplicate(context + passages)
generated_tweet = self.generate_tweet(question=question, context=context).tweet
return dspy.Prediction(generated_tweet=generated_tweet, context=context)
tweeter = Tweeter()
def has_no_hashtags(text):
return len(re.findall(r"#\w+", text)) == 0
def is_within_length_limit(text, length_limit=280):
return len(text) <= length_limit
def is_assessment_yes(assessment_answer):
"""Check if the first word of the assessment answer is 'yes'."""
return assessment_answer.split()[0].lower() == 'yes'
def has_correct_answer(text, answer):
return answer in text
class AssessTweet(dspy.Signature):
"""Assess the quality of a tweet along the specified dimension."""
context = dspy.InputField(desc='ignore if N/A')
assessed_text = dspy.InputField()
assessment_question = dspy.InputField()
assessment_answer = dspy.OutputField(desc="Yes or No")
def no_hashtags_metric(gold, pred, trace=None):
tweet = pred.generated_tweet
no_hashtags = has_no_hashtags(tweet)
score = no_hashtags
return score
def is_correct_metric(gold, pred, trace=None):
answer, tweet = gold.answer, pred.generated_tweet
correct = has_correct_answer(tweet, answer)
score = correct
return score
def within_length_metric(gold, pred, trace=None):
tweet = pred.generated_tweet
within_length_limit = is_within_length_limit(tweet, 280)
score = within_length_limit
return score
def engaging_metric(gold, pred, trace=None):
tweet = pred.generated_tweet
engaging = "Does the assessed text make for a self-contained, engaging tweet? Say no if it is not engaging."
engaging = dspy.Predict(AssessTweet)(context='N/A', assessed_text=tweet, assessment_question=engaging)
engaging = engaging.assessment_answer.split()[0].lower() == 'yes'
score = engaging
return score
def faithful_metric(gold, pred, trace=None):
context, tweet = pred.context, pred.generated_tweet
faithful = "Is the assessed text grounded in the context? Say no if it includes significant facts not in the context."
faithful = dspy.Predict(AssessTweet)(context=context, assessed_text=tweet, assessment_question=faithful)
faithful = faithful.assessment_answer.split()[0].lower() == 'yes'
score = faithful
return score
def overall_metric(gold, pred, trace=None):
answer, context, tweet = gold.answer, pred.context, pred.generated_tweet
no_hashtags = has_no_hashtags(tweet)
within_length_limit = is_within_length_limit(tweet, 280)
correct = has_correct_answer(tweet, answer)
engaging = "Does the assessed text make for a self-contained, engaging tweet? Say no if it is not engaging."
faithful = "Is the assessed text grounded in the context? Say no if it includes significant facts not in the context."
faithful = dspy.Predict(AssessTweet)(context=context, assessed_text=tweet, assessment_question=faithful)
engaging = dspy.Predict(AssessTweet)(context='N/A', assessed_text=tweet, assessment_question=engaging)
engaging, faithful = [m.assessment_answer.split()[0].lower() == 'yes' for m in [engaging, faithful]]
score = (correct + engaging + faithful + no_hashtags + within_length_limit) if correct and within_length_limit else 0
return score / 5.0
metrics = [no_hashtags_metric, is_correct_metric, within_length_metric, engaging_metric, faithful_metric, overall_metric]
for metric in metrics:
evaluate = Evaluate(metric=metric, devset=devset, num_threads=1, display_progress=True, display_table=5)
evaluate(tweeter)
example = devset[10]
tweet = tweeter(question=example.question, answer = example.answer)
print(f'Generated Tweet: ', tweet.generated_tweet)
tweet.context
for metric in metrics:
evaluate = | Evaluate(metric=metric, devset=devset[10:11], num_threads=1, display_progress=True, display_table=5) | dspy.evaluate.evaluate.Evaluate |
get_ipython().run_line_magic('load_ext', 'autoreload')
get_ipython().run_line_magic('autoreload', '2')
import sys
import pkg_resources
try: # When on Colab, let's install pyserini, Pytorch, and Faiss
import google.colab
repo_path = 'dspy'
get_ipython().system('git -C $repo_path pull origin || git clone https://github.com/stanfordnlp/dspy $repo_path')
get_ipython().run_line_magic('cd', '$repo_path')
get_ipython().system('pip install -e .')
if not "pyserini" in {pkg.key for pkg in pkg_resources.working_set}:
get_ipython().system('pip install pyserini')
if not "torch" in {pkg.key for pkg in pkg_resources.working_set}:
get_ipython().system('pip install torch')
if not "faiss-cpu" in {pkg.key for pkg in pkg_resources.working_set}:
get_ipython().system('pip install faiss-cpu')
except:
repo_path = '.'
if not "dspy-ai" in {pkg.key for pkg in pkg_resources.working_set}:
get_ipython().system('pip install -U pip')
get_ipython().system('pip install dspy-ai')
if repo_path not in sys.path:
sys.path.append(repo_path)
import dspy
pys_ret_prebuilt = | dspy.Pyserini(index='beir-v1.0.0-nfcorpus.contriever-msmarco', query_encoder='facebook/contriever-msmarco', id_field='_id', text_fields=['title', 'text']) | dspy.Pyserini |
import dspy
from dspy.evaluate.evaluate import Evaluate
from dspy.teleprompt import BootstrapFewShotWithRandomSearch
colbertv2 = dspy.ColBERTv2(url='http://20.102.90.50:2017/wiki17_abstracts')
dspy.configure(rm=colbertv2)
from langchain_openai import OpenAI
from langchain.globals import set_llm_cache
from langchain.cache import SQLiteCache
set_llm_cache(SQLiteCache(database_path="cache.db"))
llm = OpenAI(model_name="gpt-3.5-turbo-instruct", temperature=0)
retrieve = lambda x: dspy.Retrieve(k=5)(x["question"]).passages
from langchain_core.prompts import PromptTemplate
from langchain_core.output_parsers import StrOutputParser
from langchain_core.runnables import RunnablePassthrough
prompt = PromptTemplate.from_template("Given {context}, answer the question `{question}` as a tweet.")
vanilla_chain = RunnablePassthrough.assign(context=retrieve) | prompt | llm | StrOutputParser()
from dspy.predict.langchain import LangChainPredict, LangChainModule
zeroshot_chain = RunnablePassthrough.assign(context=retrieve) | LangChainPredict(prompt, llm) | StrOutputParser()
zeroshot_chain = | LangChainModule(zeroshot_chain) | dspy.predict.langchain.LangChainModule |
import dspy
from dsp.utils import deduplicate
from dspy.datasets import HotPotQA
from dspy.predict.retry import Retry
from dspy.teleprompt import BootstrapFewShot, BootstrapFewShotWithRandomSearch
from dspy.evaluate.evaluate import Evaluate
from dspy.primitives.assertions import assert_transform_module, backtrack_handler
import os
import openai
openai.api_key = os.getenv('OPENAI_API_KEY')
colbertv2_wiki17_abstracts = dspy.ColBERTv2(url='http://20.102.90.50:2017/wiki17_abstracts')
dspy.settings.configure(rm=colbertv2_wiki17_abstracts)
turbo = | dspy.OpenAI(model='gpt-3.5-turbo', max_tokens=500) | dspy.OpenAI |
import dspy
from dsp.utils import deduplicate
from dspy.datasets import HotPotQA
from dspy.predict.retry import Retry
from dspy.teleprompt import BootstrapFewShot, BootstrapFewShotWithRandomSearch
from dspy.evaluate.evaluate import Evaluate
from dspy.primitives.assertions import assert_transform_module, backtrack_handler
import os
import openai
openai.api_key = os.getenv('OPENAI_API_KEY')
colbertv2_wiki17_abstracts = dspy.ColBERTv2(url='http://20.102.90.50:2017/wiki17_abstracts')
dspy.settings.configure(rm=colbertv2_wiki17_abstracts)
turbo = dspy.OpenAI(model='gpt-3.5-turbo', max_tokens=500)
dspy.settings.configure(lm=turbo, trace=[], temperature=0.7)
dataset = HotPotQA(train_seed=1, train_size=300, eval_seed=2023, dev_size=300, test_size=0)
trainset = [x.with_inputs('question') for x in dataset.train]
devset = [x.with_inputs('question') for x in dataset.dev]
def validate_query_distinction_local(previous_queries, query):
"""check if query is distinct from previous queries"""
if previous_queries == []:
return True
if dspy.evaluate.answer_exact_match_str(query, previous_queries, frac=0.8):
return False
return True
def validate_context_and_answer_and_hops(example, pred, trace=None):
if not dspy.evaluate.answer_exact_match(example, pred):
return False
if not dspy.evaluate.answer_passage_match(example, pred):
return False
return True
def gold_passages_retrieved(example, pred, trace=None):
gold_titles = set(map(dspy.evaluate.normalize_text, example['gold_titles']))
found_titles = set(map(dspy.evaluate.normalize_text, [c.split(' | ')[0] for c in pred.context]))
return gold_titles.issubset(found_titles)
class GenerateAnswer(dspy.Signature):
"""Answer questions with short factoid answers."""
context = dspy.InputField(desc="may contain relevant facts")
question = dspy.InputField()
answer = dspy.OutputField(desc="often between 1 and 5 words")
class GenerateSearchQuery(dspy.Signature):
"""Write a simple search query that will help answer a complex question."""
context = dspy.InputField(desc="may contain relevant facts")
question = dspy.InputField()
query = dspy.OutputField()
def all_queries_distinct(prev_queries):
query_distinct = True
for i, query in enumerate(prev_queries):
if validate_query_distinction_local(prev_queries[:i], query) == False:
query_distinct = False
break
return query_distinct
class SimplifiedBaleen(dspy.Module):
def __init__(self, passages_per_hop=2, max_hops=2):
super().__init__()
self.generate_query = [dspy.ChainOfThought(GenerateSearchQuery) for _ in range(max_hops)]
self.retrieve = dspy.Retrieve(k=passages_per_hop)
self.generate_answer = dspy.ChainOfThought(GenerateAnswer)
self.max_hops = max_hops
self.passed_suggestions = 0
def forward(self, question):
context = []
prev_queries = [question]
for hop in range(self.max_hops):
query = self.generate_query[hop](context=context, question=question).query
prev_queries.append(query)
passages = self.retrieve(query).passages
context = deduplicate(context + passages)
if all_queries_distinct(prev_queries):
self.passed_suggestions += 1
pred = self.generate_answer(context=context, question=question)
pred = dspy.Prediction(context=context, answer=pred.answer)
return pred
class SimplifiedBaleenAssertions(dspy.Module):
def __init__(self, passages_per_hop=2, max_hops=2):
super().__init__()
self.generate_query = [dspy.ChainOfThought(GenerateSearchQuery) for _ in range(max_hops)]
self.retrieve = dspy.Retrieve(k=passages_per_hop)
self.generate_answer = dspy.ChainOfThought(GenerateAnswer)
self.max_hops = max_hops
self.passed_suggestions = 0
def forward(self, question):
context = []
prev_queries = [question]
for hop in range(self.max_hops):
query = self.generate_query[hop](context=context, question=question).query
dspy.Suggest(
len(query) <= 100,
"Query should be short and less than 100 characters",
)
dspy.Suggest(
validate_query_distinction_local(prev_queries, query),
"Query should be distinct from: "
+ "; ".join(f"{i+1}) {q}" for i, q in enumerate(prev_queries)),
)
prev_queries.append(query)
passages = self.retrieve(query).passages
context = deduplicate(context + passages)
if all_queries_distinct(prev_queries):
self.passed_suggestions += 1
pred = self.generate_answer(context=context, question=question)
pred = | dspy.Prediction(context=context, answer=pred.answer) | dspy.Prediction |
get_ipython().system('git clone https://huggingface.co/arnavs11/DSPy_QuizGen_Cache')
get_ipython().run_line_magic('cd', 'DSPy_QuizGen_Cache/')
get_ipython().system('git checkout master')
get_ipython().run_line_magic('cd', '..')
import os
repo_clone_path = '/content/DSPy_QuizGen_Cache'
if not os.access('/content', os.W_OK):
repo_clone_path = os.path.join(os.getcwd(), 'DSPy_QuizGen_Cache')
os.environ["DSP_NOTEBOOK_CACHEDIR"] = repo_clone_path
get_ipython().run_line_magic('load_ext', 'autoreload')
get_ipython().run_line_magic('autoreload', '2')
import sys
import os
import regex as re
import json
try: # When on google Colab, let's clone the notebook so we download the cache.
import google.colab
repo_path = 'dspy'
get_ipython().system('git -C $repo_path pull origin || git clone https://github.com/stanfordnlp/dspy $repo_path')
except:
repo_path = '.'
if repo_path not in sys.path:
sys.path.append(repo_path)
import pkg_resources # Install the package if it's not installed
if not "dspy-ai" in {pkg.key for pkg in pkg_resources.working_set}:
get_ipython().system('pip install -U pip')
get_ipython().system('pip install dspy-ai')
get_ipython().system('pip install openai~=0.28.1')
get_ipython().system('pip install -e $repo_path')
import dspy
from dspy.predict import Retry
from dspy.datasets import HotPotQA
from dspy.teleprompt import BootstrapFewShotWithRandomSearch
from dspy.evaluate.evaluate import Evaluate
from dspy.primitives.assertions import assert_transform_module, backtrack_handler
colbertv2_wiki17_abstracts = | dspy.ColBERTv2(url='http://20.102.90.50:2017/wiki17_abstracts') | dspy.ColBERTv2 |
get_ipython().system('git clone https://huggingface.co/arnavs11/DSPy_LongFormQA_Cache')
get_ipython().run_line_magic('cd', 'DSPy_LongFormQA_Cache/')
get_ipython().system('git checkout master')
get_ipython().run_line_magic('cd', '..')
import os
repo_clone_path = '/content/DSPy_LongFormQA_Cache'
if not os.access('/content', os.W_OK):
repo_clone_path = os.path.join(os.getcwd(), 'DSPy_LongFormQA_Cache')
os.environ["DSP_NOTEBOOK_CACHEDIR"] = repo_clone_path
get_ipython().run_line_magic('load_ext', 'autoreload')
get_ipython().run_line_magic('autoreload', '2')
import sys
import os
import regex as re
try: # When on google Colab, let's clone the notebook so we download the cache.
import google.colab
repo_path = 'dspy'
get_ipython().system('git -C $repo_path pull origin || git clone https://github.com/stanfordnlp/dspy $repo_path')
except:
repo_path = '.'
if repo_path not in sys.path:
sys.path.append(repo_path)
import pkg_resources # Install the package if it's not installed
if not "dspy-ai" in {pkg.key for pkg in pkg_resources.working_set}:
get_ipython().system('pip install -U pip')
get_ipython().system('pip install dspy-ai')
get_ipython().system('pip install openai~=0.28.1')
get_ipython().system('pip install -e $repo_path')
import dspy
from dspy.predict import Retry
from dspy.datasets import HotPotQA
from dspy.teleprompt import BootstrapFewShotWithRandomSearch
from dsp.utils import EM, normalize_text
from dspy.primitives.assertions import assert_transform_module, backtrack_handler
get_ipython().run_line_magic('cd', 'dspy/examples/longformqa')
from utils import extract_text_by_citation, correct_citation_format, has_citations, citations_check
colbertv2_wiki17_abstracts = | dspy.ColBERTv2(url='http://20.102.90.50:2017/wiki17_abstracts') | dspy.ColBERTv2 |
get_ipython().run_line_magic('load_ext', 'autoreload')
get_ipython().run_line_magic('autoreload', '2')
get_ipython().run_line_magic('pip', 'install datasets')
import datasets
ds = datasets.load_dataset("openai_humaneval")
ds['test'][0]
import dspy, dotenv, os
dotenv.load_dotenv(os.path.expanduser("~/.env")) # load OpenAI API key from .env file
lm = dspy.OpenAI(model="gpt-3.5-turbo", max_tokens=4000)
| dspy.settings.configure(lm=lm) | dspy.settings.configure |
get_ipython().run_line_magic('load_ext', 'autoreload')
get_ipython().run_line_magic('autoreload', '2')
import sys
import os
try: # When on google Colab, let's clone the notebook so we download the cache.
import google.colab
repo_path = 'dspy'
get_ipython().system('git -C $repo_path pull origin || git clone https://github.com/stanfordnlp/dspy $repo_path')
except:
repo_path = '.'
if repo_path not in sys.path:
sys.path.append(repo_path)
os.environ["DSP_NOTEBOOK_CACHEDIR"] = os.path.join(repo_path, 'cache')
import pkg_resources # Install the package if it's not installed
if not "dspy-ai" in {pkg.key for pkg in pkg_resources.working_set}:
get_ipython().system('pip install -U pip')
get_ipython().system('pip install dspy-ai')
get_ipython().system('pip install openai~=0.28.1')
import dspy
turbo = dspy.OpenAI(model='gpt-3.5-turbo')
colbertv2_wiki17_abstracts = dspy.ColBERTv2(url='http://20.102.90.50:2017/wiki17_abstracts')
dspy.settings.configure(lm=turbo, rm=colbertv2_wiki17_abstracts)
from dspy.datasets import HotPotQA
dataset = HotPotQA(train_seed=1, train_size=20, eval_seed=2023, dev_size=50, test_size=0)
trainset = [x.with_inputs('question') for x in dataset.train]
devset = [x.with_inputs('question') for x in dataset.dev]
len(trainset), len(devset)
train_example = trainset[0]
print(f"Question: {train_example.question}")
print(f"Answer: {train_example.answer}")
dev_example = devset[18]
print(f"Question: {dev_example.question}")
print(f"Answer: {dev_example.answer}")
print(f"Relevant Wikipedia Titles: {dev_example.gold_titles}")
print(f"For this dataset, training examples have input keys {train_example.inputs().keys()} and label keys {train_example.labels().keys()}")
print(f"For this dataset, dev examples have input keys {dev_example.inputs().keys()} and label keys {dev_example.labels().keys()}")
class BasicQA(dspy.Signature):
"""Answer questions with short factoid answers."""
question = dspy.InputField()
answer = dspy.OutputField(desc="often between 1 and 5 words")
generate_answer = dspy.Predict(BasicQA)
pred = generate_answer(question=dev_example.question)
print(f"Question: {dev_example.question}")
print(f"Predicted Answer: {pred.answer}")
turbo.inspect_history(n=1)
generate_answer_with_chain_of_thought = dspy.ChainOfThought(BasicQA)
pred = generate_answer_with_chain_of_thought(question=dev_example.question)
print(f"Question: {dev_example.question}")
print(f"Thought: {pred.rationale.split('.', 1)[1].strip()}")
print(f"Predicted Answer: {pred.answer}")
retrieve = dspy.Retrieve(k=3)
topK_passages = retrieve(dev_example.question).passages
print(f"Top {retrieve.k} passages for question: {dev_example.question} \n", '-' * 30, '\n')
for idx, passage in enumerate(topK_passages):
print(f'{idx+1}]', passage, '\n')
retrieve("When was the first FIFA World Cup held?").passages[0]
class GenerateAnswer(dspy.Signature):
"""Answer questions with short factoid answers."""
context = dspy.InputField(desc="may contain relevant facts")
question = dspy.InputField()
answer = dspy.OutputField(desc="often between 1 and 5 words")
class RAG(dspy.Module):
def __init__(self, num_passages=3):
super().__init__()
self.retrieve = dspy.Retrieve(k=num_passages)
self.generate_answer = dspy.ChainOfThought(GenerateAnswer)
def forward(self, question):
context = self.retrieve(question).passages
prediction = self.generate_answer(context=context, question=question)
return dspy.Prediction(context=context, answer=prediction.answer)
from dspy.teleprompt import BootstrapFewShot
def validate_context_and_answer(example, pred, trace=None):
answer_EM = dspy.evaluate.answer_exact_match(example, pred)
answer_PM = | dspy.evaluate.answer_passage_match(example, pred) | dspy.evaluate.answer_passage_match |
get_ipython().system('git clone https://huggingface.co/arnavs11/DSPy_QuizGen_Cache')
get_ipython().run_line_magic('cd', 'DSPy_QuizGen_Cache/')
get_ipython().system('git checkout master')
get_ipython().run_line_magic('cd', '..')
import os
repo_clone_path = '/content/DSPy_QuizGen_Cache'
if not os.access('/content', os.W_OK):
repo_clone_path = os.path.join(os.getcwd(), 'DSPy_QuizGen_Cache')
os.environ["DSP_NOTEBOOK_CACHEDIR"] = repo_clone_path
get_ipython().run_line_magic('load_ext', 'autoreload')
get_ipython().run_line_magic('autoreload', '2')
import sys
import os
import regex as re
import json
try: # When on google Colab, let's clone the notebook so we download the cache.
import google.colab
repo_path = 'dspy'
get_ipython().system('git -C $repo_path pull origin || git clone https://github.com/stanfordnlp/dspy $repo_path')
except:
repo_path = '.'
if repo_path not in sys.path:
sys.path.append(repo_path)
import pkg_resources # Install the package if it's not installed
if not "dspy-ai" in {pkg.key for pkg in pkg_resources.working_set}:
get_ipython().system('pip install -U pip')
get_ipython().system('pip install dspy-ai')
get_ipython().system('pip install openai~=0.28.1')
get_ipython().system('pip install -e $repo_path')
import dspy
from dspy.predict import Retry
from dspy.datasets import HotPotQA
from dspy.teleprompt import BootstrapFewShotWithRandomSearch
from dspy.evaluate.evaluate import Evaluate
from dspy.primitives.assertions import assert_transform_module, backtrack_handler
colbertv2_wiki17_abstracts = dspy.ColBERTv2(url='http://20.102.90.50:2017/wiki17_abstracts')
dspy.settings.configure(rm=colbertv2_wiki17_abstracts)
turbo = | dspy.OpenAI(model='gpt-3.5-turbo', max_tokens=500) | dspy.OpenAI |
get_ipython().system('git clone https://huggingface.co/arnavs11/DSPy_LongFormQA_Cache')
get_ipython().run_line_magic('cd', 'DSPy_LongFormQA_Cache/')
get_ipython().system('git checkout master')
get_ipython().run_line_magic('cd', '..')
import os
repo_clone_path = '/content/DSPy_LongFormQA_Cache'
if not os.access('/content', os.W_OK):
repo_clone_path = os.path.join(os.getcwd(), 'DSPy_LongFormQA_Cache')
os.environ["DSP_NOTEBOOK_CACHEDIR"] = repo_clone_path
get_ipython().run_line_magic('load_ext', 'autoreload')
get_ipython().run_line_magic('autoreload', '2')
import sys
import os
import regex as re
try: # When on google Colab, let's clone the notebook so we download the cache.
import google.colab
repo_path = 'dspy'
get_ipython().system('git -C $repo_path pull origin || git clone https://github.com/stanfordnlp/dspy $repo_path')
except:
repo_path = '.'
if repo_path not in sys.path:
sys.path.append(repo_path)
import pkg_resources # Install the package if it's not installed
if not "dspy-ai" in {pkg.key for pkg in pkg_resources.working_set}:
get_ipython().system('pip install -U pip')
get_ipython().system('pip install dspy-ai')
get_ipython().system('pip install openai~=0.28.1')
get_ipython().system('pip install -e $repo_path')
import dspy
from dspy.predict import Retry
from dspy.datasets import HotPotQA
from dspy.teleprompt import BootstrapFewShotWithRandomSearch
from dsp.utils import EM, normalize_text
from dspy.primitives.assertions import assert_transform_module, backtrack_handler
get_ipython().run_line_magic('cd', 'dspy/examples/longformqa')
from utils import extract_text_by_citation, correct_citation_format, has_citations, citations_check
colbertv2_wiki17_abstracts = dspy.ColBERTv2(url='http://20.102.90.50:2017/wiki17_abstracts')
dspy.settings.configure(rm=colbertv2_wiki17_abstracts)
turbo = dspy.OpenAI(model='gpt-3.5-turbo', max_tokens=500)
dspy.settings.configure(lm=turbo, trace=[], temperature=0.7)
dataset = HotPotQA(train_seed=1, train_size=300, eval_seed=2023, dev_size=300, test_size=0, keep_details=True)
trainset = [x.with_inputs('question') for x in dataset.train]
devset = [x.with_inputs('question') for x in dataset.dev]
train_example = trainset[0]
print(f"Question: {train_example.question}")
print(f"Answer: {train_example.answer}")
print(f"Relevant Wikipedia Titles: {train_example.gold_titles}")
dev_example = devset[18]
print(f"Question: {dev_example.question}")
print(f"Answer: {dev_example.answer}")
print(f"Relevant Wikipedia Titles: {dev_example.gold_titles}")
from dsp.utils import deduplicate
class GenerateSearchQuery(dspy.Signature):
"""Write a simple search query that will help answer a complex question."""
context = dspy.InputField(desc="may contain relevant facts")
question = dspy.InputField()
query = dspy.OutputField()
class GenerateCitedParagraph(dspy.Signature):
"""Generate a paragraph with citations."""
context = dspy.InputField(desc="may contain relevant facts")
question = dspy.InputField()
paragraph = dspy.OutputField(desc="includes citations")
class LongFormQA(dspy.Module):
def __init__(self, passages_per_hop=3, max_hops=2):
super().__init__()
self.generate_query = [dspy.ChainOfThought(GenerateSearchQuery) for _ in range(max_hops)]
self.retrieve = dspy.Retrieve(k=passages_per_hop)
self.generate_cited_paragraph = dspy.ChainOfThought(GenerateCitedParagraph)
self.max_hops = max_hops
def forward(self, question):
context = []
for hop in range(self.max_hops):
query = self.generate_query[hop](context=context, question=question).query
passages = self.retrieve(query).passages
context = deduplicate(context + passages)
pred = self.generate_cited_paragraph(context=context, question=question)
pred = dspy.Prediction(context=context, paragraph=pred.paragraph)
return pred
class CheckCitationFaithfulness(dspy.Signature):
"""Verify that the text is based on the provided context."""
context = dspy.InputField(desc="may contain relevant facts")
text = dspy.InputField(desc="between 1 to 2 sentences")
faithfulness = dspy.OutputField(desc="boolean indicating if text is faithful to context")
def citation_faithfulness(example, pred, trace):
paragraph, context = pred.paragraph, pred.context
citation_dict = extract_text_by_citation(paragraph)
if not citation_dict:
return False, None
context_dict = {str(i): context[i].split(' | ')[1] for i in range(len(context))}
faithfulness_results = []
unfaithful_citations = []
check_citation_faithfulness = dspy.ChainOfThought(CheckCitationFaithfulness)
for citation_num, texts in citation_dict.items():
if citation_num not in context_dict:
continue
current_context = context_dict[citation_num]
for text in texts:
try:
result = check_citation_faithfulness(context=current_context, text=text)
is_faithful = result.faithfulness.lower() == 'true'
faithfulness_results.append(is_faithful)
if not is_faithful:
unfaithful_citations.append({'paragraph': paragraph, 'text': text, 'context': current_context})
except ValueError as e:
faithfulness_results.append(False)
unfaithful_citations.append({'paragraph': paragraph, 'text': text, 'error': str(e)})
final_faithfulness = all(faithfulness_results)
if not faithfulness_results:
return False, None
return final_faithfulness, unfaithful_citations
def extract_cited_titles_from_paragraph(paragraph, context):
cited_indices = [int(m.group(1)) for m in re.finditer(r'\[(\d+)\]\.', paragraph)]
cited_indices = [index - 1 for index in cited_indices if index <= len(context)]
cited_titles = [context[index].split(' | ')[0] for index in cited_indices]
return cited_titles
def calculate_recall(example, pred, trace=None):
gold_titles = set(example['gold_titles'])
found_cited_titles = set(extract_cited_titles_from_paragraph(pred.paragraph, pred.context))
intersection = gold_titles.intersection(found_cited_titles)
recall = len(intersection) / len(gold_titles) if gold_titles else 0
return recall
def calculate_precision(example, pred, trace=None):
gold_titles = set(example['gold_titles'])
found_cited_titles = set(extract_cited_titles_from_paragraph(pred.paragraph, pred.context))
intersection = gold_titles.intersection(found_cited_titles)
precision = len(intersection) / len(found_cited_titles) if found_cited_titles else 0
return precision
def answer_correctness(example, pred, trace=None):
assert hasattr(example, 'answer'), "Example does not have 'answer'."
normalized_context = normalize_text(pred.paragraph)
if isinstance(example.answer, str):
gold_answers = [example.answer]
elif isinstance(example.answer, list):
gold_answers = example.answer
else:
raise ValueError("'example.answer' is not string or list.")
return 1 if any(normalize_text(answer) in normalized_context for answer in gold_answers) else 0
def evaluate(module):
correctness_values = []
recall_values = []
precision_values = []
citation_faithfulness_values = []
for i in range(len(devset)):
example = devset[i]
try:
pred = module(question=example.question)
correctness_values.append(answer_correctness(example, pred))
citation_faithfulness_score, _ = citation_faithfulness(None, pred, None)
citation_faithfulness_values.append(citation_faithfulness_score)
recall = calculate_recall(example, pred)
precision = calculate_precision(example, pred)
recall_values.append(recall)
precision_values.append(precision)
except Exception as e:
print(f"Failed generation with error: {e}")
average_correctness = sum(correctness_values) / len(devset) if correctness_values else 0
average_recall = sum(recall_values) / len(devset) if recall_values else 0
average_precision = sum(precision_values) / len(devset) if precision_values else 0
average_citation_faithfulness = sum(citation_faithfulness_values) / len(devset) if citation_faithfulness_values else 0
print(f"Average Correctness: {average_correctness}")
print(f"Average Recall: {average_recall}")
print(f"Average Precision: {average_precision}")
print(f"Average Citation Faithfulness: {average_citation_faithfulness}")
longformqa = LongFormQA()
evaluate(longformqa)
question = devset[6].question
pred = longformqa(question)
citation_faithfulness_score, _ = citation_faithfulness(None, pred, None)
print(f"Question: {question}")
print(f"Predicted Paragraph: {pred.paragraph}")
print(f"Citation Faithfulness: {citation_faithfulness_score}")
class LongFormQAWithAssertions(dspy.Module):
def __init__(self, passages_per_hop=3, max_hops=2):
super().__init__()
self.generate_query = [dspy.ChainOfThought(GenerateSearchQuery) for _ in range(max_hops)]
self.retrieve = dspy.Retrieve(k=passages_per_hop)
self.generate_cited_paragraph = dspy.ChainOfThought(GenerateCitedParagraph)
self.max_hops = max_hops
def forward(self, question):
context = []
for hop in range(self.max_hops):
query = self.generate_query[hop](context=context, question=question).query
passages = self.retrieve(query).passages
context = deduplicate(context + passages)
pred = self.generate_cited_paragraph(context=context, question=question)
pred = dspy.Prediction(context=context, paragraph=pred.paragraph)
dspy.Suggest(citations_check(pred.paragraph), f"Make sure every 1-2 sentences has citations. If any 1-2 sentences lack citations, add them in 'text... [x].' format.", target_module=GenerateCitedParagraph)
_, unfaithful_outputs = citation_faithfulness(None, pred, None)
if unfaithful_outputs:
unfaithful_pairs = [(output['text'], output['context']) for output in unfaithful_outputs]
for _, context in unfaithful_pairs:
dspy.Suggest(len(unfaithful_pairs) == 0, f"Make sure your output is based on the following context: '{context}'.", target_module=GenerateCitedParagraph)
else:
return pred
return pred
longformqa_with_assertions = assert_transform_module(LongFormQAWithAssertions().map_named_predictors(Retry), backtrack_handler)
evaluate(longformqa_with_assertions)
question = devset[6].question
pred = longformqa_with_assertions(question)
citation_faithfulness_score, _ = citation_faithfulness(None, pred, None)
print(f"Question: {question}")
print(f"Predicted Paragraph: {pred.paragraph}")
print(f"Citation Faithfulness: {citation_faithfulness_score}")
longformqa = LongFormQA()
teleprompter = BootstrapFewShotWithRandomSearch(metric = answer_correctness, max_bootstrapped_demos=2, num_candidate_programs=6)
cited_longformqa = teleprompter.compile(student = longformqa, teacher = longformqa, trainset=trainset, valset=devset[:100])
evaluate(cited_longformqa)
longformqa = LongFormQA()
teleprompter = BootstrapFewShotWithRandomSearch(metric = answer_correctness, max_bootstrapped_demos=2, num_candidate_programs=6)
cited_longformqa_teacher = teleprompter.compile(student=longformqa, teacher = assert_transform_module(LongFormQAWithAssertions().map_named_predictors(Retry), backtrack_handler), trainset=trainset, valset=devset[:100])
evaluate(cited_longformqa_teacher)
longformqa = LongFormQA()
teleprompter = | BootstrapFewShotWithRandomSearch(metric = answer_correctness, max_bootstrapped_demos=2, num_candidate_programs=6) | dspy.teleprompt.BootstrapFewShotWithRandomSearch |
import dspy
from dsp.utils import deduplicate
from dspy.datasets import HotPotQA
from dspy.predict.retry import Retry
from dspy.teleprompt import BootstrapFewShot, BootstrapFewShotWithRandomSearch
from dspy.evaluate.evaluate import Evaluate
from dspy.primitives.assertions import assert_transform_module, backtrack_handler
import os
import openai
openai.api_key = os.getenv('OPENAI_API_KEY')
colbertv2_wiki17_abstracts = | dspy.ColBERTv2(url='http://20.102.90.50:2017/wiki17_abstracts') | dspy.ColBERTv2 |
get_ipython().system('git clone https://huggingface.co/arnavs11/DSPy_LongFormQA_Cache')
get_ipython().run_line_magic('cd', 'DSPy_LongFormQA_Cache/')
get_ipython().system('git checkout master')
get_ipython().run_line_magic('cd', '..')
import os
repo_clone_path = '/content/DSPy_LongFormQA_Cache'
if not os.access('/content', os.W_OK):
repo_clone_path = os.path.join(os.getcwd(), 'DSPy_LongFormQA_Cache')
os.environ["DSP_NOTEBOOK_CACHEDIR"] = repo_clone_path
get_ipython().run_line_magic('load_ext', 'autoreload')
get_ipython().run_line_magic('autoreload', '2')
import sys
import os
import regex as re
try: # When on google Colab, let's clone the notebook so we download the cache.
import google.colab
repo_path = 'dspy'
get_ipython().system('git -C $repo_path pull origin || git clone https://github.com/stanfordnlp/dspy $repo_path')
except:
repo_path = '.'
if repo_path not in sys.path:
sys.path.append(repo_path)
import pkg_resources # Install the package if it's not installed
if not "dspy-ai" in {pkg.key for pkg in pkg_resources.working_set}:
get_ipython().system('pip install -U pip')
get_ipython().system('pip install dspy-ai')
get_ipython().system('pip install openai~=0.28.1')
get_ipython().system('pip install -e $repo_path')
import dspy
from dspy.predict import Retry
from dspy.datasets import HotPotQA
from dspy.teleprompt import BootstrapFewShotWithRandomSearch
from dsp.utils import EM, normalize_text
from dspy.primitives.assertions import assert_transform_module, backtrack_handler
get_ipython().run_line_magic('cd', 'dspy/examples/longformqa')
from utils import extract_text_by_citation, correct_citation_format, has_citations, citations_check
colbertv2_wiki17_abstracts = dspy.ColBERTv2(url='http://20.102.90.50:2017/wiki17_abstracts')
dspy.settings.configure(rm=colbertv2_wiki17_abstracts)
turbo = | dspy.OpenAI(model='gpt-3.5-turbo', max_tokens=500) | dspy.OpenAI |
get_ipython().system('pip install clarifai')
get_ipython().system('pip install dspy-ai')
import dspy
from dspy.retrieve.clarifai_rm import ClarifaiRM
MODEL_URL = "https://clarifai.com/meta/Llama-2/models/llama2-70b-chat"
PAT = "CLARIFAI_PAT"
USER_ID = "YOUR_ID"
APP_ID = "YOUR_APP"
from langchain.text_splitter import CharacterTextSplitter
from langchain.document_loaders import TextLoader
from langchain.vectorstores import Clarifai as clarifaivectorstore
loader = TextLoader("YOUR_TEXT_FILE") #replace with your file path
documents = loader.load()
text_splitter = CharacterTextSplitter(chunk_size=1024, chunk_overlap=200)
docs = text_splitter.split_documents(documents)
clarifai_vector_db = clarifaivectorstore.from_documents(
user_id=USER_ID,
app_id=APP_ID,
documents=docs,
pat=PAT
)
llm=dspy.Clarifai(model=MODEL_URL, api_key=PAT, n=2, inference_params={"max_tokens":100,'temperature':0.6})
retriever_model=ClarifaiRM(clarifai_user_id=USER_ID, clarfiai_app_id=APP_ID, clarifai_pat=PAT, k=2)
dspy.settings.configure(lm=llm, rm=retriever_model)
sentence = "disney again ransacks its archives for a quick-buck sequel ." # example from the SST-2 dataset.
classify = dspy.Predict('sentence -> sentiment')
print(classify(sentence=sentence).sentiment)
retrieve = dspy.Retrieve()
topK_passages = retrieve("can I test my vehicle engine in pit?").passages
print(topK_passages)
class GenerateAnswer(dspy.Signature):
"""Think and Answer questions based on the context provided."""
context = dspy.InputField(desc="may contain relevant facts about user query")
question = dspy.InputField(desc="User query")
answer = dspy.OutputField(desc="Answer in one or two lines")
class RAG(dspy.Module):
def __init__(self):
super().__init__()
self.retrieve = dspy.Retrieve()
self.generate_answer = dspy.ChainOfThought(GenerateAnswer)
def forward(self, question):
context = self.retrieve(question).passages
prediction = self.generate_answer(context=context, question=question)
return dspy.Prediction(context=context, answer=prediction.answer)
my_question = "can I test my vehicle engine in pit before inspection?"
Rag_obj= RAG()
predict_response_llama70b=Rag_obj(my_question)
print(f"Question: {my_question}")
print(f"Predicted Answer: {predict_response_llama70b.answer}")
print(f"Retrieved Contexts (truncated): {[c[:200] + '...' for c in predict_response_llama70b.context]}")
mistral_lm = dspy.Clarifai(model="https://clarifai.com/mistralai/completion/models/mistral-7B-Instruct", api_key=PAT, n=2, inference_params={'temperature':0.6})
dspy.settings.configure(lm=mistral_lm, rm=retriever_model)
my_question = "can I test my vehicle engine in pit before inspection?"
Rag_obj= RAG()
predict_response_mistral=Rag_obj(my_question)
print(f"Question: {my_question}")
print(f"Predicted Answer: {predict_response_mistral.answer}")
print(f"Retrieved Contexts (truncated): {[c[:200] + '...' for c in predict_response_mistral.context]}")
gemini_lm = | dspy.Clarifai(model="https://clarifai.com/gcp/generate/models/gemini-pro", api_key=PAT, n=2) | dspy.Clarifai |
import dspy
from dspy.evaluate.evaluate import Evaluate
from dspy.teleprompt import BootstrapFewShotWithRandomSearch
colbertv2 = dspy.ColBERTv2(url='http://20.102.90.50:2017/wiki17_abstracts')
dspy.configure(rm=colbertv2)
from langchain_openai import OpenAI
from langchain.globals import set_llm_cache
from langchain.cache import SQLiteCache
set_llm_cache(SQLiteCache(database_path="cache.db"))
llm = OpenAI(model_name="gpt-3.5-turbo-instruct", temperature=0)
retrieve = lambda x: dspy.Retrieve(k=5)(x["question"]).passages
from langchain_core.prompts import PromptTemplate
from langchain_core.output_parsers import StrOutputParser
from langchain_core.runnables import RunnablePassthrough
prompt = PromptTemplate.from_template("Given {context}, answer the question `{question}` as a tweet.")
vanilla_chain = RunnablePassthrough.assign(context=retrieve) | prompt | llm | StrOutputParser()
from dspy.predict.langchain import LangChainPredict, LangChainModule
zeroshot_chain = RunnablePassthrough.assign(context=retrieve) | LangChainPredict(prompt, llm) | StrOutputParser()
zeroshot_chain = LangChainModule(zeroshot_chain) # then wrap the chain in a DSPy module.
question = "In what region was Eddy Mazzoleni born?"
zeroshot_chain.invoke({"question": question})
from tweet_metric import metric, trainset, valset, devset
len(trainset), len(valset), len(devset)
evaluate = Evaluate(metric=metric, devset=devset, num_threads=8, display_progress=True, display_table=5)
evaluate(zeroshot_chain)
optimizer = | BootstrapFewShotWithRandomSearch(metric=metric, max_bootstrapped_demos=3, num_candidate_programs=3) | dspy.teleprompt.BootstrapFewShotWithRandomSearch |
get_ipython().system('pip install clarifai')
get_ipython().system('pip install dspy-ai')
import dspy
from dspy.retrieve.clarifai_rm import ClarifaiRM
MODEL_URL = "https://clarifai.com/meta/Llama-2/models/llama2-70b-chat"
PAT = "CLARIFAI_PAT"
USER_ID = "YOUR_ID"
APP_ID = "YOUR_APP"
from langchain.text_splitter import CharacterTextSplitter
from langchain.document_loaders import TextLoader
from langchain.vectorstores import Clarifai as clarifaivectorstore
loader = TextLoader("YOUR_TEXT_FILE") #replace with your file path
documents = loader.load()
text_splitter = CharacterTextSplitter(chunk_size=1024, chunk_overlap=200)
docs = text_splitter.split_documents(documents)
clarifai_vector_db = clarifaivectorstore.from_documents(
user_id=USER_ID,
app_id=APP_ID,
documents=docs,
pat=PAT
)
llm=dspy.Clarifai(model=MODEL_URL, api_key=PAT, n=2, inference_params={"max_tokens":100,'temperature':0.6})
retriever_model=ClarifaiRM(clarifai_user_id=USER_ID, clarfiai_app_id=APP_ID, clarifai_pat=PAT, k=2)
dspy.settings.configure(lm=llm, rm=retriever_model)
sentence = "disney again ransacks its archives for a quick-buck sequel ." # example from the SST-2 dataset.
classify = dspy.Predict('sentence -> sentiment')
print(classify(sentence=sentence).sentiment)
retrieve = dspy.Retrieve()
topK_passages = retrieve("can I test my vehicle engine in pit?").passages
print(topK_passages)
class GenerateAnswer(dspy.Signature):
"""Think and Answer questions based on the context provided."""
context = dspy.InputField(desc="may contain relevant facts about user query")
question = dspy.InputField(desc="User query")
answer = | dspy.OutputField(desc="Answer in one or two lines") | dspy.OutputField |
get_ipython().system('git clone https://huggingface.co/arnavs11/DSPy_LongFormQA_Cache')
get_ipython().run_line_magic('cd', 'DSPy_LongFormQA_Cache/')
get_ipython().system('git checkout master')
get_ipython().run_line_magic('cd', '..')
import os
repo_clone_path = '/content/DSPy_LongFormQA_Cache'
if not os.access('/content', os.W_OK):
repo_clone_path = os.path.join(os.getcwd(), 'DSPy_LongFormQA_Cache')
os.environ["DSP_NOTEBOOK_CACHEDIR"] = repo_clone_path
get_ipython().run_line_magic('load_ext', 'autoreload')
get_ipython().run_line_magic('autoreload', '2')
import sys
import os
import regex as re
try: # When on google Colab, let's clone the notebook so we download the cache.
import google.colab
repo_path = 'dspy'
get_ipython().system('git -C $repo_path pull origin || git clone https://github.com/stanfordnlp/dspy $repo_path')
except:
repo_path = '.'
if repo_path not in sys.path:
sys.path.append(repo_path)
import pkg_resources # Install the package if it's not installed
if not "dspy-ai" in {pkg.key for pkg in pkg_resources.working_set}:
get_ipython().system('pip install -U pip')
get_ipython().system('pip install dspy-ai')
get_ipython().system('pip install openai~=0.28.1')
get_ipython().system('pip install -e $repo_path')
import dspy
from dspy.predict import Retry
from dspy.datasets import HotPotQA
from dspy.teleprompt import BootstrapFewShotWithRandomSearch
from dsp.utils import EM, normalize_text
from dspy.primitives.assertions import assert_transform_module, backtrack_handler
get_ipython().run_line_magic('cd', 'dspy/examples/longformqa')
from utils import extract_text_by_citation, correct_citation_format, has_citations, citations_check
colbertv2_wiki17_abstracts = dspy.ColBERTv2(url='http://20.102.90.50:2017/wiki17_abstracts')
dspy.settings.configure(rm=colbertv2_wiki17_abstracts)
turbo = dspy.OpenAI(model='gpt-3.5-turbo', max_tokens=500)
dspy.settings.configure(lm=turbo, trace=[], temperature=0.7)
dataset = HotPotQA(train_seed=1, train_size=300, eval_seed=2023, dev_size=300, test_size=0, keep_details=True)
trainset = [x.with_inputs('question') for x in dataset.train]
devset = [x.with_inputs('question') for x in dataset.dev]
train_example = trainset[0]
print(f"Question: {train_example.question}")
print(f"Answer: {train_example.answer}")
print(f"Relevant Wikipedia Titles: {train_example.gold_titles}")
dev_example = devset[18]
print(f"Question: {dev_example.question}")
print(f"Answer: {dev_example.answer}")
print(f"Relevant Wikipedia Titles: {dev_example.gold_titles}")
from dsp.utils import deduplicate
class GenerateSearchQuery(dspy.Signature):
"""Write a simple search query that will help answer a complex question."""
context = dspy.InputField(desc="may contain relevant facts")
question = dspy.InputField()
query = dspy.OutputField()
class GenerateCitedParagraph(dspy.Signature):
"""Generate a paragraph with citations."""
context = dspy.InputField(desc="may contain relevant facts")
question = dspy.InputField()
paragraph = dspy.OutputField(desc="includes citations")
class LongFormQA(dspy.Module):
def __init__(self, passages_per_hop=3, max_hops=2):
super().__init__()
self.generate_query = [dspy.ChainOfThought(GenerateSearchQuery) for _ in range(max_hops)]
self.retrieve = dspy.Retrieve(k=passages_per_hop)
self.generate_cited_paragraph = dspy.ChainOfThought(GenerateCitedParagraph)
self.max_hops = max_hops
def forward(self, question):
context = []
for hop in range(self.max_hops):
query = self.generate_query[hop](context=context, question=question).query
passages = self.retrieve(query).passages
context = deduplicate(context + passages)
pred = self.generate_cited_paragraph(context=context, question=question)
pred = dspy.Prediction(context=context, paragraph=pred.paragraph)
return pred
class CheckCitationFaithfulness(dspy.Signature):
"""Verify that the text is based on the provided context."""
context = dspy.InputField(desc="may contain relevant facts")
text = dspy.InputField(desc="between 1 to 2 sentences")
faithfulness = dspy.OutputField(desc="boolean indicating if text is faithful to context")
def citation_faithfulness(example, pred, trace):
paragraph, context = pred.paragraph, pred.context
citation_dict = extract_text_by_citation(paragraph)
if not citation_dict:
return False, None
context_dict = {str(i): context[i].split(' | ')[1] for i in range(len(context))}
faithfulness_results = []
unfaithful_citations = []
check_citation_faithfulness = dspy.ChainOfThought(CheckCitationFaithfulness)
for citation_num, texts in citation_dict.items():
if citation_num not in context_dict:
continue
current_context = context_dict[citation_num]
for text in texts:
try:
result = check_citation_faithfulness(context=current_context, text=text)
is_faithful = result.faithfulness.lower() == 'true'
faithfulness_results.append(is_faithful)
if not is_faithful:
unfaithful_citations.append({'paragraph': paragraph, 'text': text, 'context': current_context})
except ValueError as e:
faithfulness_results.append(False)
unfaithful_citations.append({'paragraph': paragraph, 'text': text, 'error': str(e)})
final_faithfulness = all(faithfulness_results)
if not faithfulness_results:
return False, None
return final_faithfulness, unfaithful_citations
def extract_cited_titles_from_paragraph(paragraph, context):
cited_indices = [int(m.group(1)) for m in re.finditer(r'\[(\d+)\]\.', paragraph)]
cited_indices = [index - 1 for index in cited_indices if index <= len(context)]
cited_titles = [context[index].split(' | ')[0] for index in cited_indices]
return cited_titles
def calculate_recall(example, pred, trace=None):
gold_titles = set(example['gold_titles'])
found_cited_titles = set(extract_cited_titles_from_paragraph(pred.paragraph, pred.context))
intersection = gold_titles.intersection(found_cited_titles)
recall = len(intersection) / len(gold_titles) if gold_titles else 0
return recall
def calculate_precision(example, pred, trace=None):
gold_titles = set(example['gold_titles'])
found_cited_titles = set(extract_cited_titles_from_paragraph(pred.paragraph, pred.context))
intersection = gold_titles.intersection(found_cited_titles)
precision = len(intersection) / len(found_cited_titles) if found_cited_titles else 0
return precision
def answer_correctness(example, pred, trace=None):
assert hasattr(example, 'answer'), "Example does not have 'answer'."
normalized_context = normalize_text(pred.paragraph)
if isinstance(example.answer, str):
gold_answers = [example.answer]
elif isinstance(example.answer, list):
gold_answers = example.answer
else:
raise ValueError("'example.answer' is not string or list.")
return 1 if any(normalize_text(answer) in normalized_context for answer in gold_answers) else 0
def evaluate(module):
correctness_values = []
recall_values = []
precision_values = []
citation_faithfulness_values = []
for i in range(len(devset)):
example = devset[i]
try:
pred = module(question=example.question)
correctness_values.append(answer_correctness(example, pred))
citation_faithfulness_score, _ = citation_faithfulness(None, pred, None)
citation_faithfulness_values.append(citation_faithfulness_score)
recall = calculate_recall(example, pred)
precision = calculate_precision(example, pred)
recall_values.append(recall)
precision_values.append(precision)
except Exception as e:
print(f"Failed generation with error: {e}")
average_correctness = sum(correctness_values) / len(devset) if correctness_values else 0
average_recall = sum(recall_values) / len(devset) if recall_values else 0
average_precision = sum(precision_values) / len(devset) if precision_values else 0
average_citation_faithfulness = sum(citation_faithfulness_values) / len(devset) if citation_faithfulness_values else 0
print(f"Average Correctness: {average_correctness}")
print(f"Average Recall: {average_recall}")
print(f"Average Precision: {average_precision}")
print(f"Average Citation Faithfulness: {average_citation_faithfulness}")
longformqa = LongFormQA()
evaluate(longformqa)
question = devset[6].question
pred = longformqa(question)
citation_faithfulness_score, _ = citation_faithfulness(None, pred, None)
print(f"Question: {question}")
print(f"Predicted Paragraph: {pred.paragraph}")
print(f"Citation Faithfulness: {citation_faithfulness_score}")
class LongFormQAWithAssertions(dspy.Module):
def __init__(self, passages_per_hop=3, max_hops=2):
super().__init__()
self.generate_query = [dspy.ChainOfThought(GenerateSearchQuery) for _ in range(max_hops)]
self.retrieve = dspy.Retrieve(k=passages_per_hop)
self.generate_cited_paragraph = dspy.ChainOfThought(GenerateCitedParagraph)
self.max_hops = max_hops
def forward(self, question):
context = []
for hop in range(self.max_hops):
query = self.generate_query[hop](context=context, question=question).query
passages = self.retrieve(query).passages
context = deduplicate(context + passages)
pred = self.generate_cited_paragraph(context=context, question=question)
pred = dspy.Prediction(context=context, paragraph=pred.paragraph)
dspy.Suggest(citations_check(pred.paragraph), f"Make sure every 1-2 sentences has citations. If any 1-2 sentences lack citations, add them in 'text... [x].' format.", target_module=GenerateCitedParagraph)
_, unfaithful_outputs = citation_faithfulness(None, pred, None)
if unfaithful_outputs:
unfaithful_pairs = [(output['text'], output['context']) for output in unfaithful_outputs]
for _, context in unfaithful_pairs:
dspy.Suggest(len(unfaithful_pairs) == 0, f"Make sure your output is based on the following context: '{context}'.", target_module=GenerateCitedParagraph)
else:
return pred
return pred
longformqa_with_assertions = assert_transform_module(LongFormQAWithAssertions().map_named_predictors(Retry), backtrack_handler)
evaluate(longformqa_with_assertions)
question = devset[6].question
pred = longformqa_with_assertions(question)
citation_faithfulness_score, _ = citation_faithfulness(None, pred, None)
print(f"Question: {question}")
print(f"Predicted Paragraph: {pred.paragraph}")
print(f"Citation Faithfulness: {citation_faithfulness_score}")
longformqa = LongFormQA()
teleprompter = BootstrapFewShotWithRandomSearch(metric = answer_correctness, max_bootstrapped_demos=2, num_candidate_programs=6)
cited_longformqa = teleprompter.compile(student = longformqa, teacher = longformqa, trainset=trainset, valset=devset[:100])
evaluate(cited_longformqa)
longformqa = LongFormQA()
teleprompter = | BootstrapFewShotWithRandomSearch(metric = answer_correctness, max_bootstrapped_demos=2, num_candidate_programs=6) | dspy.teleprompt.BootstrapFewShotWithRandomSearch |
get_ipython().system('git clone https://huggingface.co/arnavs11/DSPy_LongFormQA_Cache')
get_ipython().run_line_magic('cd', 'DSPy_LongFormQA_Cache/')
get_ipython().system('git checkout master')
get_ipython().run_line_magic('cd', '..')
import os
repo_clone_path = '/content/DSPy_LongFormQA_Cache'
if not os.access('/content', os.W_OK):
repo_clone_path = os.path.join(os.getcwd(), 'DSPy_LongFormQA_Cache')
os.environ["DSP_NOTEBOOK_CACHEDIR"] = repo_clone_path
get_ipython().run_line_magic('load_ext', 'autoreload')
get_ipython().run_line_magic('autoreload', '2')
import sys
import os
import regex as re
try: # When on google Colab, let's clone the notebook so we download the cache.
import google.colab
repo_path = 'dspy'
get_ipython().system('git -C $repo_path pull origin || git clone https://github.com/stanfordnlp/dspy $repo_path')
except:
repo_path = '.'
if repo_path not in sys.path:
sys.path.append(repo_path)
import pkg_resources # Install the package if it's not installed
if not "dspy-ai" in {pkg.key for pkg in pkg_resources.working_set}:
get_ipython().system('pip install -U pip')
get_ipython().system('pip install dspy-ai')
get_ipython().system('pip install openai~=0.28.1')
get_ipython().system('pip install -e $repo_path')
import dspy
from dspy.predict import Retry
from dspy.datasets import HotPotQA
from dspy.teleprompt import BootstrapFewShotWithRandomSearch
from dsp.utils import EM, normalize_text
from dspy.primitives.assertions import assert_transform_module, backtrack_handler
get_ipython().run_line_magic('cd', 'dspy/examples/longformqa')
from utils import extract_text_by_citation, correct_citation_format, has_citations, citations_check
colbertv2_wiki17_abstracts = dspy.ColBERTv2(url='http://20.102.90.50:2017/wiki17_abstracts')
dspy.settings.configure(rm=colbertv2_wiki17_abstracts)
turbo = dspy.OpenAI(model='gpt-3.5-turbo', max_tokens=500)
dspy.settings.configure(lm=turbo, trace=[], temperature=0.7)
dataset = HotPotQA(train_seed=1, train_size=300, eval_seed=2023, dev_size=300, test_size=0, keep_details=True)
trainset = [x.with_inputs('question') for x in dataset.train]
devset = [x.with_inputs('question') for x in dataset.dev]
train_example = trainset[0]
print(f"Question: {train_example.question}")
print(f"Answer: {train_example.answer}")
print(f"Relevant Wikipedia Titles: {train_example.gold_titles}")
dev_example = devset[18]
print(f"Question: {dev_example.question}")
print(f"Answer: {dev_example.answer}")
print(f"Relevant Wikipedia Titles: {dev_example.gold_titles}")
from dsp.utils import deduplicate
class GenerateSearchQuery(dspy.Signature):
"""Write a simple search query that will help answer a complex question."""
context = dspy.InputField(desc="may contain relevant facts")
question = dspy.InputField()
query = dspy.OutputField()
class GenerateCitedParagraph(dspy.Signature):
"""Generate a paragraph with citations."""
context = dspy.InputField(desc="may contain relevant facts")
question = dspy.InputField()
paragraph = | dspy.OutputField(desc="includes citations") | dspy.OutputField |
get_ipython().run_line_magic('load_ext', 'autoreload')
get_ipython().run_line_magic('autoreload', '2')
import sys
import os
try: # When on google Colab, let's clone the notebook so we download the cache.
import google.colab
repo_path = 'dspy'
get_ipython().system('git -C $repo_path pull origin || git clone https://github.com/stanfordnlp/dspy $repo_path')
except:
repo_path = '.'
if repo_path not in sys.path:
sys.path.append(repo_path)
os.environ["DSP_NOTEBOOK_CACHEDIR"] = os.path.join(repo_path, 'cache')
import pkg_resources # Install the package if it's not installed
if not "dspy-ai" in {pkg.key for pkg in pkg_resources.working_set}:
get_ipython().system('pip install -U pip')
get_ipython().system('pip install -e $repo_path')
get_ipython().system('pip install transformers')
import dspy
from dspy.evaluate import Evaluate
from dspy.teleprompt import BootstrapFewShot, BootstrapFewShotWithRandomSearch, BootstrapFinetune
llama = dspy.HFClientTGI(model="meta-llama/Llama-2-13b-chat-hf", port=[7140, 7141, 7142, 7143], max_tokens=150)
colbertv2 = dspy.ColBERTv2(url='http://20.102.90.50:2017/wiki17_abstracts')
dspy.settings.configure(rm=colbertv2, lm=llama)
train = [('Who was the director of the 2009 movie featuring Peter Outerbridge as William Easton?', 'Kevin Greutert'),
('The heir to the Du Pont family fortune sponsored what wrestling team?', 'Foxcatcher'),
('In what year was the star of To Hell and Back born?', '1925'),
('Which award did the first book of Gary Zukav receive?', 'U.S. National Book Award'),
('What documentary about the Gilgo Beach Killer debuted on A&E?', 'The Killing Season'),
('Which author is English: John Braine or Studs Terkel?', 'John Braine'),
('Who produced the album that included a re-recording of "Lithium"?', 'Butch Vig')]
train = [dspy.Example(question=question, answer=answer).with_inputs('question') for question, answer in train]
dev = [('Who has a broader scope of profession: E. L. Doctorow or Julia Peterkin?', 'E. L. Doctorow'),
('Right Back At It Again contains lyrics co-written by the singer born in what city?', 'Gainesville, Florida'),
('What year was the party of the winner of the 1971 San Francisco mayoral election founded?', '1828'),
('Anthony Dirrell is the brother of which super middleweight title holder?', 'Andre Dirrell'),
('The sports nutrition business established by Oliver Cookson is based in which county in the UK?', 'Cheshire'),
('Find the birth date of the actor who played roles in First Wives Club and Searching for the Elephant.', 'February 13, 1980'),
('Kyle Moran was born in the town on what river?', 'Castletown River'),
("The actress who played the niece in the Priest film was born in what city, country?", 'Surrey, England'),
('Name the movie in which the daughter of Noel Harrison plays Violet Trefusis.', 'Portrait of a Marriage'),
('What year was the father of the Princes in the Tower born?', '1442'),
('What river is near the Crichton Collegiate Church?', 'the River Tyne'),
('Who purchased the team Michael Schumacher raced for in the 1995 Monaco Grand Prix in 2000?', 'Renault'),
('André Zucca was a French photographer who worked with a German propaganda magazine published by what Nazi organization?', 'the Wehrmacht')]
dev = [dspy.Example(question=question, answer=answer).with_inputs('question') for question, answer in dev]
predict = dspy.Predict('question -> answer')
predict(question="What is the capital of Germany?")
class CoT(dspy.Module): # let's define a new module
def __init__(self):
super().__init__()
self.generate_answer = dspy.ChainOfThought('question -> answer')
def forward(self, question):
return self.generate_answer(question=question) # here we use the module
metric_EM = dspy.evaluate.answer_exact_match
teleprompter = BootstrapFewShot(metric=metric_EM, max_bootstrapped_demos=2)
cot_compiled = teleprompter.compile(CoT(), trainset=train)
cot_compiled("What is the capital of Germany?")
llama.inspect_history(n=1)
NUM_THREADS = 32
evaluate_hotpot = Evaluate(devset=dev, metric=metric_EM, num_threads=NUM_THREADS, display_progress=True, display_table=15)
evaluate_hotpot(cot_compiled)
class RAG(dspy.Module):
def __init__(self, num_passages=3):
super().__init__()
self.retrieve = dspy.Retrieve(k=num_passages)
self.generate_query = dspy.ChainOfThought("question -> search_query")
self.generate_answer = dspy.ChainOfThought("context, question -> answer")
def forward(self, question):
search_query = self.generate_query(question=question).search_query
passages = self.retrieve(search_query).passages
return self.generate_answer(context=passages, question=question)
evaluate_hotpot(RAG(), display_table=0)
teleprompter2 = BootstrapFewShotWithRandomSearch(metric=metric_EM, max_bootstrapped_demos=2, num_candidate_programs=8, num_threads=NUM_THREADS)
rag_compiled = teleprompter2.compile(RAG(), trainset=train, valset=dev)
evaluate_hotpot(rag_compiled)
rag_compiled("What year was the party of the winner of the 1971 San Francisco mayoral election founded?")
llama.inspect_history(n=1)
from dsp.utils.utils import deduplicate
class MultiHop(dspy.Module):
def __init__(self, num_passages=3):
super().__init__()
self.retrieve = dspy.Retrieve(k=num_passages)
self.generate_query = | dspy.ChainOfThought("question -> search_query") | dspy.ChainOfThought |
get_ipython().system('git clone https://huggingface.co/arnavs11/DSPy_LongFormQA_Cache')
get_ipython().run_line_magic('cd', 'DSPy_LongFormQA_Cache/')
get_ipython().system('git checkout master')
get_ipython().run_line_magic('cd', '..')
import os
repo_clone_path = '/content/DSPy_LongFormQA_Cache'
if not os.access('/content', os.W_OK):
repo_clone_path = os.path.join(os.getcwd(), 'DSPy_LongFormQA_Cache')
os.environ["DSP_NOTEBOOK_CACHEDIR"] = repo_clone_path
get_ipython().run_line_magic('load_ext', 'autoreload')
get_ipython().run_line_magic('autoreload', '2')
import sys
import os
import regex as re
try: # When on google Colab, let's clone the notebook so we download the cache.
import google.colab
repo_path = 'dspy'
get_ipython().system('git -C $repo_path pull origin || git clone https://github.com/stanfordnlp/dspy $repo_path')
except:
repo_path = '.'
if repo_path not in sys.path:
sys.path.append(repo_path)
import pkg_resources # Install the package if it's not installed
if not "dspy-ai" in {pkg.key for pkg in pkg_resources.working_set}:
get_ipython().system('pip install -U pip')
get_ipython().system('pip install dspy-ai')
get_ipython().system('pip install openai~=0.28.1')
get_ipython().system('pip install -e $repo_path')
import dspy
from dspy.predict import Retry
from dspy.datasets import HotPotQA
from dspy.teleprompt import BootstrapFewShotWithRandomSearch
from dsp.utils import EM, normalize_text
from dspy.primitives.assertions import assert_transform_module, backtrack_handler
get_ipython().run_line_magic('cd', 'dspy/examples/longformqa')
from utils import extract_text_by_citation, correct_citation_format, has_citations, citations_check
colbertv2_wiki17_abstracts = dspy.ColBERTv2(url='http://20.102.90.50:2017/wiki17_abstracts')
dspy.settings.configure(rm=colbertv2_wiki17_abstracts)
turbo = dspy.OpenAI(model='gpt-3.5-turbo', max_tokens=500)
dspy.settings.configure(lm=turbo, trace=[], temperature=0.7)
dataset = HotPotQA(train_seed=1, train_size=300, eval_seed=2023, dev_size=300, test_size=0, keep_details=True)
trainset = [x.with_inputs('question') for x in dataset.train]
devset = [x.with_inputs('question') for x in dataset.dev]
train_example = trainset[0]
print(f"Question: {train_example.question}")
print(f"Answer: {train_example.answer}")
print(f"Relevant Wikipedia Titles: {train_example.gold_titles}")
dev_example = devset[18]
print(f"Question: {dev_example.question}")
print(f"Answer: {dev_example.answer}")
print(f"Relevant Wikipedia Titles: {dev_example.gold_titles}")
from dsp.utils import deduplicate
class GenerateSearchQuery(dspy.Signature):
"""Write a simple search query that will help answer a complex question."""
context = dspy.InputField(desc="may contain relevant facts")
question = dspy.InputField()
query = dspy.OutputField()
class GenerateCitedParagraph(dspy.Signature):
"""Generate a paragraph with citations."""
context = dspy.InputField(desc="may contain relevant facts")
question = dspy.InputField()
paragraph = dspy.OutputField(desc="includes citations")
class LongFormQA(dspy.Module):
def __init__(self, passages_per_hop=3, max_hops=2):
super().__init__()
self.generate_query = [dspy.ChainOfThought(GenerateSearchQuery) for _ in range(max_hops)]
self.retrieve = | dspy.Retrieve(k=passages_per_hop) | dspy.Retrieve |
import glob
import os
import pandas as pd
import random
import dspy
from dspy.evaluate import Evaluate
from dspy.teleprompt import BootstrapFewShotWithRandomSearch
os.environ["DSP_NOTEBOOK_CACHEDIR"] = os.path.join('.', 'cache')
turbo = dspy.OpenAI(model='gpt-3.5-turbo-1106', max_tokens=250, model_type='chat')
dspy.settings.configure(lm=turbo)
gpt4T = dspy.OpenAI(model='gpt-4-1106-preview', max_tokens=350, model_type='chat')
RUN_FROM_SCRATCH = False
get_ipython().system('git clone https://github.com/selenashe/ScoNe.git')
def load_scone(dirname):
dfs = []
for filename in glob.glob(dirname + "/*.csv"):
df = pd.read_csv(filename, index_col=0)
df['category'] = os.path.basename(filename).replace(".csv", "")
dfs.append(df)
data_df = pd.concat(dfs)
def as_example(row):
suffix = '' if row['category'] == 'one_scoped' else '_edited'
hkey = 'sentence2' + suffix
question = row[hkey][0].lower() + row[hkey][1: ].strip(".")
question = f"Can we logically conclude for sure that {question}?"
label = "Yes" if row['gold_label' + suffix] == 'entailment' else "No"
return dspy.Example({
"context": row['sentence1' + suffix],
"question": question,
"answer": label,
"category": row['category']
}).with_inputs("context", "question")
return list(data_df.apply(as_example, axis=1).values)
all_train = load_scone("ScoNe/scone_nli/train")
random.seed(1)
random.shuffle(all_train)
train, dev = all_train[: 200], all_train[200: 250]
len(train), len(dev)
random.seed(1)
test = load_scone(dirname=f"ScoNe/scone_nli/test")
test = [ex for ex in test if ex.category == "one_scoped"]
pd.Series([ex.answer for ex in test]).value_counts()
scone_accuracy = dspy.evaluate.metrics.answer_exact_match
evaluator = Evaluate(devset=test, num_threads=1, display_progress=True, display_table=0)
class ScoNeSignature(dspy.Signature):
("""You are given some context (a premise) and a question (a hypothesis). """
"""You must indicate with Yes/No answer whether we can logically """
"""conclude the hypothesis from the premise.""")
context = dspy.InputField()
question = dspy.InputField()
answer = | dspy.OutputField(desc="Yes or No") | dspy.OutputField |
get_ipython().run_line_magic('load_ext', 'autoreload')
get_ipython().run_line_magic('autoreload', '2')
import sys
import os
try: # When on google Colab, let's clone the notebook so we download the cache.
import google.colab
repo_path = 'dspy'
get_ipython().system('git -C $repo_path pull origin || git clone https://github.com/stanfordnlp/dspy $repo_path')
except:
repo_path = '.'
if repo_path not in sys.path:
sys.path.append(repo_path)
os.environ["DSP_NOTEBOOK_CACHEDIR"] = os.path.join(repo_path, 'cache')
import pkg_resources # Install the package if it's not installed
if not "dspy-ai" in {pkg.key for pkg in pkg_resources.working_set}:
get_ipython().system('pip install -U pip')
get_ipython().system('pip install -e $repo_path')
get_ipython().system('pip install transformers')
import dspy
from dspy.evaluate import Evaluate
from dspy.teleprompt import BootstrapFewShot, BootstrapFewShotWithRandomSearch, BootstrapFinetune
llama = dspy.HFClientTGI(model="meta-llama/Llama-2-13b-chat-hf", port=[7140, 7141, 7142, 7143], max_tokens=150)
colbertv2 = dspy.ColBERTv2(url='http://20.102.90.50:2017/wiki17_abstracts')
dspy.settings.configure(rm=colbertv2, lm=llama)
train = [('Who was the director of the 2009 movie featuring Peter Outerbridge as William Easton?', 'Kevin Greutert'),
('The heir to the Du Pont family fortune sponsored what wrestling team?', 'Foxcatcher'),
('In what year was the star of To Hell and Back born?', '1925'),
('Which award did the first book of Gary Zukav receive?', 'U.S. National Book Award'),
('What documentary about the Gilgo Beach Killer debuted on A&E?', 'The Killing Season'),
('Which author is English: John Braine or Studs Terkel?', 'John Braine'),
('Who produced the album that included a re-recording of "Lithium"?', 'Butch Vig')]
train = [dspy.Example(question=question, answer=answer).with_inputs('question') for question, answer in train]
dev = [('Who has a broader scope of profession: E. L. Doctorow or Julia Peterkin?', 'E. L. Doctorow'),
('Right Back At It Again contains lyrics co-written by the singer born in what city?', 'Gainesville, Florida'),
('What year was the party of the winner of the 1971 San Francisco mayoral election founded?', '1828'),
('Anthony Dirrell is the brother of which super middleweight title holder?', 'Andre Dirrell'),
('The sports nutrition business established by Oliver Cookson is based in which county in the UK?', 'Cheshire'),
('Find the birth date of the actor who played roles in First Wives Club and Searching for the Elephant.', 'February 13, 1980'),
('Kyle Moran was born in the town on what river?', 'Castletown River'),
("The actress who played the niece in the Priest film was born in what city, country?", 'Surrey, England'),
('Name the movie in which the daughter of Noel Harrison plays Violet Trefusis.', 'Portrait of a Marriage'),
('What year was the father of the Princes in the Tower born?', '1442'),
('What river is near the Crichton Collegiate Church?', 'the River Tyne'),
('Who purchased the team Michael Schumacher raced for in the 1995 Monaco Grand Prix in 2000?', 'Renault'),
('André Zucca was a French photographer who worked with a German propaganda magazine published by what Nazi organization?', 'the Wehrmacht')]
dev = [dspy.Example(question=question, answer=answer).with_inputs('question') for question, answer in dev]
predict = dspy.Predict('question -> answer')
predict(question="What is the capital of Germany?")
class CoT(dspy.Module): # let's define a new module
def __init__(self):
super().__init__()
self.generate_answer = dspy.ChainOfThought('question -> answer')
def forward(self, question):
return self.generate_answer(question=question) # here we use the module
metric_EM = dspy.evaluate.answer_exact_match
teleprompter = BootstrapFewShot(metric=metric_EM, max_bootstrapped_demos=2)
cot_compiled = teleprompter.compile(CoT(), trainset=train)
cot_compiled("What is the capital of Germany?")
llama.inspect_history(n=1)
NUM_THREADS = 32
evaluate_hotpot = Evaluate(devset=dev, metric=metric_EM, num_threads=NUM_THREADS, display_progress=True, display_table=15)
evaluate_hotpot(cot_compiled)
class RAG(dspy.Module):
def __init__(self, num_passages=3):
super().__init__()
self.retrieve = | dspy.Retrieve(k=num_passages) | dspy.Retrieve |
import dspy
from dsp.utils import deduplicate
from dspy.datasets import HotPotQA
from dspy.predict.retry import Retry
from dspy.teleprompt import BootstrapFewShot, BootstrapFewShotWithRandomSearch
from dspy.evaluate.evaluate import Evaluate
from dspy.primitives.assertions import assert_transform_module, backtrack_handler
import os
import openai
openai.api_key = os.getenv('OPENAI_API_KEY')
colbertv2_wiki17_abstracts = dspy.ColBERTv2(url='http://20.102.90.50:2017/wiki17_abstracts')
dspy.settings.configure(rm=colbertv2_wiki17_abstracts)
turbo = dspy.OpenAI(model='gpt-3.5-turbo', max_tokens=500)
dspy.settings.configure(lm=turbo, trace=[], temperature=0.7)
dataset = HotPotQA(train_seed=1, train_size=300, eval_seed=2023, dev_size=300, test_size=0)
trainset = [x.with_inputs('question') for x in dataset.train]
devset = [x.with_inputs('question') for x in dataset.dev]
def validate_query_distinction_local(previous_queries, query):
"""check if query is distinct from previous queries"""
if previous_queries == []:
return True
if dspy.evaluate.answer_exact_match_str(query, previous_queries, frac=0.8):
return False
return True
def validate_context_and_answer_and_hops(example, pred, trace=None):
if not | dspy.evaluate.answer_exact_match(example, pred) | dspy.evaluate.answer_exact_match |
get_ipython().system('git clone https://huggingface.co/arnavs11/DSPy_TweetGen_Cache')
get_ipython().run_line_magic('cd', 'DSPy_TweetGen_Cache/')
get_ipython().system('git checkout master')
get_ipython().run_line_magic('cd', '..')
import os
repo_clone_path = '/content/DSPy_TweetGen_Cache'
if not os.access('/content', os.W_OK):
repo_clone_path = os.path.join(os.getcwd(), 'DSPy_TweetGen_Cache')
os.environ["DSP_NOTEBOOK_CACHEDIR"] = repo_clone_path
get_ipython().run_line_magic('load_ext', 'autoreload')
get_ipython().run_line_magic('autoreload', '2')
import sys
import os
import regex as re
import json
try: # When on google Colab, let's clone the notebook so we download the cache.
import google.colab
repo_path = 'dspy'
get_ipython().system('git -C $repo_path pull origin || git clone https://github.com/stanfordnlp/dspy $repo_path')
except:
repo_path = '.'
if repo_path not in sys.path:
sys.path.append(repo_path)
import pkg_resources # Install the package if it's not installed
if not "dspy-ai" in {pkg.key for pkg in pkg_resources.working_set}:
get_ipython().system('pip install -U pip')
get_ipython().system('pip install dspy-ai')
get_ipython().system('pip install openai~=0.28.1')
get_ipython().system('pip install -e $repo_path')
import dspy
from dspy.predict import Retry
from dspy.datasets import HotPotQA
from dspy.teleprompt import BootstrapFewShotWithRandomSearch
from dsp.utils import deduplicate
from dspy.evaluate.evaluate import Evaluate
from dspy.primitives.assertions import assert_transform_module, backtrack_handler
colbertv2_wiki17_abstracts = dspy.ColBERTv2(url='http://20.102.90.50:2017/wiki17_abstracts')
dspy.settings.configure(rm=colbertv2_wiki17_abstracts)
turbo = dspy.OpenAI(model='gpt-3.5-turbo', max_tokens=500)
dspy.settings.configure(lm=turbo, trace=[], temperature=0.7)
dataset = HotPotQA(train_seed=1, train_size=300, eval_seed=2023, dev_size=300, test_size=0, keep_details=True)
trainset = [x.with_inputs('question', 'answer') for x in dataset.train]
devset = [x.with_inputs('question', 'answer') for x in dataset.dev]
class GenerateSearchQuery(dspy.Signature):
"""Write a simple search query that will help answer a complex question."""
context = dspy.InputField(desc="may contain relevant facts")
question = dspy.InputField()
query = dspy.OutputField()
class GenerateTweet(dspy.Signature):
"""Generate an engaging tweet that effectively answers a question staying faithful to the context, is less than 280 characters, and has no hashtags."""
question = dspy.InputField()
context = dspy.InputField(desc="may contain relevant facts")
tweet = dspy.OutputField()
class Tweeter(dspy.Module):
def __init__(self):
super().__init__()
self.generate_tweet = dspy.ChainOfThought(GenerateTweet)
def forward(self, question, answer):
context = []
max_hops=2
passages_per_hop=3
generate_query = [dspy.ChainOfThought(GenerateSearchQuery) for _ in range(max_hops)]
retrieve = dspy.Retrieve(k=passages_per_hop)
for hop in range(max_hops):
query = generate_query[hop](context=context, question=question).query
passages = retrieve(query).passages
context = deduplicate(context + passages)
generated_tweet = self.generate_tweet(question=question, context=context).tweet
return dspy.Prediction(generated_tweet=generated_tweet, context=context)
tweeter = Tweeter()
def has_no_hashtags(text):
return len(re.findall(r"#\w+", text)) == 0
def is_within_length_limit(text, length_limit=280):
return len(text) <= length_limit
def is_assessment_yes(assessment_answer):
"""Check if the first word of the assessment answer is 'yes'."""
return assessment_answer.split()[0].lower() == 'yes'
def has_correct_answer(text, answer):
return answer in text
class AssessTweet(dspy.Signature):
"""Assess the quality of a tweet along the specified dimension."""
context = dspy.InputField(desc='ignore if N/A')
assessed_text = dspy.InputField()
assessment_question = dspy.InputField()
assessment_answer = dspy.OutputField(desc="Yes or No")
def no_hashtags_metric(gold, pred, trace=None):
tweet = pred.generated_tweet
no_hashtags = has_no_hashtags(tweet)
score = no_hashtags
return score
def is_correct_metric(gold, pred, trace=None):
answer, tweet = gold.answer, pred.generated_tweet
correct = has_correct_answer(tweet, answer)
score = correct
return score
def within_length_metric(gold, pred, trace=None):
tweet = pred.generated_tweet
within_length_limit = is_within_length_limit(tweet, 280)
score = within_length_limit
return score
def engaging_metric(gold, pred, trace=None):
tweet = pred.generated_tweet
engaging = "Does the assessed text make for a self-contained, engaging tweet? Say no if it is not engaging."
engaging = dspy.Predict(AssessTweet)(context='N/A', assessed_text=tweet, assessment_question=engaging)
engaging = engaging.assessment_answer.split()[0].lower() == 'yes'
score = engaging
return score
def faithful_metric(gold, pred, trace=None):
context, tweet = pred.context, pred.generated_tweet
faithful = "Is the assessed text grounded in the context? Say no if it includes significant facts not in the context."
faithful = dspy.Predict(AssessTweet)(context=context, assessed_text=tweet, assessment_question=faithful)
faithful = faithful.assessment_answer.split()[0].lower() == 'yes'
score = faithful
return score
def overall_metric(gold, pred, trace=None):
answer, context, tweet = gold.answer, pred.context, pred.generated_tweet
no_hashtags = has_no_hashtags(tweet)
within_length_limit = is_within_length_limit(tweet, 280)
correct = has_correct_answer(tweet, answer)
engaging = "Does the assessed text make for a self-contained, engaging tweet? Say no if it is not engaging."
faithful = "Is the assessed text grounded in the context? Say no if it includes significant facts not in the context."
faithful = dspy.Predict(AssessTweet)(context=context, assessed_text=tweet, assessment_question=faithful)
engaging = dspy.Predict(AssessTweet)(context='N/A', assessed_text=tweet, assessment_question=engaging)
engaging, faithful = [m.assessment_answer.split()[0].lower() == 'yes' for m in [engaging, faithful]]
score = (correct + engaging + faithful + no_hashtags + within_length_limit) if correct and within_length_limit else 0
return score / 5.0
metrics = [no_hashtags_metric, is_correct_metric, within_length_metric, engaging_metric, faithful_metric, overall_metric]
for metric in metrics:
evaluate = Evaluate(metric=metric, devset=devset, num_threads=1, display_progress=True, display_table=5)
evaluate(tweeter)
example = devset[10]
tweet = tweeter(question=example.question, answer = example.answer)
print(f'Generated Tweet: ', tweet.generated_tweet)
tweet.context
for metric in metrics:
evaluate = Evaluate(metric=metric, devset=devset[10:11], num_threads=1, display_progress=True, display_table=5)
evaluate(tweeter)
class TweeterWithAssertions(dspy.Module):
def __init__(self):
super().__init__()
self.generate_tweet = dspy.ChainOfThought(GenerateTweet)
def forward(self, question, answer):
context = []
max_hops=2
passages_per_hop=3
generate_query = [dspy.ChainOfThought(GenerateSearchQuery) for _ in range(max_hops)]
retrieve = dspy.Retrieve(k=passages_per_hop)
for hop in range(max_hops):
query = generate_query[hop](context=context, question=question).query
passages = retrieve(query).passages
context = deduplicate(context + passages)
generated_tweet = self.generate_tweet(question=question, context=context).tweet
dspy.Suggest(has_no_hashtags(generated_tweet), f"Please revise the tweet to remove hashtag phrases following it.", target_module=GenerateTweet)
dspy.Suggest(is_within_length_limit(generated_tweet, 280), f"Please ensure the tweet is within {280} characters.", target_module=GenerateTweet)
dspy.Suggest(has_correct_answer(generated_tweet, answer), "The tweet does not include the correct answer to the question. Please revise accordingly.", target_module=GenerateTweet)
engaging_question = "Does the assessed text make for a self-contained, engaging tweet? Say no if it is not engaging."
engaging_assessment = dspy.Predict(AssessTweet)(context=context, assessed_text=generated_tweet, assessment_question=engaging_question)
dspy.Suggest(is_assessment_yes(engaging_assessment.assessment_answer), "The text is not engaging enough. Please revise to make it more captivating.", target_module=GenerateTweet)
faithful_question = "Is the assessed text grounded in the context? Say no if it includes significant facts not in the context."
faithful_assessment = dspy.Predict(AssessTweet)(context='N/A', assessed_text=generated_tweet, assessment_question=faithful_question)
dspy.Suggest(is_assessment_yes(faithful_assessment.assessment_answer), "The text contains unfaithful elements or significant facts not in the context. Please revise for accuracy.", target_module=GenerateTweet)
return dspy.Prediction(generated_tweet=generated_tweet, context=context)
tweeter_with_assertions = assert_transform_module(TweeterWithAssertions().map_named_predictors(Retry), backtrack_handler)
metrics = [no_hashtags_metric, is_correct_metric, within_length_metric, engaging_metric, faithful_metric, overall_metric]
for metric in metrics:
evaluate = Evaluate(metric=metric, devset=devset, num_threads=1, display_progress=True, display_table=5)
evaluate(tweeter_with_assertions)
example = devset[10]
tweet = tweeter_with_assertions(question=example.question, answer = example.answer)
print(f'Generated Tweet: ', tweet.generated_tweet)
tweet.context
for metric in metrics:
evaluate = Evaluate(metric=metric, devset=devset[10:11], num_threads=1, display_progress=True, display_table=5)
evaluate(tweeter_with_assertions)
teleprompter = BootstrapFewShotWithRandomSearch(metric = overall_metric, max_bootstrapped_demos=2, num_candidate_programs=6)
compiled_tweeter = teleprompter.compile(student = tweeter, teacher = tweeter, trainset=trainset, valset=devset[:100])
for metric in metrics:
evaluate = | Evaluate(metric=metric, devset=devset, num_threads=1, display_progress=True, display_table=5) | dspy.evaluate.evaluate.Evaluate |
import dspy
from dspy.evaluate.evaluate import Evaluate
from dspy.teleprompt import BootstrapFewShotWithRandomSearch
colbertv2 = dspy.ColBERTv2(url='http://20.102.90.50:2017/wiki17_abstracts')
dspy.configure(rm=colbertv2)
from langchain_openai import OpenAI
from langchain.globals import set_llm_cache
from langchain.cache import SQLiteCache
set_llm_cache(SQLiteCache(database_path="cache.db"))
llm = OpenAI(model_name="gpt-3.5-turbo-instruct", temperature=0)
retrieve = lambda x: dspy.Retrieve(k=5)(x["question"]).passages
from langchain_core.prompts import PromptTemplate
from langchain_core.output_parsers import StrOutputParser
from langchain_core.runnables import RunnablePassthrough
prompt = PromptTemplate.from_template("Given {context}, answer the question `{question}` as a tweet.")
vanilla_chain = RunnablePassthrough.assign(context=retrieve) | prompt | llm | StrOutputParser()
from dspy.predict.langchain import LangChainPredict, LangChainModule
zeroshot_chain = RunnablePassthrough.assign(context=retrieve) | | LangChainPredict(prompt, llm) | dspy.predict.langchain.LangChainPredict |
get_ipython().run_line_magic('load_ext', 'autoreload')
get_ipython().run_line_magic('autoreload', '2')
import sys
import os
try: # When on google Colab, let's clone the notebook so we download the cache.
import google.colab
repo_path = 'dspy'
get_ipython().system('git -C $repo_path pull origin || git clone https://github.com/stanfordnlp/dspy $repo_path')
except:
repo_path = '.'
if repo_path not in sys.path:
sys.path.append(repo_path)
os.environ["DSP_NOTEBOOK_CACHEDIR"] = os.path.join(repo_path, 'cache')
import dspy
get_ipython().run_line_magic('load_ext', 'autoreload')
get_ipython().run_line_magic('autoreload', '2')
import sys; sys.path.append('/future/u/okhattab/repos/public/stanfordnlp/dspy')
import dspy
from dspy.evaluate import Evaluate
from dspy.datasets.hotpotqa import HotPotQA
from dspy.teleprompt import BootstrapFewShot, BootstrapFewShotWithRandomSearch, BootstrapFinetune
llama = | dspy.HFClientTGI(model="meta-llama/Llama-2-13b-chat-hf", port=[7140, 7141, 7142, 7143], max_tokens=150) | dspy.HFClientTGI |
import dspy
from dsp.utils import deduplicate
from dspy.datasets import HotPotQA
from dspy.predict.retry import Retry
from dspy.teleprompt import BootstrapFewShot, BootstrapFewShotWithRandomSearch
from dspy.evaluate.evaluate import Evaluate
from dspy.primitives.assertions import assert_transform_module, backtrack_handler
import os
import openai
openai.api_key = os.getenv('OPENAI_API_KEY')
colbertv2_wiki17_abstracts = dspy.ColBERTv2(url='http://20.102.90.50:2017/wiki17_abstracts')
dspy.settings.configure(rm=colbertv2_wiki17_abstracts)
turbo = dspy.OpenAI(model='gpt-3.5-turbo', max_tokens=500)
dspy.settings.configure(lm=turbo, trace=[], temperature=0.7)
dataset = HotPotQA(train_seed=1, train_size=300, eval_seed=2023, dev_size=300, test_size=0)
trainset = [x.with_inputs('question') for x in dataset.train]
devset = [x.with_inputs('question') for x in dataset.dev]
def validate_query_distinction_local(previous_queries, query):
"""check if query is distinct from previous queries"""
if previous_queries == []:
return True
if dspy.evaluate.answer_exact_match_str(query, previous_queries, frac=0.8):
return False
return True
def validate_context_and_answer_and_hops(example, pred, trace=None):
if not dspy.evaluate.answer_exact_match(example, pred):
return False
if not dspy.evaluate.answer_passage_match(example, pred):
return False
return True
def gold_passages_retrieved(example, pred, trace=None):
gold_titles = set(map(dspy.evaluate.normalize_text, example['gold_titles']))
found_titles = set(map(dspy.evaluate.normalize_text, [c.split(' | ')[0] for c in pred.context]))
return gold_titles.issubset(found_titles)
class GenerateAnswer(dspy.Signature):
"""Answer questions with short factoid answers."""
context = dspy.InputField(desc="may contain relevant facts")
question = dspy.InputField()
answer = dspy.OutputField(desc="often between 1 and 5 words")
class GenerateSearchQuery(dspy.Signature):
"""Write a simple search query that will help answer a complex question."""
context = dspy.InputField(desc="may contain relevant facts")
question = dspy.InputField()
query = dspy.OutputField()
def all_queries_distinct(prev_queries):
query_distinct = True
for i, query in enumerate(prev_queries):
if validate_query_distinction_local(prev_queries[:i], query) == False:
query_distinct = False
break
return query_distinct
class SimplifiedBaleen(dspy.Module):
def __init__(self, passages_per_hop=2, max_hops=2):
super().__init__()
self.generate_query = [dspy.ChainOfThought(GenerateSearchQuery) for _ in range(max_hops)]
self.retrieve = dspy.Retrieve(k=passages_per_hop)
self.generate_answer = dspy.ChainOfThought(GenerateAnswer)
self.max_hops = max_hops
self.passed_suggestions = 0
def forward(self, question):
context = []
prev_queries = [question]
for hop in range(self.max_hops):
query = self.generate_query[hop](context=context, question=question).query
prev_queries.append(query)
passages = self.retrieve(query).passages
context = deduplicate(context + passages)
if all_queries_distinct(prev_queries):
self.passed_suggestions += 1
pred = self.generate_answer(context=context, question=question)
pred = dspy.Prediction(context=context, answer=pred.answer)
return pred
class SimplifiedBaleenAssertions(dspy.Module):
def __init__(self, passages_per_hop=2, max_hops=2):
super().__init__()
self.generate_query = [dspy.ChainOfThought(GenerateSearchQuery) for _ in range(max_hops)]
self.retrieve = dspy.Retrieve(k=passages_per_hop)
self.generate_answer = dspy.ChainOfThought(GenerateAnswer)
self.max_hops = max_hops
self.passed_suggestions = 0
def forward(self, question):
context = []
prev_queries = [question]
for hop in range(self.max_hops):
query = self.generate_query[hop](context=context, question=question).query
dspy.Suggest(
len(query) <= 100,
"Query should be short and less than 100 characters",
)
dspy.Suggest(
validate_query_distinction_local(prev_queries, query),
"Query should be distinct from: "
+ "; ".join(f"{i+1}) {q}" for i, q in enumerate(prev_queries)),
)
prev_queries.append(query)
passages = self.retrieve(query).passages
context = deduplicate(context + passages)
if all_queries_distinct(prev_queries):
self.passed_suggestions += 1
pred = self.generate_answer(context=context, question=question)
pred = dspy.Prediction(context=context, answer=pred.answer)
return pred
evaluate_on_hotpotqa = | Evaluate(devset=devset, num_threads=10, display_progress=True, display_table=False) | dspy.evaluate.evaluate.Evaluate |
get_ipython().run_line_magic('load_ext', 'autoreload')
get_ipython().run_line_magic('autoreload', '2')
import dspy
from dspy.evaluate import Evaluate
from dspy.datasets.hotpotqa import HotPotQA
from dspy.teleprompt import BootstrapFewShotWithRandomSearch, BootstrapFinetune
ports = [7140, 7141, 7142, 7143, 7144, 7145]
llamaChat = dspy.HFClientTGI(model="meta-llama/Llama-2-13b-chat-hf", port=ports, max_tokens=150)
colbertv2 = dspy.ColBERTv2(url='http://20.102.90.50:2017/wiki17_abstracts')
dspy.settings.configure(rm=colbertv2, lm=llamaChat)
dataset = HotPotQA(train_seed=1, train_size=200, eval_seed=2023, dev_size=1000, test_size=0)
trainset = [x.with_inputs('question') for x in dataset.train]
devset = [x.with_inputs('question') for x in dataset.dev]
testset = [x.with_inputs('question') for x in dataset.test]
len(trainset), len(devset), len(testset)
trainset[0]
from dsp.utils.utils import deduplicate
class BasicMH(dspy.Module):
def __init__(self, passages_per_hop=3):
super().__init__()
self.retrieve = dspy.Retrieve(k=passages_per_hop)
self.generate_query = [dspy.ChainOfThought("context, question -> search_query") for _ in range(2)]
self.generate_answer = dspy.ChainOfThought("context, question -> answer")
def forward(self, question):
context = []
for hop in range(2):
search_query = self.generate_query[hop](context=context, question=question).search_query
passages = self.retrieve(search_query).passages
context = deduplicate(context + passages)
return self.generate_answer(context=context, question=question).copy(context=context)
RECOMPILE_INTO_LLAMA_FROM_SCRATCH = False
NUM_THREADS = 24
metric_EM = dspy.evaluate.answer_exact_match
if RECOMPILE_INTO_LLAMA_FROM_SCRATCH:
tp = BootstrapFewShotWithRandomSearch(metric=metric_EM, max_bootstrapped_demos=2, num_threads=NUM_THREADS)
basicmh_bs = tp.compile(BasicMH(), trainset=trainset[:50], valset=trainset[50:200])
ensemble = [prog for *_, prog in basicmh_bs.candidate_programs[:4]]
for idx, prog in enumerate(ensemble):
pass
if not RECOMPILE_INTO_LLAMA_FROM_SCRATCH:
ensemble = []
for idx in range(4):
prog = BasicMH()
prog.load(f'multihop_llama213b_{idx}.json')
ensemble.append(prog)
llama_program = ensemble[0]
evaluate_hotpot = Evaluate(devset=devset[:1000], metric=metric_EM, num_threads=NUM_THREADS, display_progress=True, display_table=0)
evaluate_hotpot(llama_program)
llama_program(question="How many storeys are in the castle that David Gregory inherited?")
llamaChat.inspect_history(n=3)
unlabeled_train = | HotPotQA(train_seed=1, train_size=3000, eval_seed=2023, dev_size=0, test_size=0) | dspy.datasets.hotpotqa.HotPotQA |
get_ipython().system('git clone https://huggingface.co/arnavs11/DSPy_TweetGen_Cache')
get_ipython().run_line_magic('cd', 'DSPy_TweetGen_Cache/')
get_ipython().system('git checkout master')
get_ipython().run_line_magic('cd', '..')
import os
repo_clone_path = '/content/DSPy_TweetGen_Cache'
if not os.access('/content', os.W_OK):
repo_clone_path = os.path.join(os.getcwd(), 'DSPy_TweetGen_Cache')
os.environ["DSP_NOTEBOOK_CACHEDIR"] = repo_clone_path
get_ipython().run_line_magic('load_ext', 'autoreload')
get_ipython().run_line_magic('autoreload', '2')
import sys
import os
import regex as re
import json
try: # When on google Colab, let's clone the notebook so we download the cache.
import google.colab
repo_path = 'dspy'
get_ipython().system('git -C $repo_path pull origin || git clone https://github.com/stanfordnlp/dspy $repo_path')
except:
repo_path = '.'
if repo_path not in sys.path:
sys.path.append(repo_path)
import pkg_resources # Install the package if it's not installed
if not "dspy-ai" in {pkg.key for pkg in pkg_resources.working_set}:
get_ipython().system('pip install -U pip')
get_ipython().system('pip install dspy-ai')
get_ipython().system('pip install openai~=0.28.1')
get_ipython().system('pip install -e $repo_path')
import dspy
from dspy.predict import Retry
from dspy.datasets import HotPotQA
from dspy.teleprompt import BootstrapFewShotWithRandomSearch
from dsp.utils import deduplicate
from dspy.evaluate.evaluate import Evaluate
from dspy.primitives.assertions import assert_transform_module, backtrack_handler
colbertv2_wiki17_abstracts = dspy.ColBERTv2(url='http://20.102.90.50:2017/wiki17_abstracts')
dspy.settings.configure(rm=colbertv2_wiki17_abstracts)
turbo = dspy.OpenAI(model='gpt-3.5-turbo', max_tokens=500)
dspy.settings.configure(lm=turbo, trace=[], temperature=0.7)
dataset = HotPotQA(train_seed=1, train_size=300, eval_seed=2023, dev_size=300, test_size=0, keep_details=True)
trainset = [x.with_inputs('question', 'answer') for x in dataset.train]
devset = [x.with_inputs('question', 'answer') for x in dataset.dev]
class GenerateSearchQuery(dspy.Signature):
"""Write a simple search query that will help answer a complex question."""
context = dspy.InputField(desc="may contain relevant facts")
question = dspy.InputField()
query = dspy.OutputField()
class GenerateTweet(dspy.Signature):
"""Generate an engaging tweet that effectively answers a question staying faithful to the context, is less than 280 characters, and has no hashtags."""
question = dspy.InputField()
context = dspy.InputField(desc="may contain relevant facts")
tweet = dspy.OutputField()
class Tweeter(dspy.Module):
def __init__(self):
super().__init__()
self.generate_tweet = dspy.ChainOfThought(GenerateTweet)
def forward(self, question, answer):
context = []
max_hops=2
passages_per_hop=3
generate_query = [dspy.ChainOfThought(GenerateSearchQuery) for _ in range(max_hops)]
retrieve = dspy.Retrieve(k=passages_per_hop)
for hop in range(max_hops):
query = generate_query[hop](context=context, question=question).query
passages = retrieve(query).passages
context = deduplicate(context + passages)
generated_tweet = self.generate_tweet(question=question, context=context).tweet
return dspy.Prediction(generated_tweet=generated_tweet, context=context)
tweeter = Tweeter()
def has_no_hashtags(text):
return len(re.findall(r"#\w+", text)) == 0
def is_within_length_limit(text, length_limit=280):
return len(text) <= length_limit
def is_assessment_yes(assessment_answer):
"""Check if the first word of the assessment answer is 'yes'."""
return assessment_answer.split()[0].lower() == 'yes'
def has_correct_answer(text, answer):
return answer in text
class AssessTweet(dspy.Signature):
"""Assess the quality of a tweet along the specified dimension."""
context = dspy.InputField(desc='ignore if N/A')
assessed_text = dspy.InputField()
assessment_question = dspy.InputField()
assessment_answer = dspy.OutputField(desc="Yes or No")
def no_hashtags_metric(gold, pred, trace=None):
tweet = pred.generated_tweet
no_hashtags = has_no_hashtags(tweet)
score = no_hashtags
return score
def is_correct_metric(gold, pred, trace=None):
answer, tweet = gold.answer, pred.generated_tweet
correct = has_correct_answer(tweet, answer)
score = correct
return score
def within_length_metric(gold, pred, trace=None):
tweet = pred.generated_tweet
within_length_limit = is_within_length_limit(tweet, 280)
score = within_length_limit
return score
def engaging_metric(gold, pred, trace=None):
tweet = pred.generated_tweet
engaging = "Does the assessed text make for a self-contained, engaging tweet? Say no if it is not engaging."
engaging = dspy.Predict(AssessTweet)(context='N/A', assessed_text=tweet, assessment_question=engaging)
engaging = engaging.assessment_answer.split()[0].lower() == 'yes'
score = engaging
return score
def faithful_metric(gold, pred, trace=None):
context, tweet = pred.context, pred.generated_tweet
faithful = "Is the assessed text grounded in the context? Say no if it includes significant facts not in the context."
faithful = dspy.Predict(AssessTweet)(context=context, assessed_text=tweet, assessment_question=faithful)
faithful = faithful.assessment_answer.split()[0].lower() == 'yes'
score = faithful
return score
def overall_metric(gold, pred, trace=None):
answer, context, tweet = gold.answer, pred.context, pred.generated_tweet
no_hashtags = has_no_hashtags(tweet)
within_length_limit = is_within_length_limit(tweet, 280)
correct = has_correct_answer(tweet, answer)
engaging = "Does the assessed text make for a self-contained, engaging tweet? Say no if it is not engaging."
faithful = "Is the assessed text grounded in the context? Say no if it includes significant facts not in the context."
faithful = dspy.Predict(AssessTweet)(context=context, assessed_text=tweet, assessment_question=faithful)
engaging = dspy.Predict(AssessTweet)(context='N/A', assessed_text=tweet, assessment_question=engaging)
engaging, faithful = [m.assessment_answer.split()[0].lower() == 'yes' for m in [engaging, faithful]]
score = (correct + engaging + faithful + no_hashtags + within_length_limit) if correct and within_length_limit else 0
return score / 5.0
metrics = [no_hashtags_metric, is_correct_metric, within_length_metric, engaging_metric, faithful_metric, overall_metric]
for metric in metrics:
evaluate = Evaluate(metric=metric, devset=devset, num_threads=1, display_progress=True, display_table=5)
evaluate(tweeter)
example = devset[10]
tweet = tweeter(question=example.question, answer = example.answer)
print(f'Generated Tweet: ', tweet.generated_tweet)
tweet.context
for metric in metrics:
evaluate = Evaluate(metric=metric, devset=devset[10:11], num_threads=1, display_progress=True, display_table=5)
evaluate(tweeter)
class TweeterWithAssertions(dspy.Module):
def __init__(self):
super().__init__()
self.generate_tweet = dspy.ChainOfThought(GenerateTweet)
def forward(self, question, answer):
context = []
max_hops=2
passages_per_hop=3
generate_query = [dspy.ChainOfThought(GenerateSearchQuery) for _ in range(max_hops)]
retrieve = dspy.Retrieve(k=passages_per_hop)
for hop in range(max_hops):
query = generate_query[hop](context=context, question=question).query
passages = retrieve(query).passages
context = deduplicate(context + passages)
generated_tweet = self.generate_tweet(question=question, context=context).tweet
dspy.Suggest(has_no_hashtags(generated_tweet), f"Please revise the tweet to remove hashtag phrases following it.", target_module=GenerateTweet)
dspy.Suggest(is_within_length_limit(generated_tweet, 280), f"Please ensure the tweet is within {280} characters.", target_module=GenerateTweet)
dspy.Suggest(has_correct_answer(generated_tweet, answer), "The tweet does not include the correct answer to the question. Please revise accordingly.", target_module=GenerateTweet)
engaging_question = "Does the assessed text make for a self-contained, engaging tweet? Say no if it is not engaging."
engaging_assessment = dspy.Predict(AssessTweet)(context=context, assessed_text=generated_tweet, assessment_question=engaging_question)
dspy.Suggest(is_assessment_yes(engaging_assessment.assessment_answer), "The text is not engaging enough. Please revise to make it more captivating.", target_module=GenerateTweet)
faithful_question = "Is the assessed text grounded in the context? Say no if it includes significant facts not in the context."
faithful_assessment = dspy.Predict(AssessTweet)(context='N/A', assessed_text=generated_tweet, assessment_question=faithful_question)
dspy.Suggest(is_assessment_yes(faithful_assessment.assessment_answer), "The text contains unfaithful elements or significant facts not in the context. Please revise for accuracy.", target_module=GenerateTweet)
return dspy.Prediction(generated_tweet=generated_tweet, context=context)
tweeter_with_assertions = assert_transform_module(TweeterWithAssertions().map_named_predictors(Retry), backtrack_handler)
metrics = [no_hashtags_metric, is_correct_metric, within_length_metric, engaging_metric, faithful_metric, overall_metric]
for metric in metrics:
evaluate = Evaluate(metric=metric, devset=devset, num_threads=1, display_progress=True, display_table=5)
evaluate(tweeter_with_assertions)
example = devset[10]
tweet = tweeter_with_assertions(question=example.question, answer = example.answer)
print(f'Generated Tweet: ', tweet.generated_tweet)
tweet.context
for metric in metrics:
evaluate = Evaluate(metric=metric, devset=devset[10:11], num_threads=1, display_progress=True, display_table=5)
evaluate(tweeter_with_assertions)
teleprompter = BootstrapFewShotWithRandomSearch(metric = overall_metric, max_bootstrapped_demos=2, num_candidate_programs=6)
compiled_tweeter = teleprompter.compile(student = tweeter, teacher = tweeter, trainset=trainset, valset=devset[:100])
for metric in metrics:
evaluate = Evaluate(metric=metric, devset=devset, num_threads=1, display_progress=True, display_table=5)
evaluate(compiled_tweeter)
teleprompter = BootstrapFewShotWithRandomSearch(metric = overall_metric, max_bootstrapped_demos=2, num_candidate_programs=6)
compiled_with_assertions_tweeter = teleprompter.compile(student=tweeter, teacher = tweeter_with_assertions, trainset=trainset, valset=devset[:100])
for metric in metrics:
evaluate = Evaluate(metric=metric, devset=devset, num_threads=1, display_progress=True, display_table=5)
evaluate(compiled_with_assertions_tweeter)
teleprompter = BootstrapFewShotWithRandomSearch(metric = overall_metric, max_bootstrapped_demos=2, num_candidate_programs=6, num_threads=1)
compiled_tweeter_with_assertions = teleprompter.compile(student=tweeter_with_assertions, teacher = tweeter_with_assertions, trainset=trainset, valset=devset[:100])
for metric in metrics:
evaluate = | Evaluate(metric=metric, devset=devset, num_threads=1, display_progress=True, display_table=5) | dspy.evaluate.evaluate.Evaluate |
get_ipython().system('git clone https://huggingface.co/arnavs11/DSPy_TweetGen_Cache')
get_ipython().run_line_magic('cd', 'DSPy_TweetGen_Cache/')
get_ipython().system('git checkout master')
get_ipython().run_line_magic('cd', '..')
import os
repo_clone_path = '/content/DSPy_TweetGen_Cache'
if not os.access('/content', os.W_OK):
repo_clone_path = os.path.join(os.getcwd(), 'DSPy_TweetGen_Cache')
os.environ["DSP_NOTEBOOK_CACHEDIR"] = repo_clone_path
get_ipython().run_line_magic('load_ext', 'autoreload')
get_ipython().run_line_magic('autoreload', '2')
import sys
import os
import regex as re
import json
try: # When on google Colab, let's clone the notebook so we download the cache.
import google.colab
repo_path = 'dspy'
get_ipython().system('git -C $repo_path pull origin || git clone https://github.com/stanfordnlp/dspy $repo_path')
except:
repo_path = '.'
if repo_path not in sys.path:
sys.path.append(repo_path)
import pkg_resources # Install the package if it's not installed
if not "dspy-ai" in {pkg.key for pkg in pkg_resources.working_set}:
get_ipython().system('pip install -U pip')
get_ipython().system('pip install dspy-ai')
get_ipython().system('pip install openai~=0.28.1')
get_ipython().system('pip install -e $repo_path')
import dspy
from dspy.predict import Retry
from dspy.datasets import HotPotQA
from dspy.teleprompt import BootstrapFewShotWithRandomSearch
from dsp.utils import deduplicate
from dspy.evaluate.evaluate import Evaluate
from dspy.primitives.assertions import assert_transform_module, backtrack_handler
colbertv2_wiki17_abstracts = dspy.ColBERTv2(url='http://20.102.90.50:2017/wiki17_abstracts')
dspy.settings.configure(rm=colbertv2_wiki17_abstracts)
turbo = dspy.OpenAI(model='gpt-3.5-turbo', max_tokens=500)
dspy.settings.configure(lm=turbo, trace=[], temperature=0.7)
dataset = | HotPotQA(train_seed=1, train_size=300, eval_seed=2023, dev_size=300, test_size=0, keep_details=True) | dspy.datasets.HotPotQA |
get_ipython().run_line_magic('load_ext', 'autoreload')
get_ipython().run_line_magic('autoreload', '2')
import sys
import os
try: # When on google Colab, let's clone the notebook so we download the cache.
import google.colab
repo_path = 'dspy'
get_ipython().system('git -C $repo_path pull origin || git clone https://github.com/stanfordnlp/dspy $repo_path')
except:
repo_path = '.'
if repo_path not in sys.path:
sys.path.append(repo_path)
os.environ["DSP_NOTEBOOK_CACHEDIR"] = os.path.join(repo_path, 'cache')
import dspy
get_ipython().run_line_magic('load_ext', 'autoreload')
get_ipython().run_line_magic('autoreload', '2')
import sys; sys.path.append('/future/u/okhattab/repos/public/stanfordnlp/dspy')
import dspy
from dspy.evaluate import Evaluate
from dspy.datasets.hotpotqa import HotPotQA
from dspy.teleprompt import BootstrapFewShot, BootstrapFewShotWithRandomSearch, BootstrapFinetune
llama = dspy.HFClientTGI(model="meta-llama/Llama-2-13b-chat-hf", port=[7140, 7141, 7142, 7143], max_tokens=150)
colbertv2 = dspy.ColBERTv2(url='http://20.102.90.50:2017/wiki17_abstracts')
dspy.settings.configure(rm=colbertv2, lm=llama)
train = [('Who was the director of the 2009 movie featuring Peter Outerbridge as William Easton?', 'Kevin Greutert'),
('The heir to the Du Pont family fortune sponsored what wrestling team?', 'Foxcatcher'),
('In what year was the star of To Hell and Back born?', '1925'),
('Which award did the first book of Gary Zukav receive?', 'U.S. National Book Award'),
('What documentary about the Gilgo Beach Killer debuted on A&E?', 'The Killing Season'),
('Which author is English: John Braine or Studs Terkel?', 'John Braine'),
('Who produced the album that included a re-recording of "Lithium"?', 'Butch Vig')]
train = [ | dspy.Example(question=question, answer=answer) | dspy.Example |
import dspy
from dsp.utils import deduplicate
from dspy.datasets import HotPotQA
from dspy.predict.retry import Retry
from dspy.teleprompt import BootstrapFewShot, BootstrapFewShotWithRandomSearch
from dspy.evaluate.evaluate import Evaluate
from dspy.primitives.assertions import assert_transform_module, backtrack_handler
import os
import openai
openai.api_key = os.getenv('OPENAI_API_KEY')
colbertv2_wiki17_abstracts = dspy.ColBERTv2(url='http://20.102.90.50:2017/wiki17_abstracts')
dspy.settings.configure(rm=colbertv2_wiki17_abstracts)
turbo = dspy.OpenAI(model='gpt-3.5-turbo', max_tokens=500)
dspy.settings.configure(lm=turbo, trace=[], temperature=0.7)
dataset = HotPotQA(train_seed=1, train_size=300, eval_seed=2023, dev_size=300, test_size=0)
trainset = [x.with_inputs('question') for x in dataset.train]
devset = [x.with_inputs('question') for x in dataset.dev]
def validate_query_distinction_local(previous_queries, query):
"""check if query is distinct from previous queries"""
if previous_queries == []:
return True
if dspy.evaluate.answer_exact_match_str(query, previous_queries, frac=0.8):
return False
return True
def validate_context_and_answer_and_hops(example, pred, trace=None):
if not dspy.evaluate.answer_exact_match(example, pred):
return False
if not dspy.evaluate.answer_passage_match(example, pred):
return False
return True
def gold_passages_retrieved(example, pred, trace=None):
gold_titles = set(map(dspy.evaluate.normalize_text, example['gold_titles']))
found_titles = set(map(dspy.evaluate.normalize_text, [c.split(' | ')[0] for c in pred.context]))
return gold_titles.issubset(found_titles)
class GenerateAnswer(dspy.Signature):
"""Answer questions with short factoid answers."""
context = dspy.InputField(desc="may contain relevant facts")
question = dspy.InputField()
answer = dspy.OutputField(desc="often between 1 and 5 words")
class GenerateSearchQuery(dspy.Signature):
"""Write a simple search query that will help answer a complex question."""
context = dspy.InputField(desc="may contain relevant facts")
question = dspy.InputField()
query = | dspy.OutputField() | dspy.OutputField |
get_ipython().system('git clone https://huggingface.co/arnavs11/DSPy_LongFormQA_Cache')
get_ipython().run_line_magic('cd', 'DSPy_LongFormQA_Cache/')
get_ipython().system('git checkout master')
get_ipython().run_line_magic('cd', '..')
import os
repo_clone_path = '/content/DSPy_LongFormQA_Cache'
if not os.access('/content', os.W_OK):
repo_clone_path = os.path.join(os.getcwd(), 'DSPy_LongFormQA_Cache')
os.environ["DSP_NOTEBOOK_CACHEDIR"] = repo_clone_path
get_ipython().run_line_magic('load_ext', 'autoreload')
get_ipython().run_line_magic('autoreload', '2')
import sys
import os
import regex as re
try: # When on google Colab, let's clone the notebook so we download the cache.
import google.colab
repo_path = 'dspy'
get_ipython().system('git -C $repo_path pull origin || git clone https://github.com/stanfordnlp/dspy $repo_path')
except:
repo_path = '.'
if repo_path not in sys.path:
sys.path.append(repo_path)
import pkg_resources # Install the package if it's not installed
if not "dspy-ai" in {pkg.key for pkg in pkg_resources.working_set}:
get_ipython().system('pip install -U pip')
get_ipython().system('pip install dspy-ai')
get_ipython().system('pip install openai~=0.28.1')
get_ipython().system('pip install -e $repo_path')
import dspy
from dspy.predict import Retry
from dspy.datasets import HotPotQA
from dspy.teleprompt import BootstrapFewShotWithRandomSearch
from dsp.utils import EM, normalize_text
from dspy.primitives.assertions import assert_transform_module, backtrack_handler
get_ipython().run_line_magic('cd', 'dspy/examples/longformqa')
from utils import extract_text_by_citation, correct_citation_format, has_citations, citations_check
colbertv2_wiki17_abstracts = dspy.ColBERTv2(url='http://20.102.90.50:2017/wiki17_abstracts')
dspy.settings.configure(rm=colbertv2_wiki17_abstracts)
turbo = dspy.OpenAI(model='gpt-3.5-turbo', max_tokens=500)
dspy.settings.configure(lm=turbo, trace=[], temperature=0.7)
dataset = HotPotQA(train_seed=1, train_size=300, eval_seed=2023, dev_size=300, test_size=0, keep_details=True)
trainset = [x.with_inputs('question') for x in dataset.train]
devset = [x.with_inputs('question') for x in dataset.dev]
train_example = trainset[0]
print(f"Question: {train_example.question}")
print(f"Answer: {train_example.answer}")
print(f"Relevant Wikipedia Titles: {train_example.gold_titles}")
dev_example = devset[18]
print(f"Question: {dev_example.question}")
print(f"Answer: {dev_example.answer}")
print(f"Relevant Wikipedia Titles: {dev_example.gold_titles}")
from dsp.utils import deduplicate
class GenerateSearchQuery(dspy.Signature):
"""Write a simple search query that will help answer a complex question."""
context = dspy.InputField(desc="may contain relevant facts")
question = dspy.InputField()
query = dspy.OutputField()
class GenerateCitedParagraph(dspy.Signature):
"""Generate a paragraph with citations."""
context = dspy.InputField(desc="may contain relevant facts")
question = dspy.InputField()
paragraph = dspy.OutputField(desc="includes citations")
class LongFormQA(dspy.Module):
def __init__(self, passages_per_hop=3, max_hops=2):
super().__init__()
self.generate_query = [dspy.ChainOfThought(GenerateSearchQuery) for _ in range(max_hops)]
self.retrieve = dspy.Retrieve(k=passages_per_hop)
self.generate_cited_paragraph = dspy.ChainOfThought(GenerateCitedParagraph)
self.max_hops = max_hops
def forward(self, question):
context = []
for hop in range(self.max_hops):
query = self.generate_query[hop](context=context, question=question).query
passages = self.retrieve(query).passages
context = deduplicate(context + passages)
pred = self.generate_cited_paragraph(context=context, question=question)
pred = dspy.Prediction(context=context, paragraph=pred.paragraph)
return pred
class CheckCitationFaithfulness(dspy.Signature):
"""Verify that the text is based on the provided context."""
context = dspy.InputField(desc="may contain relevant facts")
text = dspy.InputField(desc="between 1 to 2 sentences")
faithfulness = dspy.OutputField(desc="boolean indicating if text is faithful to context")
def citation_faithfulness(example, pred, trace):
paragraph, context = pred.paragraph, pred.context
citation_dict = extract_text_by_citation(paragraph)
if not citation_dict:
return False, None
context_dict = {str(i): context[i].split(' | ')[1] for i in range(len(context))}
faithfulness_results = []
unfaithful_citations = []
check_citation_faithfulness = dspy.ChainOfThought(CheckCitationFaithfulness)
for citation_num, texts in citation_dict.items():
if citation_num not in context_dict:
continue
current_context = context_dict[citation_num]
for text in texts:
try:
result = check_citation_faithfulness(context=current_context, text=text)
is_faithful = result.faithfulness.lower() == 'true'
faithfulness_results.append(is_faithful)
if not is_faithful:
unfaithful_citations.append({'paragraph': paragraph, 'text': text, 'context': current_context})
except ValueError as e:
faithfulness_results.append(False)
unfaithful_citations.append({'paragraph': paragraph, 'text': text, 'error': str(e)})
final_faithfulness = all(faithfulness_results)
if not faithfulness_results:
return False, None
return final_faithfulness, unfaithful_citations
def extract_cited_titles_from_paragraph(paragraph, context):
cited_indices = [int(m.group(1)) for m in re.finditer(r'\[(\d+)\]\.', paragraph)]
cited_indices = [index - 1 for index in cited_indices if index <= len(context)]
cited_titles = [context[index].split(' | ')[0] for index in cited_indices]
return cited_titles
def calculate_recall(example, pred, trace=None):
gold_titles = set(example['gold_titles'])
found_cited_titles = set(extract_cited_titles_from_paragraph(pred.paragraph, pred.context))
intersection = gold_titles.intersection(found_cited_titles)
recall = len(intersection) / len(gold_titles) if gold_titles else 0
return recall
def calculate_precision(example, pred, trace=None):
gold_titles = set(example['gold_titles'])
found_cited_titles = set(extract_cited_titles_from_paragraph(pred.paragraph, pred.context))
intersection = gold_titles.intersection(found_cited_titles)
precision = len(intersection) / len(found_cited_titles) if found_cited_titles else 0
return precision
def answer_correctness(example, pred, trace=None):
assert hasattr(example, 'answer'), "Example does not have 'answer'."
normalized_context = normalize_text(pred.paragraph)
if isinstance(example.answer, str):
gold_answers = [example.answer]
elif isinstance(example.answer, list):
gold_answers = example.answer
else:
raise ValueError("'example.answer' is not string or list.")
return 1 if any(normalize_text(answer) in normalized_context for answer in gold_answers) else 0
def evaluate(module):
correctness_values = []
recall_values = []
precision_values = []
citation_faithfulness_values = []
for i in range(len(devset)):
example = devset[i]
try:
pred = module(question=example.question)
correctness_values.append(answer_correctness(example, pred))
citation_faithfulness_score, _ = citation_faithfulness(None, pred, None)
citation_faithfulness_values.append(citation_faithfulness_score)
recall = calculate_recall(example, pred)
precision = calculate_precision(example, pred)
recall_values.append(recall)
precision_values.append(precision)
except Exception as e:
print(f"Failed generation with error: {e}")
average_correctness = sum(correctness_values) / len(devset) if correctness_values else 0
average_recall = sum(recall_values) / len(devset) if recall_values else 0
average_precision = sum(precision_values) / len(devset) if precision_values else 0
average_citation_faithfulness = sum(citation_faithfulness_values) / len(devset) if citation_faithfulness_values else 0
print(f"Average Correctness: {average_correctness}")
print(f"Average Recall: {average_recall}")
print(f"Average Precision: {average_precision}")
print(f"Average Citation Faithfulness: {average_citation_faithfulness}")
longformqa = LongFormQA()
evaluate(longformqa)
question = devset[6].question
pred = longformqa(question)
citation_faithfulness_score, _ = citation_faithfulness(None, pred, None)
print(f"Question: {question}")
print(f"Predicted Paragraph: {pred.paragraph}")
print(f"Citation Faithfulness: {citation_faithfulness_score}")
class LongFormQAWithAssertions(dspy.Module):
def __init__(self, passages_per_hop=3, max_hops=2):
super().__init__()
self.generate_query = [dspy.ChainOfThought(GenerateSearchQuery) for _ in range(max_hops)]
self.retrieve = dspy.Retrieve(k=passages_per_hop)
self.generate_cited_paragraph = dspy.ChainOfThought(GenerateCitedParagraph)
self.max_hops = max_hops
def forward(self, question):
context = []
for hop in range(self.max_hops):
query = self.generate_query[hop](context=context, question=question).query
passages = self.retrieve(query).passages
context = deduplicate(context + passages)
pred = self.generate_cited_paragraph(context=context, question=question)
pred = dspy.Prediction(context=context, paragraph=pred.paragraph)
dspy.Suggest(citations_check(pred.paragraph), f"Make sure every 1-2 sentences has citations. If any 1-2 sentences lack citations, add them in 'text... [x].' format.", target_module=GenerateCitedParagraph)
_, unfaithful_outputs = citation_faithfulness(None, pred, None)
if unfaithful_outputs:
unfaithful_pairs = [(output['text'], output['context']) for output in unfaithful_outputs]
for _, context in unfaithful_pairs:
dspy.Suggest(len(unfaithful_pairs) == 0, f"Make sure your output is based on the following context: '{context}'.", target_module=GenerateCitedParagraph)
else:
return pred
return pred
longformqa_with_assertions = assert_transform_module(LongFormQAWithAssertions().map_named_predictors(Retry), backtrack_handler)
evaluate(longformqa_with_assertions)
question = devset[6].question
pred = longformqa_with_assertions(question)
citation_faithfulness_score, _ = citation_faithfulness(None, pred, None)
print(f"Question: {question}")
print(f"Predicted Paragraph: {pred.paragraph}")
print(f"Citation Faithfulness: {citation_faithfulness_score}")
longformqa = LongFormQA()
teleprompter = | BootstrapFewShotWithRandomSearch(metric = answer_correctness, max_bootstrapped_demos=2, num_candidate_programs=6) | dspy.teleprompt.BootstrapFewShotWithRandomSearch |
get_ipython().run_line_magic('load_ext', 'autoreload')
get_ipython().run_line_magic('autoreload', '2')
import sys
import os
try: # When on google Colab, let's clone the notebook so we download the cache.
import google.colab
repo_path = 'dspy'
get_ipython().system('git -C $repo_path pull origin || git clone https://github.com/stanfordnlp/dspy $repo_path')
except:
repo_path = '.'
if repo_path not in sys.path:
sys.path.append(repo_path)
os.environ["DSP_NOTEBOOK_CACHEDIR"] = os.path.join(repo_path, 'cache')
import pkg_resources # Install the package if it's not installed
if not "dspy-ai" in {pkg.key for pkg in pkg_resources.working_set}:
get_ipython().system('pip install -U pip')
get_ipython().system('pip install -e $repo_path')
get_ipython().system('pip install transformers')
import dspy
from dspy.evaluate import Evaluate
from dspy.teleprompt import BootstrapFewShot, BootstrapFewShotWithRandomSearch, BootstrapFinetune
llama = dspy.HFClientTGI(model="meta-llama/Llama-2-13b-chat-hf", port=[7140, 7141, 7142, 7143], max_tokens=150)
colbertv2 = dspy.ColBERTv2(url='http://20.102.90.50:2017/wiki17_abstracts')
dspy.settings.configure(rm=colbertv2, lm=llama)
train = [('Who was the director of the 2009 movie featuring Peter Outerbridge as William Easton?', 'Kevin Greutert'),
('The heir to the Du Pont family fortune sponsored what wrestling team?', 'Foxcatcher'),
('In what year was the star of To Hell and Back born?', '1925'),
('Which award did the first book of Gary Zukav receive?', 'U.S. National Book Award'),
('What documentary about the Gilgo Beach Killer debuted on A&E?', 'The Killing Season'),
('Which author is English: John Braine or Studs Terkel?', 'John Braine'),
('Who produced the album that included a re-recording of "Lithium"?', 'Butch Vig')]
train = [dspy.Example(question=question, answer=answer).with_inputs('question') for question, answer in train]
dev = [('Who has a broader scope of profession: E. L. Doctorow or Julia Peterkin?', 'E. L. Doctorow'),
('Right Back At It Again contains lyrics co-written by the singer born in what city?', 'Gainesville, Florida'),
('What year was the party of the winner of the 1971 San Francisco mayoral election founded?', '1828'),
('Anthony Dirrell is the brother of which super middleweight title holder?', 'Andre Dirrell'),
('The sports nutrition business established by Oliver Cookson is based in which county in the UK?', 'Cheshire'),
('Find the birth date of the actor who played roles in First Wives Club and Searching for the Elephant.', 'February 13, 1980'),
('Kyle Moran was born in the town on what river?', 'Castletown River'),
("The actress who played the niece in the Priest film was born in what city, country?", 'Surrey, England'),
('Name the movie in which the daughter of Noel Harrison plays Violet Trefusis.', 'Portrait of a Marriage'),
('What year was the father of the Princes in the Tower born?', '1442'),
('What river is near the Crichton Collegiate Church?', 'the River Tyne'),
('Who purchased the team Michael Schumacher raced for in the 1995 Monaco Grand Prix in 2000?', 'Renault'),
('André Zucca was a French photographer who worked with a German propaganda magazine published by what Nazi organization?', 'the Wehrmacht')]
dev = [dspy.Example(question=question, answer=answer).with_inputs('question') for question, answer in dev]
predict = dspy.Predict('question -> answer')
predict(question="What is the capital of Germany?")
class CoT(dspy.Module): # let's define a new module
def __init__(self):
super().__init__()
self.generate_answer = dspy.ChainOfThought('question -> answer')
def forward(self, question):
return self.generate_answer(question=question) # here we use the module
metric_EM = dspy.evaluate.answer_exact_match
teleprompter = BootstrapFewShot(metric=metric_EM, max_bootstrapped_demos=2)
cot_compiled = teleprompter.compile(CoT(), trainset=train)
cot_compiled("What is the capital of Germany?")
llama.inspect_history(n=1)
NUM_THREADS = 32
evaluate_hotpot = Evaluate(devset=dev, metric=metric_EM, num_threads=NUM_THREADS, display_progress=True, display_table=15)
evaluate_hotpot(cot_compiled)
class RAG(dspy.Module):
def __init__(self, num_passages=3):
super().__init__()
self.retrieve = dspy.Retrieve(k=num_passages)
self.generate_query = dspy.ChainOfThought("question -> search_query")
self.generate_answer = | dspy.ChainOfThought("context, question -> answer") | dspy.ChainOfThought |
get_ipython().run_line_magic('load_ext', 'autoreload')
get_ipython().run_line_magic('autoreload', '2')
import dspy
from dspy.evaluate import Evaluate
from dspy.datasets.hotpotqa import HotPotQA
from dspy.teleprompt import BootstrapFewShotWithRandomSearch, BootstrapFinetune
ports = [7140, 7141, 7142, 7143, 7144, 7145]
llamaChat = dspy.HFClientTGI(model="meta-llama/Llama-2-13b-chat-hf", port=ports, max_tokens=150)
colbertv2 = dspy.ColBERTv2(url='http://20.102.90.50:2017/wiki17_abstracts')
dspy.settings.configure(rm=colbertv2, lm=llamaChat)
dataset = HotPotQA(train_seed=1, train_size=200, eval_seed=2023, dev_size=1000, test_size=0)
trainset = [x.with_inputs('question') for x in dataset.train]
devset = [x.with_inputs('question') for x in dataset.dev]
testset = [x.with_inputs('question') for x in dataset.test]
len(trainset), len(devset), len(testset)
trainset[0]
from dsp.utils.utils import deduplicate
class BasicMH(dspy.Module):
def __init__(self, passages_per_hop=3):
super().__init__()
self.retrieve = dspy.Retrieve(k=passages_per_hop)
self.generate_query = [dspy.ChainOfThought("context, question -> search_query") for _ in range(2)]
self.generate_answer = dspy.ChainOfThought("context, question -> answer")
def forward(self, question):
context = []
for hop in range(2):
search_query = self.generate_query[hop](context=context, question=question).search_query
passages = self.retrieve(search_query).passages
context = deduplicate(context + passages)
return self.generate_answer(context=context, question=question).copy(context=context)
RECOMPILE_INTO_LLAMA_FROM_SCRATCH = False
NUM_THREADS = 24
metric_EM = dspy.evaluate.answer_exact_match
if RECOMPILE_INTO_LLAMA_FROM_SCRATCH:
tp = BootstrapFewShotWithRandomSearch(metric=metric_EM, max_bootstrapped_demos=2, num_threads=NUM_THREADS)
basicmh_bs = tp.compile(BasicMH(), trainset=trainset[:50], valset=trainset[50:200])
ensemble = [prog for *_, prog in basicmh_bs.candidate_programs[:4]]
for idx, prog in enumerate(ensemble):
pass
if not RECOMPILE_INTO_LLAMA_FROM_SCRATCH:
ensemble = []
for idx in range(4):
prog = BasicMH()
prog.load(f'multihop_llama213b_{idx}.json')
ensemble.append(prog)
llama_program = ensemble[0]
evaluate_hotpot = Evaluate(devset=devset[:1000], metric=metric_EM, num_threads=NUM_THREADS, display_progress=True, display_table=0)
evaluate_hotpot(llama_program)
llama_program(question="How many storeys are in the castle that David Gregory inherited?")
llamaChat.inspect_history(n=3)
unlabeled_train = HotPotQA(train_seed=1, train_size=3000, eval_seed=2023, dev_size=0, test_size=0).train
unlabeled_train = [dspy.Example(question=x.question).with_inputs('question') for x in unlabeled_train]
len(unlabeled_train)
always_true = lambda g, p, trace=None: True
for prog_ in ensemble:
evaluate_hotpot(prog_, devset=unlabeled_train[:3000], metric=always_true)
RECOMPILE_INTO_T5_FROM_SCRATCH = False
if RECOMPILE_INTO_T5_FROM_SCRATCH:
config = dict(target='t5-large', epochs=2, bf16=True, bsize=6, accumsteps=2, lr=5e-5)
tp = BootstrapFinetune(metric=None)
t5_program = tp.compile(BasicMH(), teacher=ensemble, trainset=unlabeled_train[:3000], **config)
for p in t5_program.predictors(): p.activated = False
if not RECOMPILE_INTO_T5_FROM_SCRATCH:
t5_program = BasicMH()
ckpt_path = "colbert-ir/dspy-Oct11-T5-Large-MH-3k-v1"
LM = | dspy.HFModel(checkpoint=ckpt_path, model='t5-large') | dspy.HFModel |
get_ipython().system('git clone https://huggingface.co/arnavs11/DSPy_TweetGen_Cache')
get_ipython().run_line_magic('cd', 'DSPy_TweetGen_Cache/')
get_ipython().system('git checkout master')
get_ipython().run_line_magic('cd', '..')
import os
repo_clone_path = '/content/DSPy_TweetGen_Cache'
if not os.access('/content', os.W_OK):
repo_clone_path = os.path.join(os.getcwd(), 'DSPy_TweetGen_Cache')
os.environ["DSP_NOTEBOOK_CACHEDIR"] = repo_clone_path
get_ipython().run_line_magic('load_ext', 'autoreload')
get_ipython().run_line_magic('autoreload', '2')
import sys
import os
import regex as re
import json
try: # When on google Colab, let's clone the notebook so we download the cache.
import google.colab
repo_path = 'dspy'
get_ipython().system('git -C $repo_path pull origin || git clone https://github.com/stanfordnlp/dspy $repo_path')
except:
repo_path = '.'
if repo_path not in sys.path:
sys.path.append(repo_path)
import pkg_resources # Install the package if it's not installed
if not "dspy-ai" in {pkg.key for pkg in pkg_resources.working_set}:
get_ipython().system('pip install -U pip')
get_ipython().system('pip install dspy-ai')
get_ipython().system('pip install openai~=0.28.1')
get_ipython().system('pip install -e $repo_path')
import dspy
from dspy.predict import Retry
from dspy.datasets import HotPotQA
from dspy.teleprompt import BootstrapFewShotWithRandomSearch
from dsp.utils import deduplicate
from dspy.evaluate.evaluate import Evaluate
from dspy.primitives.assertions import assert_transform_module, backtrack_handler
colbertv2_wiki17_abstracts = dspy.ColBERTv2(url='http://20.102.90.50:2017/wiki17_abstracts')
dspy.settings.configure(rm=colbertv2_wiki17_abstracts)
turbo = dspy.OpenAI(model='gpt-3.5-turbo', max_tokens=500)
dspy.settings.configure(lm=turbo, trace=[], temperature=0.7)
dataset = HotPotQA(train_seed=1, train_size=300, eval_seed=2023, dev_size=300, test_size=0, keep_details=True)
trainset = [x.with_inputs('question', 'answer') for x in dataset.train]
devset = [x.with_inputs('question', 'answer') for x in dataset.dev]
class GenerateSearchQuery(dspy.Signature):
"""Write a simple search query that will help answer a complex question."""
context = dspy.InputField(desc="may contain relevant facts")
question = dspy.InputField()
query = dspy.OutputField()
class GenerateTweet(dspy.Signature):
"""Generate an engaging tweet that effectively answers a question staying faithful to the context, is less than 280 characters, and has no hashtags."""
question = dspy.InputField()
context = dspy.InputField(desc="may contain relevant facts")
tweet = dspy.OutputField()
class Tweeter(dspy.Module):
def __init__(self):
super().__init__()
self.generate_tweet = dspy.ChainOfThought(GenerateTweet)
def forward(self, question, answer):
context = []
max_hops=2
passages_per_hop=3
generate_query = [dspy.ChainOfThought(GenerateSearchQuery) for _ in range(max_hops)]
retrieve = dspy.Retrieve(k=passages_per_hop)
for hop in range(max_hops):
query = generate_query[hop](context=context, question=question).query
passages = retrieve(query).passages
context = deduplicate(context + passages)
generated_tweet = self.generate_tweet(question=question, context=context).tweet
return dspy.Prediction(generated_tweet=generated_tweet, context=context)
tweeter = Tweeter()
def has_no_hashtags(text):
return len(re.findall(r"#\w+", text)) == 0
def is_within_length_limit(text, length_limit=280):
return len(text) <= length_limit
def is_assessment_yes(assessment_answer):
"""Check if the first word of the assessment answer is 'yes'."""
return assessment_answer.split()[0].lower() == 'yes'
def has_correct_answer(text, answer):
return answer in text
class AssessTweet(dspy.Signature):
"""Assess the quality of a tweet along the specified dimension."""
context = dspy.InputField(desc='ignore if N/A')
assessed_text = dspy.InputField()
assessment_question = dspy.InputField()
assessment_answer = dspy.OutputField(desc="Yes or No")
def no_hashtags_metric(gold, pred, trace=None):
tweet = pred.generated_tweet
no_hashtags = has_no_hashtags(tweet)
score = no_hashtags
return score
def is_correct_metric(gold, pred, trace=None):
answer, tweet = gold.answer, pred.generated_tweet
correct = has_correct_answer(tweet, answer)
score = correct
return score
def within_length_metric(gold, pred, trace=None):
tweet = pred.generated_tweet
within_length_limit = is_within_length_limit(tweet, 280)
score = within_length_limit
return score
def engaging_metric(gold, pred, trace=None):
tweet = pred.generated_tweet
engaging = "Does the assessed text make for a self-contained, engaging tweet? Say no if it is not engaging."
engaging = dspy.Predict(AssessTweet)(context='N/A', assessed_text=tweet, assessment_question=engaging)
engaging = engaging.assessment_answer.split()[0].lower() == 'yes'
score = engaging
return score
def faithful_metric(gold, pred, trace=None):
context, tweet = pred.context, pred.generated_tweet
faithful = "Is the assessed text grounded in the context? Say no if it includes significant facts not in the context."
faithful = dspy.Predict(AssessTweet)(context=context, assessed_text=tweet, assessment_question=faithful)
faithful = faithful.assessment_answer.split()[0].lower() == 'yes'
score = faithful
return score
def overall_metric(gold, pred, trace=None):
answer, context, tweet = gold.answer, pred.context, pred.generated_tweet
no_hashtags = has_no_hashtags(tweet)
within_length_limit = is_within_length_limit(tweet, 280)
correct = has_correct_answer(tweet, answer)
engaging = "Does the assessed text make for a self-contained, engaging tweet? Say no if it is not engaging."
faithful = "Is the assessed text grounded in the context? Say no if it includes significant facts not in the context."
faithful = dspy.Predict(AssessTweet)(context=context, assessed_text=tweet, assessment_question=faithful)
engaging = dspy.Predict(AssessTweet)(context='N/A', assessed_text=tweet, assessment_question=engaging)
engaging, faithful = [m.assessment_answer.split()[0].lower() == 'yes' for m in [engaging, faithful]]
score = (correct + engaging + faithful + no_hashtags + within_length_limit) if correct and within_length_limit else 0
return score / 5.0
metrics = [no_hashtags_metric, is_correct_metric, within_length_metric, engaging_metric, faithful_metric, overall_metric]
for metric in metrics:
evaluate = Evaluate(metric=metric, devset=devset, num_threads=1, display_progress=True, display_table=5)
evaluate(tweeter)
example = devset[10]
tweet = tweeter(question=example.question, answer = example.answer)
print(f'Generated Tweet: ', tweet.generated_tweet)
tweet.context
for metric in metrics:
evaluate = Evaluate(metric=metric, devset=devset[10:11], num_threads=1, display_progress=True, display_table=5)
evaluate(tweeter)
class TweeterWithAssertions(dspy.Module):
def __init__(self):
super().__init__()
self.generate_tweet = dspy.ChainOfThought(GenerateTweet)
def forward(self, question, answer):
context = []
max_hops=2
passages_per_hop=3
generate_query = [dspy.ChainOfThought(GenerateSearchQuery) for _ in range(max_hops)]
retrieve = dspy.Retrieve(k=passages_per_hop)
for hop in range(max_hops):
query = generate_query[hop](context=context, question=question).query
passages = retrieve(query).passages
context = deduplicate(context + passages)
generated_tweet = self.generate_tweet(question=question, context=context).tweet
dspy.Suggest(has_no_hashtags(generated_tweet), f"Please revise the tweet to remove hashtag phrases following it.", target_module=GenerateTweet)
dspy.Suggest(is_within_length_limit(generated_tweet, 280), f"Please ensure the tweet is within {280} characters.", target_module=GenerateTweet)
dspy.Suggest(has_correct_answer(generated_tweet, answer), "The tweet does not include the correct answer to the question. Please revise accordingly.", target_module=GenerateTweet)
engaging_question = "Does the assessed text make for a self-contained, engaging tweet? Say no if it is not engaging."
engaging_assessment = dspy.Predict(AssessTweet)(context=context, assessed_text=generated_tweet, assessment_question=engaging_question)
dspy.Suggest(is_assessment_yes(engaging_assessment.assessment_answer), "The text is not engaging enough. Please revise to make it more captivating.", target_module=GenerateTweet)
faithful_question = "Is the assessed text grounded in the context? Say no if it includes significant facts not in the context."
faithful_assessment = dspy.Predict(AssessTweet)(context='N/A', assessed_text=generated_tweet, assessment_question=faithful_question)
dspy.Suggest(is_assessment_yes(faithful_assessment.assessment_answer), "The text contains unfaithful elements or significant facts not in the context. Please revise for accuracy.", target_module=GenerateTweet)
return | dspy.Prediction(generated_tweet=generated_tweet, context=context) | dspy.Prediction |
get_ipython().run_line_magic('load_ext', 'autoreload')
get_ipython().run_line_magic('autoreload', '2')
import sys
import os
try: # When on google Colab, let's clone the notebook so we download the cache.
import google.colab
repo_path = 'dspy'
get_ipython().system('git -C $repo_path pull origin || git clone https://github.com/stanfordnlp/dspy $repo_path')
except:
repo_path = '.'
if repo_path not in sys.path:
sys.path.append(repo_path)
os.environ["DSP_NOTEBOOK_CACHEDIR"] = os.path.join(repo_path, 'cache')
import pkg_resources # Install the package if it's not installed
if not "dspy-ai" in {pkg.key for pkg in pkg_resources.working_set}:
get_ipython().system('pip install -U pip')
get_ipython().system('pip install -e $repo_path')
get_ipython().system('pip install transformers')
import dspy
from dspy.evaluate import Evaluate
from dspy.teleprompt import BootstrapFewShot, BootstrapFewShotWithRandomSearch, BootstrapFinetune
llama = dspy.HFClientTGI(model="meta-llama/Llama-2-13b-chat-hf", port=[7140, 7141, 7142, 7143], max_tokens=150)
colbertv2 = dspy.ColBERTv2(url='http://20.102.90.50:2017/wiki17_abstracts')
| dspy.settings.configure(rm=colbertv2, lm=llama) | dspy.settings.configure |
End of preview. Expand
in Dataset Viewer.
README.md exists but content is empty.
Use the Edit dataset card button to edit it.
- Downloads last month
- 44