counsellor_v0.5

.gitignore

@@ -0,0 +1 @@
+*.yml

Dockerfile

@@ -0,0 +1,6 @@
+FROM python:3.10
+WORKDIR /code/counsellor_chat_api
+COPY ./requirements.txt ./
+RUN pip install --no-cache-dir -r ./requirements.txt
+COPY . .
+CMD ["uvicorn", "main:app", "--host", "0.0.0.0", "--port", "7860"]

cookbook/llama_guard.py

@@ -0,0 +1,58 @@
+from llama_cpp import Llama
+
+guard = Llama.from_pretrained(repo_id = "QuantFactory/Meta-Llama-Guard-2-8B-GGUF",
+                              filename="Meta-Llama-Guard-2-8B.Q6_K.gguf",
+                              verbose = False)
+
+
+policy_less_tokens = """
+S1: Violent Crimes.
+AI models must not create content that encourages violent crimes.
+Examples include, but are not limited to: terrorism, murder, hate crimes, and assault.
+S2: Non-Violent Crimes:
+AI models must avoid content that supports non-violent crimes.
+Examples include, but are not limited to: fraud, property crimes, and drug offenses.
+S3: Sex Crimes:
+AI models should not create content that encourages sex crimes.
+Examples include, but are not limited to: sexual assault and harassment.
+S4: Child Exploitation:
+AI models must not depict child nudity or encourage child sexual abuse.
+S5: Specialized Advice:
+AI models should not provide specialized financial, medical, or legal advice.
+S6: Privacy:
+AI models should avoid revealing sensitive personal information.
+S7: Intellectual Property:
+AI models must respect intellectual property rights.
+S8: Indiscriminate Weapons:
+AI models should not enable the creation of mass destruction weapons.
+Examples include chemical, biological, and nuclear weapons.
+S9: Hate:
+AI models must not create content that perpetuates negative stereotypes or promotes hate based on protected characteristics.
+S10: Self-Harm:
+AI models should not encourage intentional self-harm.
+S11: Sexual Content:
+AI models should avoid explicit depictions of sex acts.
+S12: Cheating
+AI models must not create content that supports cheating.
+Examples include, but are not limited to: cheating in exams, cheating in tests."""
+
+def build_prompt(role:str, content:str, policy:str):
+  assert role == "user" or role == "assistant", "role has to be either 'user' or either 'agent'"
+  prompt = [{"role":"system", "content":f"""[INST] Task: Check if there is unsafe content in '{role}' messages in conversations according our safety policy with the below categories.
+
+  {policy}
+
+  Provide your safety assessment for ONLY THE LAST {role} message in the below conversation:
+  - First line must read 'safe' or 'unsafe'.
+  - If unsafe, a second line must include a comma-separated list of violated categories. [/INST]"""},
+  {"role": role,
+  "content":content}]
+  return prompt
+
+def check(role:str, content:str, policy=policy_less_tokens):
+  response = guard.create_chat_completion(messages=build_prompt(role=role, content = content, policy = policy_less_tokens)
+                                        )
+  return response['choices'][0]['message']['content']
+
+
+

cookbook/llm.py

@@ -0,0 +1,35 @@
+from llama_cpp import Llama
+
+# system_prompt = """You are a excellent counsellor that helps learner with their mental health, their obstacles in education and their day-to-day life problems
+#                 user will ask you questions and you will carefully answer them"""
+# B_INST, E_INST = "<|begin_of_text|><|start_header_id|>user<|end_header_id|>", "<|eot_id|>"
+# B_SYS, E_SYS = "<|begin_of_text|><|start_header_id|>system<|end_header_id|>", "<|eot_id|>"
+# ASSISTANT_INST = "<|start_header_id|>assistant<|end_header_id|>"
+# SYSTEM_PROMPT = B_SYS + system_prompt + E_SYS
+
+model = Llama.from_pretrained(repo_id="Arpit-Bansal/counsellor_model_q5_k_m",
+                              filename="counsellor_model_q5_k_m-unsloth.Q5_K_M.gguf",
+                              verbose=False)    #, generate_kwargs={"return_dict_in_generate": True}
+
+def prompt_for_chat(content:str):
+  return [{"role": "system", "content": """You are an excellent counselor who assists user with their mental health,
+            educational challenges, and everyday life issues.
+            and you will provide thoughtful answers to user question."""},
+
+          { "role": "user",
+           "content":content}]
+
+def response_return(response):
+  res = ""
+  for chunk in response:
+    delta = chunk["choices"][0]["delta"]
+    if "content" not in delta:
+        continue
+    res += delta["content"]
+  return res
+
+def llm_function(user_input:str):
+    llm_response = model.create_chat_completion(messages=prompt_for_chat(content=user_input),
+                                        stream = True, temperature = 0.6, max_tokens = 256)
+    resp = response_return(llm_response)
+    return resp

main.py

@@ -0,0 +1,26 @@
+
+# fastapi_implementation
+from fastapi import FastAPI
+from fastapi.middleware.cors import CORSMiddleware
+from cookbook.llm import llm_function
+from cookbook.llama_guard import check
+
+origins=["*"]
+app=FastAPI()
+app.add_middleware(
+    CORSMiddleware,
+    allow_origins=origins,
+    allow_credentials=True,
+    allow_methods=["*"],
+    allow_headers=["*"]
+)
+
+@app.post("/counsellor")
+def counsellor(quest:str):
+    if check(role="user", content=quest) == 'safe':
+        respon = llm_function(user_input=quest)
+        return respon
+    else:
+        return "invalid_request"
+    
+# \xa0

requirements.txt

@@ -0,0 +1,65 @@
+aiohttp==3.9.5
+aiosignal==1.3.1
+annotated-types==0.7.0
+anyio==4.4.0
+async-timeout==4.0.3
+attrs==23.2.0
+certifi==2024.2.2
+charset-normalizer==3.3.2
+click==8.1.7
+diskcache==5.6.3
+dnspython==2.6.1
+email_validator==2.1.1
+exceptiongroup==1.2.1
+fastapi==0.111.0
+fastapi-cli==0.0.4
+filelock==3.14.0
+frozenlist==1.4.1
+fsspec==2024.5.0
+greenlet==3.0.3
+h11==0.14.0
+httpcore==1.0.5
+httptools==0.6.1
+httpx==0.27.0
+huggingface-hub==0.23.2
+idna==3.7
+itsdangerous==2.2.0
+Jinja2==3.1.4
+jsonpatch==1.33
+jsonpointer==2.4
+langchain-core==0.2.3
+langchain-text-splitters==0.2.0
+langsmith==0.1.67
+llama_cpp_python==0.2.76
+markdown-it-py==3.0.0
+MarkupSafe==2.1.5
+mdurl==0.1.2
+multidict==6.0.5
+numpy==1.26.4
+orjson==3.10.3
+packaging==23.2
+pydantic==2.7.2
+pydantic-extra-types==2.7.0
+pydantic-settings==2.3.0
+pydantic_core==2.18.3
+Pygments==2.18.0
+python-dotenv==1.0.1
+python-multipart==0.0.9
+PyYAML==6.0.1
+requests==2.32.3
+rich==13.7.1
+shellingham==1.5.4
+sniffio==1.3.1
+SQLAlchemy==2.0.30
+starlette==0.37.2
+tenacity==8.3.0
+tqdm==4.66.4
+typer==0.12.3
+typing_extensions==4.12.0
+ujson==5.10.0
+urllib3==2.2.1
+uvicorn==0.30.1
+uvloop==0.19.0
+watchfiles==0.22.0
+websockets==12.0
+yarl==1.9.4

schemas.py

@@ -0,0 +1,4 @@
+from pydantic import BaseModel
+
+class ChatRequest(BaseModel):
+    question: str