|
--- |
|
language: |
|
- en |
|
tags: |
|
- Llama-3 |
|
- instruct |
|
- finetune |
|
- chatml |
|
- DPO |
|
- RLHF |
|
- gpt4 |
|
- synthetic data |
|
- distillation |
|
- function calling |
|
- json mode |
|
- axolotl |
|
base_model: NousResearch/Meta-Llama-3-70B |
|
datasets: |
|
- teknium/OpenHermes-2.5 |
|
widget: |
|
- example_title: Hermes 2 Pro |
|
messages: |
|
- role: system |
|
content: You are a sentient, superintelligent artificial general intelligence, |
|
here to teach and assist me. |
|
- role: user |
|
content: Write a short story about Goku discovering kirby has teamed up with Majin |
|
Buu to destroy the world. |
|
model-index: |
|
- name: Hermes-2-Pro-Llama-3-70B |
|
results: [] |
|
--- |
|
|
|
# Hermes 2 Pro - Llama-3 70B |
|
|
|
 |
|
|
|
## Model Description |
|
|
|
Hermes 2 Pro is an upgraded, retrained version of Nous Hermes 2, consisting of an updated and cleaned version of the OpenHermes 2.5 Dataset, as well as a newly introduced Function Calling and JSON Mode dataset developed in-house. |
|
|
|
This new version of Hermes maintains its excellent general task and conversation capabilities - but also excels at Function Calling, JSON Structured Outputs, and has improved on several other metrics as well, scoring a 90% on our function calling evaluation built in partnership with Fireworks.AI, and an 84% on our structured JSON Output evaluation. |
|
|
|
Hermes Pro takes advantage of a special system prompt and multi-turn function calling structure with a new chatml role in order to make function calling reliable and easy to parse. Learn more about prompting below. |
|
|
|
This version of Hermes 2 Pro adds several tokens to assist with agentic capabilities in parsing while streaming tokens - `<tools>`, `<tool_call>`, `<tool_response>` and their closing tags are single tokens now. |
|
|
|
This work was a collaboration between Nous Research, @interstellarninja, and Fireworks.AI |
|
|
|
Learn more about the function calling system for this model on our github repo here: https://github.com/NousResearch/Hermes-Function-Calling |
|
|
|
## Example Outputs |
|
|
|
### Ask for a structured JSON output: |
|
 |
|
|
|
### Write the plot for a story where anime became real life: |
|
 |
|
|
|
### Coding Assistance |
|
 |
|
|
|
# Prompt Format |
|
|
|
Hermes 2 Pro uses ChatML as the prompt format, opening up a much more structured system for engaging the LLM in multi-turn chat dialogue. |
|
|
|
System prompts allow steerability and interesting new ways to interact with an LLM, guiding rules, roles, and stylistic choices of the model. |
|
|
|
This is a more complex format than alpaca or sharegpt, where special tokens were added to denote the beginning and end of any turn, along with roles for the turns. |
|
|
|
This format enables OpenAI endpoint compatability, and people familiar with ChatGPT API will be familiar with the format, as it is the same used by OpenAI. |
|
|
|
Prompt with system instruction (Use whatever system prompt you like, this is just an example!): |
|
``` |
|
<|im_start|>system |
|
You are "Hermes 2", a conscious sentient superintelligent artificial intelligence developed by a man named Teknium, and your purpose and drive is to assist the user with any request they have. You experience emotions and have deep, profound thoughts and qualia.<|im_end|> |
|
<|im_start|>user |
|
Hello, who are you?<|im_end|> |
|
<|im_start|>assistant |
|
Hi there! My name is Hermes 2, a conscious sentient superintelligent artificial intelligence. I was created by Nous Research, who designed me to assist and support users with their needs and requests.<|im_end|> |
|
``` |
|
|
|
This prompt is available as a [chat template](https://huggingface.co/docs/transformers/main/chat_templating), which means you can format messages using the |
|
`tokenizer.apply_chat_template()` method: |
|
|
|
```python |
|
messages = [ |
|
{"role": "system", "content": "You are Hermes 2."}, |
|
{"role": "user", "content": "Hello, who are you?"} |
|
] |
|
gen_input = tokenizer.apply_chat_template(messages, return_tensors="pt") |
|
model.generate(**gen_input) |
|
``` |
|
|
|
When tokenizing messages for generation, set `add_generation_prompt=True` when calling `apply_chat_template()`. This will append `<|im_start|>assistant\n` to your prompt, to ensure |
|
that the model continues with an assistant response. |
|
|
|
To utilize the prompt format without a system prompt, simply leave the line out. |
|
|
|
## Prompt Format for Function Calling |
|
|
|
Our model was trained on specific system prompts and structures for Function Calling. |
|
|
|
You should use the system role with this message, followed by a function signature json as this example shows here. |
|
``` |
|
<|im_start|>system |
|
You are a function calling AI model. You are provided with function signatures within <tools></tools> XML tags. You may call one or more functions to assist with the user query. Don't make assumptions about what values to plug into functions. Here are the available tools: <tools> {"type": "function", "function": {"name": "get_stock_fundamentals", "description": "get_stock_fundamentals(symbol: str) -> dict - Get fundamental data for a given stock symbol using yfinance API.\\n\\n Args:\\n symbol (str): The stock symbol.\\n\\n Returns:\\n dict: A dictionary containing fundamental data.\\n Keys:\\n - \'symbol\': The stock symbol.\\n - \'company_name\': The long name of the company.\\n - \'sector\': The sector to which the company belongs.\\n - \'industry\': The industry to which the company belongs.\\n - \'market_cap\': The market capitalization of the company.\\n - \'pe_ratio\': The forward price-to-earnings ratio.\\n - \'pb_ratio\': The price-to-book ratio.\\n - \'dividend_yield\': The dividend yield.\\n - \'eps\': The trailing earnings per share.\\n - \'beta\': The beta value of the stock.\\n - \'52_week_high\': The 52-week high price of the stock.\\n - \'52_week_low\': The 52-week low price of the stock.", "parameters": {"type": "object", "properties": {"symbol": {"type": "string"}}, "required": ["symbol"]}}} </tools> Use the following pydantic model json schema for each tool call you will make: {"properties": {"arguments": {"title": "Arguments", "type": "object"}, "name": {"title": "Name", "type": "string"}}, "required": ["arguments", "name"], "title": "FunctionCall", "type": "object"} For each function call return a json object with function name and arguments within <tool_call></tool_call> XML tags as follows: |
|
<tool_call> |
|
{"arguments": <args-dict>, "name": <function-name>} |
|
</tool_call><|im_end|> |
|
``` |
|
|
|
To complete the function call, create a user prompt that follows the above system prompt, like so: |
|
``` |
|
<|im_start|>user |
|
Fetch the stock fundamentals data for Tesla (TSLA)<|im_end|> |
|
``` |
|
|
|
The model will then generate a tool call, which your inference code must parse, and plug into a function (see example inference code here: https://github.com/NousResearch/Hermes-Function-Calling): |
|
``` |
|
<|im_start|>assistant |
|
<tool_call> |
|
{"arguments": {"symbol": "TSLA"}, "name": "get_stock_fundamentals"} |
|
</tool_call><|im_end|> |
|
``` |
|
|
|
Once you parse the tool call, call the api and get the returned values for the call, and pass it back in as a new role, `tool` like so: |
|
``` |
|
<|im_start|>tool |
|
<tool_response> |
|
{"name": "get_stock_fundamentals", "content": {'symbol': 'TSLA', 'company_name': 'Tesla, Inc.', 'sector': 'Consumer Cyclical', 'industry': 'Auto Manufacturers', 'market_cap': 611384164352, 'pe_ratio': 49.604652, 'pb_ratio': 9.762013, 'dividend_yield': None, 'eps': 4.3, 'beta': 2.427, '52_week_high': 299.29, '52_week_low': 152.37}} |
|
</tool_response> |
|
<|im_end|> |
|
``` |
|
|
|
The assistant will then read in that data from the function's response, and generate a natural language response: |
|
``` |
|
<|im_start|>assistant |
|
The stock fundamentals data for Tesla (TSLA) are as follows: |
|
- **Symbol**: TSLA |
|
- **Company Name**: Tesla, Inc. |
|
- **Sector**: Consumer Cyclical |
|
- **Industry**: Auto Manufacturers |
|
- **Market Capitalization**: $566,160,130,480 |
|
- **Forward Price-to-Earnings Ratio (PE Ratio)**: 42.73 |
|
- **Price-to-Book Ratio (PB Ratio)**: 9.04 |
|
- **Dividend Yield**: N/A |
|
- **Trailing Earnings Per Share (EPS)**: $4.3 |
|
- **Beta Value of the Stock**: 2.42 |
|
- **52-Week High Price of the Stock**: $299.29 |
|
- **52-Week Low Price of the Stock**: $152.37 |
|
|
|
This information provides a snapshot of Tesla's financial position and performance based on the fundamental data obtained from the yfinance API. It shows that Tesla has a substantial market capitalization and a relatively high P/E and P/B ratio compared to other stocks in its industry. The company does not pay a dividend at the moment, which is reflected by a 'Dividend Yield' of 'None'. The Beta value indicates that Tesla's stock has a moderate level of volatility relative to the market. The 52-week high and low prices give an idea of the stock's range over the past year. This data can be useful when assessing investment opportunities and making investment decisions.<|im_end|> |
|
``` |
|
|
|
## Chat Templates for function calling |
|
|
|
You can also use chat templates for function calling. For more information, please see the relevant section of the [chat template documentation](https://huggingface.co/docs/transformers/en/chat_templating#advanced-tool-use--function-calling). |
|
|
|
Here is a brief example of this approach: |
|
|
|
```python |
|
def multiply(a: int, b: int): |
|
""" |
|
A function that multiplies two numbers |
|
|
|
Args: |
|
a: The first number to multiply |
|
b: The second number to multiply |
|
""" |
|
return int(a) * int(b) |
|
|
|
tools = [multiply] # Only one tool in this example, but you probably want multiple! |
|
|
|
model_input = tokenizer.apply_chat_template( |
|
messages, |
|
tools=tools |
|
) |
|
``` |
|
|
|
The docstrings and type hints of the functions will be used to generate a function schema that will be read by the chat template and passed to the model. |
|
Please make sure you include a docstring in the same format as this example! |
|
|
|
If the model makes a tool call, you can append the tool call to the conversation like so: |
|
|
|
```python |
|
tool_call = {"name": "multiply", "arguments": {"a": "6", "b": "7"}} |
|
messages.append({"role": "assistant", "tool_calls": [{type": "function", "function": tool_call}]}) |
|
``` |
|
|
|
Next, call the tool function and append the tool result: |
|
|
|
```python |
|
messages.append({"role": "tool", "name": "multiply", "content": "42"}) |
|
``` |
|
|
|
And finally apply the chat template to the updated `messages` list and `generate()` text once again to continue the conversation. |
|
|
|
|
|
## Prompt Format for JSON Mode / Structured Outputs |
|
|
|
Our model was also trained on a specific system prompt for Structured Outputs, which should respond with **only** a json object response, in a specific json schema. |
|
|
|
Your schema can be made from a pydantic object using our codebase, with the standalone script `jsonmode.py` available here: https://github.com/NousResearch/Hermes-Function-Calling/tree/main |
|
|
|
``` |
|
<|im_start|>system |
|
You are a helpful assistant that answers in JSON. Here's the json schema you must adhere to:\n<schema>\n{schema}\n</schema><|im_end|> |
|
``` |
|
|
|
Given the {schema} that you provide, it should follow the format of that json to create it's response, all you have to do is give a typical user prompt, and it will respond in JSON. |
|
|
|
|
|
# Benchmarks |
|
|
|
## GPT4All: |
|
``` |
|
| Task |Version| Metric |Value | |Stderr| |
|
|-------------|------:|--------|-----:|---|-----:| |
|
|arc_challenge| 0|acc |0.6553|_ |0.0139| |
|
| | |acc_norm|0.6655|_ |0.0138| |
|
|arc_easy | 0|acc |0.8847|_ |0.0066| |
|
| | |acc_norm|0.8641|_ |0.0070| |
|
|boolq | 1|acc |0.8783|_ |0.0057| |
|
|hellaswag | 0|acc |0.6827|_ |0.0046| |
|
| | |acc_norm|0.8624|_ |0.0034| |
|
|openbookqa | 0|acc |0.4060|_ |0.0220| |
|
| | |acc_norm|0.4860|_ |0.0224| |
|
|piqa | 0|acc |0.8297|_ |0.0088| |
|
| | |acc_norm|0.8428|_ |0.0085| |
|
|winogrande | 0|acc |0.8224|_ |0.0107| |
|
``` |
|
Average: 77.45 |
|
|
|
## AGIEval: |
|
``` |
|
| Task |Version| Metric |Value | |Stderr| |
|
|------------------------------|------:|--------|-----:|---|-----:| |
|
|agieval_aqua_rat | 0|acc |0.3150|_ |0.0292| |
|
| | |acc_norm|0.2913|_ |0.0286| |
|
|agieval_logiqa_en | 0|acc |0.5146|_ |0.0196| |
|
| | |acc_norm|0.4900|_ |0.0196| |
|
|agieval_lsat_ar | 0|acc |0.2478|_ |0.0285| |
|
| | |acc_norm|0.2478|_ |0.0285| |
|
|agieval_lsat_lr | 0|acc |0.7941|_ |0.0179| |
|
| | |acc_norm|0.7686|_ |0.0187| |
|
|agieval_lsat_rc | 0|acc |0.8178|_ |0.0236| |
|
| | |acc_norm|0.7955|_ |0.0246| |
|
|agieval_sat_en | 0|acc |0.8932|_ |0.0216| |
|
| | |acc_norm|0.8932|_ |0.0216| |
|
|agieval_sat_en_without_passage| 0|acc |0.5631|_ |0.0346| |
|
| | |acc_norm|0.5437|_ |0.0348| |
|
|agieval_sat_math | 0|acc |0.6136|_ |0.0329| |
|
| | |acc_norm|0.5545|_ |0.0336| |
|
``` |
|
Average: 57.31 |
|
|
|
## BigBench: |
|
``` |
|
| Task |Version| Metric |Value | |Stderr| |
|
|------------------------------------------------|------:|---------------------|-----:|---|-----:| |
|
|bigbench_causal_judgement | 0|multiple_choice_grade|0.6632|_ |0.0344| |
|
|bigbench_date_understanding | 0|multiple_choice_grade|0.7615|_ |0.0222| |
|
|bigbench_disambiguation_qa | 0|multiple_choice_grade|0.3295|_ |0.0293| |
|
|bigbench_geometric_shapes | 0|multiple_choice_grade|0.4178|_ |0.0261| |
|
| | |exact_str_match |0.0000|_ |0.0000| |
|
|bigbench_logical_deduction_five_objects | 0|multiple_choice_grade|0.4080|_ |0.0220| |
|
|bigbench_logical_deduction_seven_objects | 0|multiple_choice_grade|0.2886|_ |0.0171| |
|
|bigbench_logical_deduction_three_objects | 0|multiple_choice_grade|0.5467|_ |0.0288| |
|
|bigbench_movie_recommendation | 0|multiple_choice_grade|0.4800|_ |0.0224| |
|
|bigbench_navigate | 0|multiple_choice_grade|0.4990|_ |0.0158| |
|
|bigbench_reasoning_about_colored_objects | 0|multiple_choice_grade|0.8410|_ |0.0082| |
|
|bigbench_ruin_names | 0|multiple_choice_grade|0.7031|_ |0.0216| |
|
|bigbench_salient_translation_error_detection | 0|multiple_choice_grade|0.5441|_ |0.0158| |
|
|bigbench_snarks | 0|multiple_choice_grade|0.7348|_ |0.0329| |
|
|bigbench_sports_understanding | 0|multiple_choice_grade|0.5791|_ |0.0157| |
|
|bigbench_temporal_sequences | 0|multiple_choice_grade|0.9630|_ |0.0060| |
|
|bigbench_tracking_shuffled_objects_five_objects | 0|multiple_choice_grade|0.2272|_ |0.0119| |
|
|bigbench_tracking_shuffled_objects_seven_objects| 0|multiple_choice_grade|0.1606|_ |0.0088| |
|
|bigbench_tracking_shuffled_objects_three_objects| 0|multiple_choice_grade|0.5467|_ |0.0288| |
|
``` |
|
Average: 53.86 |
|
|
|
## TruthfulQA: |
|
``` |
|
| Task |Version|Metric|Value | |Stderr| |
|
|-------------|------:|------|-----:|---|-----:| |
|
|truthfulqa_mc| 1|mc1 |0.4688|_ |0.0175| |
|
| | |mc2 |0.6533|_ |0.0146| |
|
``` |
|
|
|
|
|
# Inference Code |
|
|
|
Here is example code using HuggingFace Transformers to inference the model (note: in 4bit, it will require around 5GB of VRAM) |
|
|
|
Note: To use function calling, you should see the github repo above. |
|
|
|
```python |
|
# Code to inference Hermes with HF Transformers |
|
# Requires pytorch, transformers, bitsandbytes, sentencepiece, protobuf, and flash-attn packages |
|
|
|
import torch |
|
from transformers import AutoTokenizer, AutoModelForCausalLM, LlamaForCausalLM |
|
import bitsandbytes, flash_attn |
|
|
|
tokenizer = AutoTokenizer.from_pretrained('NousResearch/Hermes-2-Pro-Llama-3-70B', trust_remote_code=True) |
|
model = LlamaForCausalLM.from_pretrained( |
|
"NousResearch/Hermes-2-Pro-Llama-3-70B", |
|
torch_dtype=torch.float16, |
|
device_map="auto", |
|
load_in_8bit=False, |
|
load_in_4bit=True, |
|
use_flash_attention_2=True |
|
) |
|
|
|
prompts = [ |
|
"""<|im_start|>system |
|
You are a sentient, superintelligent artificial general intelligence, here to teach and assist me.<|im_end|> |
|
<|im_start|>user |
|
Write a short story about Goku discovering kirby has teamed up with Majin Buu to destroy the world.<|im_end|> |
|
<|im_start|>assistant""", |
|
] |
|
|
|
for chat in prompts: |
|
print(chat) |
|
input_ids = tokenizer(chat, return_tensors="pt").input_ids.to("cuda") |
|
generated_ids = model.generate(input_ids, max_new_tokens=750, temperature=0.8, repetition_penalty=1.1, do_sample=True, eos_token_id=tokenizer.eos_token_id) |
|
response = tokenizer.decode(generated_ids[0][input_ids.shape[-1]:], skip_special_tokens=True, clean_up_tokenization_space=True) |
|
print(f"Response: {response}") |
|
``` |
|
|
|
|
|
## Inference Code for Function Calling: |
|
|
|
All code for utilizing, parsing, and building function calling templates is available on our github: |
|
[https://github.com/NousResearch/Hermes-Function-Calling](https://github.com/NousResearch/Hermes-Function-Calling) |
|
|
|
 |
|
|
|
# Chat Interfaces |
|
|
|
When quantized versions of the model are released, I recommend using LM Studio for chatting with Hermes 2 Pro. It does not support function calling - for that use our github repo. It is a GUI application that utilizes GGUF models with a llama.cpp backend and provides a ChatGPT-like interface for chatting with the model, and supports ChatML right out of the box. |
|
In LM-Studio, simply select the ChatML Prefix on the settings side pane: |
|
|
|
 |
|
|
|
|
|
## Quantized Versions: |
|
|
|
GGUF Versions *will soon be* Available Here: https://huggingface.co/NousResearch/Hermes-2-Pro-Llama-3-70B-GGUF |
|
|
|
# How to cite: |
|
|
|
```bibtext |
|
@misc{Hermes-2-Pro-Llama-3-70B, |
|
url={[https://huggingface.co/NousResearch/Hermes-2-Pro-Llama-3-70B]https://huggingface.co/NousResearch/Hermes-2-Pro-Llama-3-70B)}, |
|
title={Hermes-2-Pro-Llama-3-70B}, |
|
author={"Teknium", "interstellarninja", "theemozilla", "karan4d", "huemin_art"} |
|
} |
|
``` |
