koesn commited on
Commit
809a158
1 Parent(s): be4c25c

Update README.md

Browse files
Files changed (1) hide show
  1. README.md +255 -3
README.md CHANGED
@@ -1,3 +1,255 @@
1
- ---
2
- license: llama3.1
3
- ---
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ ---
2
+ language:
3
+ - en
4
+ license: llama3
5
+ tags:
6
+ - Llama-3
7
+ - instruct
8
+ - finetune
9
+ - chatml
10
+ - gpt4
11
+ - synthetic data
12
+ - distillation
13
+ - function calling
14
+ - json mode
15
+ - axolotl
16
+ - roleplaying
17
+ - chat
18
+ base_model: meta-llama/Meta-Llama-3.1-8B
19
+ widget:
20
+ - example_title: Hermes 3
21
+ messages:
22
+ - role: system
23
+ content: You are a sentient, superintelligent artificial general intelligence,
24
+ here to teach and assist me.
25
+ - role: user
26
+ content: What is the meaning of life?
27
+ model-index:
28
+ - name: Hermes-3-Llama-3.1-70B
29
+ results: []
30
+ ---
31
+
32
+ ## Description
33
+ This repo contains GGUF format model files for Hermes-3-Llama-3.1-8B.
34
+
35
+ ## Files Provided
36
+ | Name | Quant | Bits | File Size | Remark |
37
+ | ---------------------------- | ------- | ---- | --------- | -------------------------------- |
38
+ | hermes-3-llama-3.1-8b.Q2_K.gguf | Q2_K | 2 | 3.18 GB | 2.96G, +3.5199 ppl @ Llama-3-8B |
39
+ | hermes-3-llama-3.1-8b.Q3_K.gguf | Q3_K | 3 | 4.02 GB | 3.74G, +0.6569 ppl @ Llama-3-8B |
40
+ | hermes-3-llama-3.1-8b.Q4_0.gguf | Q4_0 | 4 | 4.66 GB | 4.34G, +0.4685 ppl @ Llama-3-8B |
41
+ | hermes-3-llama-3.1-8b.Q4_K.gguf | Q4_K | 4 | 4.92 GB | 4.58G, +0.1754 ppl @ Llama-3-8B |
42
+ | hermes-3-llama-3.1-8b.Q5_K.gguf | Q5_K | 5 | 5.73 GB | 5.33G, +0.0569 ppl @ Llama-3-8B |
43
+ | hermes-3-llama-3.1-8b.Q6_K.gguf | Q6_K | 6 | 6.60 GB | 6.14G, +0.0217 ppl @ Llama-3-8B |
44
+ | hermes-3-llama-3.1-8b.Q8_0.gguf | Q8_0 | 8 | 8.54 GB | 7.96G, +0.0026 ppl @ Llama-3-8B |
45
+
46
+ ## Parameters
47
+ | path | type | architecture | rope_theta | sliding_win | max_pos_embed |
48
+ | ---------------------------------- | ----- | ---------------- | ---------- | ----------- | ------------- |
49
+ | NousResearch/Hermes-3-Llama-3.1-8B | llama | LlamaForCausalLM | 500000.0 | null | 131072 |
50
+
51
+
52
+ # Original Model Card
53
+
54
+ # Hermes 3 - Llama-3.1 8B
55
+
56
+ ![image/jpeg](https://cdn-uploads.huggingface.co/production/uploads/6317aade83d8d2fd903192d9/bMcZ3sNNQK8SRZpHXBmwM.jpeg)
57
+
58
+ ## Model Description
59
+
60
+ Hermes 3 is the latest version of our flagship Hermes series of LLMs by Nous Research.
61
+
62
+ For more details on new capabilities, training results, and more, see the [**Hermes 3 Technical Report**](https://arxiv.org/abs/2408.11857).
63
+
64
+ Hermes 3 is a generalist language model with many improvements over Hermes 2, including advanced agentic capabilities, much better roleplaying, reasoning, multi-turn conversation, long context coherence, and improvements across the board.
65
+
66
+ The ethos of the Hermes series of models is focused on aligning LLMs to the user, with powerful steering capabilities and control given to the end user.
67
+
68
+ The Hermes 3 series builds and expands on the Hermes 2 set of capabilities, including more powerful and reliable function calling and structured output capabilities, generalist assistant capabilities, and improved code generation skills.
69
+
70
+
71
+ # Benchmarks
72
+
73
+ Hermes 3 is competitive, if not superior, to Llama-3.1 Instruct models at general capabilities, with varying strengths and weaknesses attributable between the two.
74
+
75
+ Full benchmark comparisons below:
76
+
77
+ ![image/png](https://cdn-uploads.huggingface.co/production/uploads/6317aade83d8d2fd903192d9/DIMca3M0U-ArWwtyIbF-k.png)
78
+
79
+
80
+ # Prompt Format
81
+
82
+ Hermes 3 uses ChatML as the prompt format, opening up a much more structured system for engaging the LLM in multi-turn chat dialogue.
83
+
84
+ System prompts allow steerability and interesting new ways to interact with an LLM, guiding rules, roles, and stylistic choices of the model.
85
+
86
+ 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.
87
+
88
+ 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.
89
+
90
+ Prompt with system instruction (Use whatever system prompt you like, this is just an example!):
91
+ ```
92
+ <|im_start|>system
93
+ You are Hermes 3, 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|>
94
+ <|im_start|>user
95
+ Hello, who are you?<|im_end|>
96
+ <|im_start|>assistant
97
+ Hi there! My name is Hermes 3, 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|>
98
+ ```
99
+
100
+ This prompt is available as a [chat template](https://huggingface.co/docs/transformers/main/chat_templating), which means you can format messages using the
101
+ `tokenizer.apply_chat_template()` method:
102
+
103
+ ```python
104
+ messages = [
105
+ {"role": "system", "content": "You are Hermes 3."},
106
+ {"role": "user", "content": "Hello, who are you?"}
107
+ ]
108
+ gen_input = tokenizer.apply_chat_template(messages, return_tensors="pt")
109
+ model.generate(**gen_input)
110
+ ```
111
+
112
+ 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
113
+ that the model continues with an assistant response.
114
+
115
+ To utilize the prompt format without a system prompt, simply leave the line out.
116
+
117
+
118
+ ## Prompt Format for Function Calling
119
+
120
+ Our model was trained on specific system prompts and structures for Function Calling.
121
+
122
+ You should use the system role with this message, followed by a function signature json as this example shows here.
123
+ ```
124
+ <|im_start|>system
125
+ 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:
126
+ <tool_call>
127
+ {"arguments": <args-dict>, "name": <function-name>}
128
+ </tool_call><|im_end|>
129
+ ```
130
+
131
+ To complete the function call, create a user prompt that follows the above system prompt, like so:
132
+ ```
133
+ <|im_start|>user
134
+ Fetch the stock fundamentals data for Tesla (TSLA)<|im_end|>
135
+ ```
136
+
137
+ 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):
138
+ ```
139
+ <|im_start|>assistant
140
+ <tool_call>
141
+ {"arguments": {"symbol": "TSLA"}, "name": "get_stock_fundamentals"}
142
+ </tool_call><|im_end|>
143
+ ```
144
+
145
+ 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:
146
+ ```
147
+ <|im_start|>tool
148
+ <tool_response>
149
+ {"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}}
150
+ </tool_response>
151
+ <|im_end|>
152
+ ```
153
+
154
+ The assistant will then read in that data from the function's response, and generate a natural language response:
155
+ ```
156
+ <|im_start|>assistant
157
+ The stock fundamentals data for Tesla (TSLA) are as follows:
158
+ - **Symbol**: TSLA
159
+ - **Company Name**: Tesla, Inc.
160
+ - **Sector**: Consumer Cyclical
161
+ - **Industry**: Auto Manufacturers
162
+ - **Market Capitalization**: $566,160,130,480
163
+ - **Forward Price-to-Earnings Ratio (PE Ratio)**: 42.73
164
+ - **Price-to-Book Ratio (PB Ratio)**: 9.04
165
+ - **Dividend Yield**: N/A
166
+ - **Trailing Earnings Per Share (EPS)**: $4.3
167
+ - **Beta Value of the Stock**: 2.42
168
+ - **52-Week High Price of the Stock**: $299.29
169
+ - **52-Week Low Price of the Stock**: $152.37
170
+
171
+ 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|>
172
+ ```
173
+
174
+
175
+ ## Prompt Format for JSON Mode / Structured Outputs
176
+
177
+ 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.
178
+
179
+ 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
180
+
181
+ ```
182
+ <|im_start|>system
183
+ 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|>
184
+ ```
185
+
186
+ 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.
187
+
188
+
189
+ # Inference
190
+
191
+ Here is example code using HuggingFace Transformers to inference the model
192
+
193
+ ```python
194
+ # Code to inference Hermes with HF Transformers
195
+ # Requires pytorch, transformers, bitsandbytes, sentencepiece, protobuf, and flash-attn packages
196
+
197
+ import torch
198
+ from transformers import AutoTokenizer, AutoModelForCausalLM, LlamaForCausalLM
199
+ import bitsandbytes, flash_attn
200
+
201
+ tokenizer = AutoTokenizer.from_pretrained('NousResearch/Hermes-3-Llama-3.1-8B', trust_remote_code=True)
202
+ model = LlamaForCausalLM.from_pretrained(
203
+ "NousResearch/Hermes-3-Llama-3.1-8B",
204
+ torch_dtype=torch.float16,
205
+ device_map="auto",
206
+ load_in_8bit=False,
207
+ load_in_4bit=True,
208
+ use_flash_attention_2=True
209
+ )
210
+
211
+ prompts = [
212
+ """<|im_start|>system
213
+ You are a sentient, superintelligent artificial general intelligence, here to teach and assist me.<|im_end|>
214
+ <|im_start|>user
215
+ Write a short story about Goku discovering kirby has teamed up with Majin Buu to destroy the world.<|im_end|>
216
+ <|im_start|>assistant""",
217
+ ]
218
+
219
+ for chat in prompts:
220
+ print(chat)
221
+ input_ids = tokenizer(chat, return_tensors="pt").input_ids.to("cuda")
222
+ 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)
223
+ response = tokenizer.decode(generated_ids[0][input_ids.shape[-1]:], skip_special_tokens=True, clean_up_tokenization_space=True)
224
+ print(f"Response: {response}")
225
+ ```
226
+
227
+ You can also run this model with vLLM, by running the following in your terminal after `pip install vllm`
228
+
229
+ `vllm serve NousResearch/Hermes-3-Llama-3.1-8B`
230
+
231
+ ## Inference Code for Function Calling:
232
+
233
+ All code for utilizing, parsing, and building function calling templates is available on our github:
234
+ [https://github.com/NousResearch/Hermes-Function-Calling](https://github.com/NousResearch/Hermes-Function-Calling)
235
+
236
+ ![image/png](https://cdn-uploads.huggingface.co/production/uploads/6317aade83d8d2fd903192d9/oi4CiGh50xmoviUQnh8R3.png)
237
+
238
+
239
+ ## Quantized Versions:
240
+
241
+ GGUF Quants: https://huggingface.co/NousResearch/Hermes-3-Llama-3.1-8B-GGUF
242
+
243
+ # How to cite:
244
+
245
+ ```bibtext
246
+ @misc{teknium2024hermes3technicalreport,
247
+ title={Hermes 3 Technical Report},
248
+ author={Ryan Teknium and Jeffrey Quesnelle and Chen Guang},
249
+ year={2024},
250
+ eprint={2408.11857},
251
+ archivePrefix={arXiv},
252
+ primaryClass={cs.CL},
253
+ url={https://arxiv.org/abs/2408.11857},
254
+ }
255
+ ```