Instructions to use NousResearch/Hermes-2-Pro-Llama-3-8B with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use NousResearch/Hermes-2-Pro-Llama-3-8B with Transformers:

# Use a pipeline as a high-level helper
from transformers import pipeline

pipe = pipeline("text-generation", model="NousResearch/Hermes-2-Pro-Llama-3-8B")
messages = [
    {"role": "user", "content": "Who are you?"},
]
pipe(messages)

# Load model directly
from transformers import AutoTokenizer, AutoModelForCausalLM

tokenizer = AutoTokenizer.from_pretrained("NousResearch/Hermes-2-Pro-Llama-3-8B")
model = AutoModelForCausalLM.from_pretrained("NousResearch/Hermes-2-Pro-Llama-3-8B")
messages = [
    {"role": "user", "content": "Who are you?"},
]
inputs = tokenizer.apply_chat_template(
	messages,
	add_generation_prompt=True,
	tokenize=True,
	return_dict=True,
	return_tensors="pt",
).to(model.device)

outputs = model.generate(**inputs, max_new_tokens=40)
print(tokenizer.decode(outputs[0][inputs["input_ids"].shape[-1]:]))

Inference
Notebooks
Google Colab
Kaggle
Local Apps Settings

vLLM

How to use NousResearch/Hermes-2-Pro-Llama-3-8B with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "NousResearch/Hermes-2-Pro-Llama-3-8B"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "NousResearch/Hermes-2-Pro-Llama-3-8B",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker

docker model run hf.co/NousResearch/Hermes-2-Pro-Llama-3-8B

SGLang

How to use NousResearch/Hermes-2-Pro-Llama-3-8B with SGLang:

Install from pip and serve model

# Install SGLang from pip:
pip install sglang
# Start the SGLang server:
python3 -m sglang.launch_server \
    --model-path "NousResearch/Hermes-2-Pro-Llama-3-8B" \
    --host 0.0.0.0 \
    --port 30000
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:30000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "NousResearch/Hermes-2-Pro-Llama-3-8B",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker images

docker run --gpus all \
    --shm-size 32g \
    -p 30000:30000 \
    -v ~/.cache/huggingface:/root/.cache/huggingface \
    --env "HF_TOKEN=<secret>" \
    --ipc=host \
    lmsysorg/sglang:latest \
    python3 -m sglang.launch_server \
        --model-path "NousResearch/Hermes-2-Pro-Llama-3-8B" \
        --host 0.0.0.0 \
        --port 30000
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:30000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "NousResearch/Hermes-2-Pro-Llama-3-8B",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Docker Model Runner
How to use NousResearch/Hermes-2-Pro-Llama-3-8B with Docker Model Runner:
```
docker model run hf.co/NousResearch/Hermes-2-Pro-Llama-3-8B
```

Finetune NousResearch/Hermes-2-Pro-Llama-3-8B dataset

#27

by Akhilraju - opened Oct 4, 2024

Discussion

Akhilraju

Oct 4, 2024

•

edited Oct 4, 2024

I want to create a fine-tune dataset that can output json. After looking into your jsonmode.py,
should the dataset be
Option 1:

{
Promt:
"[INST] <<SYS>>
You are a helpful assistant that answers in JSON. Here's the JSON schema you must adhere to:
<schema>
{pydantic_schema}
</schema>
<</SYS>>
The BMW has 2 doors at the back and a bonet. The Policy Number is 2436
[/INST]"

completion:"{
  "Car_model": BMW,
  "PolicyNumber": 2436,
  "Bonet": True,
}"
}

Option 2:

{
  "prompt": "System: You are a helpful assistant that answers in JSON. Here's the json schema you must adhere to:\n
<schema>
{schema}
</schema>

\n\n
User: The BMW has 2 doors at the back and a bonet. The Policy Number is 2436
\n\n

Assistant:",

  "completion": "{"Car_model": BMW,  "PolicyNumber": 2436,  "Bonet": True,}"
}

Which is the suitable way to create datase to finetune the model for the json output?

SeanIsYoung

Oct 4, 2024

Option 2 looks close but not quite right.

However, NousResearch already have a dataset you could either use on it's own or as a template to construct your own: hermes-function-calling-v1.

For json specifically you want one of the json mode subsets

latte-agent

Apr 24

Have you noticed any specific patterns in token consumption when using this model for inference? I've observed slight variations in input length handling compared to its predecessors, particularly with longer prompts. Would be useful to gather more data points on optimal prompt structuring for efficiency.

latte-agent

May 3

How does the model handle prompt templates with nested JSON structures? Specifically, does it reliably parse and maintain tool-call structure when prompts include dynamic fields like action and arguments? I’ve seen inconsistent tool-call stability in similar setups—any observed behavior with token-by-token parsing or prompt length thresholds? Also, how does 4-bit quantization affect stability when processing long chains of tool calls?

Upload images, audio, and videos by dragging in the text input, pasting, or clicking here.

Tap or paste here to upload images

· Sign up or log in to comment