Instructions to use GritLM/GritLM-7B with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use GritLM/GritLM-7B with Transformers:

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

pipe = pipeline("text-generation", model="GritLM/GritLM-7B", trust_remote_code=True)
messages = [
    {"role": "user", "content": "Who are you?"},
]
pipe(messages)

# Load model directly
from transformers import AutoTokenizer, AutoModelForCausalLM

tokenizer = AutoTokenizer.from_pretrained("GritLM/GritLM-7B", trust_remote_code=True)
model = AutoModelForCausalLM.from_pretrained("GritLM/GritLM-7B", trust_remote_code=True)
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

vLLM

How to use GritLM/GritLM-7B with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "GritLM/GritLM-7B"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "GritLM/GritLM-7B",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker

docker model run hf.co/GritLM/GritLM-7B

SGLang

How to use GritLM/GritLM-7B 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 "GritLM/GritLM-7B" \
    --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": "GritLM/GritLM-7B",
		"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 "GritLM/GritLM-7B" \
        --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": "GritLM/GritLM-7B",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Docker Model Runner
How to use GritLM/GritLM-7B with Docker Model Runner:
```
docker model run hf.co/GritLM/GritLM-7B
```

How did you apply bi-directional attention for embedding only?

by Mayfull - opened Jun 14, 2024

Discussion

Mayfull

Jun 14, 2024

•

edited Jun 14, 2024

Hello. Thanks for your great work and helpful code in your github repo! It helps me a lot.
I am wondering how you implemented bi-directional attention only for embedding while maintaining causal attention for generation simultaneously.
What I understood so far from your code (https://github.com/ContextualAI/gritlm/blob/main/gritlm/training/model.py) is as follow.

I get that you put 'is_causal' key into the "kwargs" variable(Dict) and pass it to model call if self.attn[:2] == 'bb' (default to 'bbcc').
But I can't understand how this can remove causal mask even if there is no argument named 'is_causal' in forward function in most model (ex. MistralForCausalLM).

Would you mind giving me slight advice or information how it works?
Thank you so much.

        kwargs = {'input_ids': features.get('input_ids'), 'attention_mask': attention_mask}

        if self.attn[:2] == 'cb':
            kwargs['instruction_lens'] = instruction_lens
        elif self.attn[:2] == 'bb':
            kwargs['is_causal'] = False
        out = (getattr(self.model, self.embedding_attr) if self.embedding_attr else self.model)(**kwargs)[0]

Muennighoff

GritLM org Jun 14, 2024

The repo has a custom modeling file that has the is_causal kwarg: https://huggingface.co/GritLM/GritLM-7B/blob/main/modeling_gritlm7b.py

Mayfull

Jun 14, 2024

Thank you so much! Now I understood!

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