hf-llava-v1.6-34b / README.md
Yazhou Cao
added example in README
8399353
metadata
inference: false
license: apache-2.0


LLaVA Model Card

SGLang

This contains the necessary files to run LLaVA-1.6 34B on SGLang. You can run the server with the following command:

python -m sglang.launch_server --model-path dillonlaird/hf-llava-v1.6-34b --port 30000

There seems to be issues with the chat formatting when using the sglang interface so I recommend querying the server directly and formatting the string yourself:

import requests
from transformers import AutoTokenizer


def generate(image_path: str, prompt: str, tokenizer):
    chat = [
        {"role": "system", "content": "Answer the question."},
        {"role": "user", "content": "<image>\n" + prompt},
    ]
    chat_str = tokenizer.apply_chat_template(chat, tokenize=False)
    chat_str += "<|img_start|>assistant\n"
    sampling_params = {"temperature": 0.2, "max_new_tokens": 1536}
    res = requests.post(
        "http://localhost:30000/generate",
        json={
            "text": chat_str,
            "image_data": image_path,
            "sampling_params": sampling_params,
        },
    )
    return res.json()["text"]


if __name__ == "__main__":
    tokenizer = AutoTokenizer.from_pretrained("liuhaotian/llava-v1.6-34b")
    image_path = "path/to/image.jpg"
    prompt = "What is the name of the mountain?"
    desc = generate(image_path, prompt, tokenizer)

Model details

Model type: LLaVA is an open-source chatbot trained by fine-tuning LLM on multimodal instruction-following data. It is an auto-regressive language model, based on the transformer architecture. Base LLM: NousResearch/Nous-Hermes-2-Yi-34B

Model date: LLaVA-v1.6-34B was trained in December 2023.

Paper or resources for more information: https://llava-vl.github.io/

License

NousResearch/Nous-Hermes-2-Yi-34B license.

Where to send questions or comments about the model: https://github.com/haotian-liu/LLaVA/issues

Intended use

Primary intended uses: The primary use of LLaVA is research on large multimodal models and chatbots.

Primary intended users: The primary intended users of the model are researchers and hobbyists in computer vision, natural language processing, machine learning, and artificial intelligence.

Training dataset

  • 558K filtered image-text pairs from LAION/CC/SBU, captioned by BLIP.
  • 158K GPT-generated multimodal instruction-following data.
  • 500K academic-task-oriented VQA data mixture.
  • 50K GPT-4V data mixture.
  • 40K ShareGPT data.

Evaluation dataset

A collection of 12 benchmarks, including 5 academic VQA benchmarks and 7 recent benchmarks specifically proposed for instruction-following LMMs.