llava llama 3 8b v1_1 gguf

Introduction

The llava-llama-3-8b-v1_1 model is a variant of the LLaVA model fine-tuned from Meta-Llama-3-8B-Instruct and CLIP-ViT-Large-patch14-336 using datasets like ShareGPT4V-PT and InternVL-SFT. It is designed for image-to-text tasks and is available in GGUF format.

Architecture

The model employs a visual encoder with CLIP-Large and uses a multi-layer perceptron (MLP) for projecting visual features. It maintains a resolution of 336 and follows a strategy of frozen language and vision transformers during pretraining, while fully utilizing the language model and LoRA for vision transformers in fine-tuning.

Training

The fine-tuning of llava-llama-3-8b-v1_1 is performed with datasets such as ShareGPT4V-PT and InternVL-SFT, using a strategy that leverages frozen vision transformers and fully engaged language models.

Guide: Running Locally

1. Download Models:

   • wget https://huggingface.co/xtuner/llava-llama-3-8b-v1_1-gguf/resolve/main/llava-llama-3-8b-v1_1-mmproj-f16.gguf
   • wget https://huggingface.co/xtuner/llava-llama-3-8b-v1_1-gguf/resolve/main/llava-llama-3-8b-v1_1-f16.gguf
   • wget https://huggingface.co/xtuner/llava-llama-3-8b-v1_1-gguf/resolve/main/llava-llama-3-8b-v1_1-int4.gguf

2. Using Ollama:

   • For FP16:

     ollama create llava-llama3-f16 -f ./OLLAMA_MODELFILE_F16
     ollama run llava-llama3-f16 "xx.png Describe this image"
     

   • For INT4:

     ollama create llava-llama3-int4 -f ./OLLAMA_MODELFILE_INT4
     ollama run llava-llama3-int4 "xx.png Describe this image"
     

3. Using Llama.cpp:

   • Build llama.cpp and llava-cli.
   • For FP16:

     ./llava-cli -m ./llava-llama-3-8b-v1_1-f16.gguf --mmproj ./llava-llama-3-8b-v1_1-mmproj-f16.gguf --image YOUR_IMAGE.jpg -c 4096 -e -p "<|start_header_id|>user<|end_header_id|>\n\n<image>\nDescribe this image<|eot_id|><|start_header_id|>assistant<|end_header_id|>\n\n"
     

   • For INT4:

     ./llava-cli -m ./llava-llama-3-8b-v1_1-int4.gguf --mmproj ./llava-llama-3-8b-v1_1-mmproj-f16.gguf --image YOUR_IMAGE.jpg -c 4096 -e -p "<|start_header_id|>user<|end_header_id|>\n\n<image>\nDescribe this image<|eot_id|><|start_header_id|>assistant<|end_header_id|>\n\n"
     

Suggested Cloud GPUs: Consider using cloud services like AWS, GCP, or Azure for enhanced performance and scalability.

License

Refer to the XTuner GitHub repository for licensing details.