Xuan-Son Nguyen 36667c8edc clip : revert the change of BOI/EOI token for GLM-edge (⚠️ breaking change) (#13259) 8 月之前
..
android 1d36b3670b llama : move end-user examples to tools directory (#13249) 8 月之前
CMakeLists.txt 1d36b3670b llama : move end-user examples to tools directory (#13249) 8 月之前
README-quantize.md 1d36b3670b llama : move end-user examples to tools directory (#13249) 8 月之前
README.md 1d36b3670b llama : move end-user examples to tools directory (#13249) 8 月之前
clip-impl.h 36667c8edc clip : revert the change of BOI/EOI token for GLM-edge (⚠️ breaking change) (#13259) 8 月之前
clip-quantize-cli.cpp 1d36b3670b llama : move end-user examples to tools directory (#13249) 8 月之前
clip.cpp 36667c8edc clip : revert the change of BOI/EOI token for GLM-edge (⚠️ breaking change) (#13259) 8 月之前
clip.h 1d36b3670b llama : move end-user examples to tools directory (#13249) 8 月之前
convert_image_encoder_to_gguf.py 1d36b3670b llama : move end-user examples to tools directory (#13249) 8 月之前
deprecation-warning.cpp 1d36b3670b llama : move end-user examples to tools directory (#13249) 8 月之前
glmedge-convert-image-encoder-to-gguf.py 1d36b3670b llama : move end-user examples to tools directory (#13249) 8 月之前
glmedge-surgery.py 1d36b3670b llama : move end-user examples to tools directory (#13249) 8 月之前
llava.cpp 1d36b3670b llama : move end-user examples to tools directory (#13249) 8 月之前
llava.h 1d36b3670b llama : move end-user examples to tools directory (#13249) 8 月之前
llava_surgery.py 1d36b3670b llama : move end-user examples to tools directory (#13249) 8 月之前
llava_surgery_v2.py 1d36b3670b llama : move end-user examples to tools directory (#13249) 8 月之前
minicpmv-convert-image-encoder-to-gguf.py 1d36b3670b llama : move end-user examples to tools directory (#13249) 8 月之前
minicpmv-surgery.py 1d36b3670b llama : move end-user examples to tools directory (#13249) 8 月之前
mtmd-cli.cpp 1d36b3670b llama : move end-user examples to tools directory (#13249) 8 月之前
mtmd.cpp 36667c8edc clip : revert the change of BOI/EOI token for GLM-edge (⚠️ breaking change) (#13259) 8 月之前
mtmd.h 1d36b3670b llama : move end-user examples to tools directory (#13249) 8 月之前
qwen2vl-test.cpp 1d36b3670b llama : move end-user examples to tools directory (#13249) 8 月之前
requirements.txt 1d36b3670b llama : move end-user examples to tools directory (#13249) 8 月之前
test-1.jpeg 1d36b3670b llama : move end-user examples to tools directory (#13249) 8 月之前
tests.sh 1d36b3670b llama : move end-user examples to tools directory (#13249) 8 月之前

README-quantize.md

Quantizing CLIP Visual Projector

This is the tool for quantizing the CLIP visual projector model. Quantization reduces the precision of the model's weights, which can significantly decrease the model size and improve inference speed, often with minimal impact on performance.

Usage

To quantize a CLIP visual projector model, use the following command:

./bin/llama-llava-clip-quantize-cli /path/to/ggml-model-f32.gguf /path/to/ggml-model-quantized.gguf <type>

After the quantization, the visual projector can be used freely with the existing LLAVA cli (LLAVA, Qwen2VL, etc).

Arguments

  • /path/to/ggml-model-f32.gguf: The path to the input model file in FP32 or FP16 format.
  • /path/to/ggml-model-quantized.gguf: The path where the quantized model will be saved.
  • <type>: The quantization type to apply. This should be an integer corresponding to one of the quantization types defined in the enum ggml_type.

Quantization Types

The following quantization types are supported, based on the enum ggml_type definition:

  • 2 - q4_0: 4-bit quantization with a single scale value.
  • 3 - q4_1: 4-bit quantization with a separate scale value for each block.
  • 6 - q5_0: 5-bit quantization with a single scale value.
  • 7 - q5_1: 5-bit quantization with a separate scale value for each block.
  • 8 - q8_0: 8-bit quantization with a single scale value.

Example

To quantize a model using the q4_0 quantization type, you would run:

./bin/llama-llava-clip-quantize-cli /path/to/ggml-model-f32.gguf /path/to/ggml-model-quantized.gguf 2

This command will generate a quantized model at /path/to/ggml-model-quantized.gguf using the q4_0 quantization method.

Notes

  • Quantization can lead to a loss in model accuracy, depending on the chosen quantization type. It is recommended to evaluate the quantized model's performance on your specific task to ensure it meets your requirements.
  • The quantized model will typically be smaller in size and faster to run, making it more suitable for deployment in resource-constrained environments.