legraphista's picture
Upload README.md with huggingface_hub
8b77516 verified
|
raw
history blame
6.47 kB
metadata
base_model: mistralai/Mistral-7B-v0.3
inference: false
library_name: gguf
license: apache-2.0
pipeline_tag: text-generation
quantized_by: legraphista
tags:
  - quantized
  - GGUF
  - imatrix
  - quantization
  - imat
  - imatrix
  - static

Mistral-7B-v0.3-IMat-GGUF

Llama.cpp imatrix quantization of mistralai/Mistral-7B-v0.3

Original Model: mistralai/Mistral-7B-v0.3
Original dtype: BF16 (bfloat16)
Quantized by: llama.cpp b3003
IMatrix dataset: here


Files

IMatrix

Status: βœ… Available
Link: here

Common Quants

Filename Quant type File Size Status Uses IMatrix Is Split
Mistral-7B-v0.3.Q8_0.gguf Q8_0 7.70GB βœ… Available βšͺ No πŸ“¦ No
Mistral-7B-v0.3.Q6_K.gguf Q6_K 5.95GB βœ… Available βšͺ No πŸ“¦ No
Mistral-7B-v0.3.Q4_K.gguf Q4_K 4.37GB βœ… Available 🟒 Yes πŸ“¦ No
Mistral-7B-v0.3.Q3_K.gguf Q3_K 3.52GB βœ… Available 🟒 Yes πŸ“¦ No
Mistral-7B-v0.3.Q2_K.gguf Q2_K 2.72GB βœ… Available 🟒 Yes πŸ“¦ No

All Quants

Filename Quant type File Size Status Uses IMatrix Is Split
Mistral-7B-v0.3.FP16.gguf F16 14.50GB βœ… Available βšͺ No πŸ“¦ No
Mistral-7B-v0.3.BF16.gguf BF16 14.50GB βœ… Available βšͺ No πŸ“¦ No
Mistral-7B-v0.3.Q5_K.gguf Q5_K 5.14GB βœ… Available βšͺ No πŸ“¦ No
Mistral-7B-v0.3.Q5_K_S.gguf Q5_K_S 5.00GB βœ… Available βšͺ No πŸ“¦ No
Mistral-7B-v0.3.Q4_K_S.gguf Q4_K_S 4.14GB βœ… Available 🟒 Yes πŸ“¦ No
Mistral-7B-v0.3.Q3_K_L.gguf Q3_K_L 3.83GB βœ… Available 🟒 Yes πŸ“¦ No
Mistral-7B-v0.3.Q3_K_S.gguf Q3_K_S 3.17GB βœ… Available 🟒 Yes πŸ“¦ No
Mistral-7B-v0.3.Q2_K_S Q2_K_S - ⏳ Processing 🟒 Yes -
Mistral-7B-v0.3.IQ4_NL IQ4_NL - ⏳ Processing 🟒 Yes -
Mistral-7B-v0.3.IQ4_XS IQ4_XS - ⏳ Processing 🟒 Yes -
Mistral-7B-v0.3.IQ3_M IQ3_M - ⏳ Processing 🟒 Yes -
Mistral-7B-v0.3.IQ3_S IQ3_S - ⏳ Processing 🟒 Yes -
Mistral-7B-v0.3.IQ3_XS IQ3_XS - ⏳ Processing 🟒 Yes -
Mistral-7B-v0.3.IQ3_XXS IQ3_XXS - ⏳ Processing 🟒 Yes -
Mistral-7B-v0.3.IQ2_M IQ2_M - ⏳ Processing 🟒 Yes -
Mistral-7B-v0.3.IQ2_S IQ2_S - ⏳ Processing 🟒 Yes -
Mistral-7B-v0.3.IQ2_XS IQ2_XS - ⏳ Processing 🟒 Yes -
Mistral-7B-v0.3.IQ2_XXS IQ2_XXS - ⏳ Processing 🟒 Yes -
Mistral-7B-v0.3.IQ1_M IQ1_M - ⏳ Processing 🟒 Yes -
Mistral-7B-v0.3.IQ1_S IQ1_S - ⏳ Processing 🟒 Yes -

Downloading using huggingface-cli

If you do not have hugginface-cli installed:

pip install -U "huggingface_hub[cli]"

Download the specific file you want:

huggingface-cli download legraphista/Mistral-7B-v0.3-IMat-GGUF --include "Mistral-7B-v0.3.Q8_0.gguf" --local-dir ./

If the model file is big, it has been split into multiple files. In order to download them all to a local folder, run:

huggingface-cli download legraphista/Mistral-7B-v0.3-IMat-GGUF --include "Mistral-7B-v0.3.Q8_0/*" --local-dir Mistral-7B-v0.3.Q8_0
# see FAQ for merging GGUF's

Inference

Llama.cpp

llama.cpp/main -m Mistral-7B-v0.3.Q8_0.gguf --color -i -p "prompt here"

FAQ

Why is the IMatrix not applied everywhere?

According to this investigation, it appears that lower quantizations are the only ones that benefit from the imatrix input (as per hellaswag results).

How do I merge a split GGUF?

  1. Make sure you have gguf-split available
  2. Locate your GGUF chunks folder (ex: Mistral-7B-v0.3.Q8_0)
  3. Run gguf-split --merge Mistral-7B-v0.3.Q8_0/Mistral-7B-v0.3.Q8_0-00001-of-XXXXX.gguf Mistral-7B-v0.3.Q8_0.gguf
    • Make sure to point gguf-split to the first chunk of the split.

Got a suggestion? Ping me @legraphista!