Gemma-7b-it-GGUF / README.md
Xin Liu
Update
ff58cce
|
raw
history blame
3.74 kB
metadata
base_model: google/gemma-7b-it
inference: false
library_name: transformers
license: other
license_name: gemma-terms-of-use
license_link: https://ai.google.dev/gemma/terms
model_creator: Google
model_name: gemma 7b it
quantized_by: Second State Inc.

Gemma-7b-it

Original Model

google/gemma-7b-it

Run with LlamaEdge

  • LlamaEdge version: v0.3.2 (coming soon)

  • Prompt template

    • Prompt type: gemma-instruct

    • Prompt string

      <start_of_turn>user
      {user_message}<end_of_turn>
      <start_of_turn>model
      {model_message}<end_of_turn>model
      
  • Run as LlamaEdge service

    wasmedge --dir .:. --nn-preload default:GGML:AUTO:gemma-7b-it-Q5_K_M.gguf llama-api-server.wasm -p gemma-instruct -c 4096
    
  • Run as LlamaEdge command app

    wasmedge --dir .:. --nn-preload default:GGML:AUTO:gemma-7b-it-Q5_K_M.gguf llama-chat.wasm -p gemma-instruct -c 4096
    

Quantized GGUF Models

Name Quant method Bits Size Use case
gemma-7b-it-Q2_K.gguf Q2_K 2 3.09 GB smallest, significant quality loss - not recommended for most purposes
gemma-7b-it-Q3_K_L.gguf Q3_K_L 3 4.4 GB small, substantial quality loss
gemma-7b-it-Q3_K_M.gguf Q3_K_M 3 4.06 GB very small, high quality loss
gemma-7b-it-Q3_K_S.gguf Q3_K_S 3 3.68 GB very small, high quality loss
gemma-7b-it-Q4_0.gguf Q4_0 4 4.81 GB legacy; small, very high quality loss - prefer using Q3_K_M
gemma-7b-it-Q4_K_M.gguf Q4_K_M 4 5.13 GB medium, balanced quality - recommended
gemma-7b-it-Q4_K_S.gguf Q4_K_S 4 4.84 GB small, greater quality loss
gemma-7b-it-Q5_0.gguf Q5_0 5 5.88 GB legacy; medium, balanced quality - prefer using Q4_K_M
gemma-7b-it-Q5_K_M.gguf Q5_K_M 5 6.04 GB large, very low quality loss - recommended
gemma-7b-it-Q5_K_S.gguf Q5_K_S 5 5.88 GB large, low quality loss - recommended
gemma-7b-it-Q6_K.gguf Q6_K 6 7.01 GB very large, extremely low quality loss
gemma-7b-it-Q8_0.gguf Q8_0 8 9.08 GB very large, extremely low quality loss - not recommended

Quantized with llama.cpp b2230