metadata

base_model: google/gemma-2-9b-it
inference: false
license: gemma
library_name: transformers
pipeline_tag: text-generation
model_creator: Google
model_name: gemma-2-9b-it
quantized_by: Second State Inc.
tags:
  - conversational

Gemma-2-9b-it-GGUF

Original Model

google/gemma-2-9b-it

Run with LlamaEdge

LlamaEdge version: v0.12.1 and above

Prompt template

Prompt type: gemma-instruct

Prompt string

<bos><start_of_turn>user
{user_message}<end_of_turn>
<start_of_turn>model
{model_message}<end_of_turn>model

Context size: 8192

Run as LlamaEdge service

wasmedge --dir .:. --nn-preload default:GGML:AUTO:gemma-2-9b-it-Q5_K_M.gguf \
  llama-api-server.wasm \
  --prompt-template gemma-instruct \
  --ctx-size 8192 \
  --model-name gemma-2-9b

Run as LlamaEdge command app

wasmedge --dir .:. \
  --nn-preload default:GGML:AUTO:gemma-2-9b-it-Q5_K_M.gguf \
  llama-chat.wasm \
  --prompt-template gemma-instruct \
  --ctx-size 8192

Quantized GGUF Models

Name	Quant method	Bits	Size	Use case
gemma-2-9b-it-Q2_K.gguf	Q2_K	2	3.81 GB	smallest, significant quality loss - not recommended for most purposes
gemma-2-9b-it-Q3_K_L.gguf	Q3_K_L	3	5.13 GB	small, substantial quality loss
gemma-2-9b-it-Q3_K_M.gguf	Q3_K_M	3	4.76 GB	very small, high quality loss
gemma-2-9b-it-Q3_K_S.gguf	Q3_K_S	3	4.34 GB	very small, high quality loss
gemma-2-9b-it-Q4_0.gguf	Q4_0	4	5.44 GB	legacy; small, very high quality loss - prefer using Q3_K_M
gemma-2-9b-it-Q4_K_M.gguf	Q4_K_M	4	5.76 GB	medium, balanced quality - recommended
gemma-2-9b-it-Q4_K_S.gguf	Q4_K_S	4	5.48 GB	small, greater quality loss
gemma-2-9b-it-Q5_0.gguf	Q5_0	5	6.48 GB	legacy; medium, balanced quality - prefer using Q4_K_M
gemma-2-9b-it-Q5_K_M.gguf	Q5_K_M	5	6.65 GB	large, very low quality loss - recommended
gemma-2-9b-it-Q5_K_S.gguf	Q5_K_S	5	6.48 GB	large, low quality loss - recommended
gemma-2-9b-it-Q6_K.gguf	Q6_K	6	7.59 GB	very large, extremely low quality loss
gemma-2-9b-it-Q8_0.gguf	Q8_0	8	9.83 GB	very large, extremely low quality loss - not recommended
gemma-2-9b-it-f16.gguf	f16	16	18.5 GB

Quantized with llama.cpp b3259