gemma-3n-E4B-it-GGUF
Original Model
Run with LlamaEdge
-
LlamaEdge version: v0.23.1 and above
-
Prompt template
-
Prompt type:
gemma-3 -
Prompt string
<bos><start_of_turn>user {user_message}<end_of_turn> <start_of_turn>model {model_message}<end_of_turn>model
-
-
Context size:
32000 -
Run as LlamaEdge service
wasmedge --dir .:. --nn-preload default:GGML:AUTO:gemma-3n-E4B-it-Q5_K_M.gguf \ llama-api-server.wasm \ --prompt-template gemma-3 \ --ctx-size 32000 \ --model-name gemma-3n-E4B-it -
Run as LlamaEdge command app
wasmedge --dir .:. \ --nn-preload default:GGML:AUTO:gemma-3n-E4B-it-Q5_K_M.gguf \ llama-chat.wasm \ --prompt-template gemma-3 \ --ctx-size 32000
Quantized GGUF Models
| Name | Quant method | Bits | Size | Use case |
|---|---|---|---|---|
| gemma-3n-E4B-it-Q2_K.gguf | Q2_K | 2 | 2.76 GB | smallest, significant quality loss - not recommended for most purposes |
| gemma-3n-E4B-it-Q3_K_L.gguf | Q3_K_L | 3 | 3.60 GB | small, substantial quality loss |
| gemma-3n-E4B-it-Q3_K_M.gguf | Q3_K_M | 3 | 3.44 GB | very small, high quality loss |
| gemma-3n-E4B-it-Q3_K_S.gguf | Q3_K_S | 3 | 3.25 GB | very small, high quality loss |
| gemma-3n-E4B-it-Q4_0.gguf | Q4_0 | 4 | 4.09 GB | legacy; small, very high quality loss - prefer using Q3_K_M |
| gemma-3n-E4B-it-Q4_K_M.gguf | Q4_K_M | 4 | 4.24 GB | medium, balanced quality - recommended |
| gemma-3n-E4B-it-Q4_K_S.gguf | Q4_K_S | 4 | 4.10 GB | small, greater quality loss |
| gemma-3n-E4B-it-Q5_0.gguf | Q5_0 | 5 | 4.87 GB | legacy; medium, balanced quality - prefer using Q4_K_M |
| gemma-3n-E4B-it-Q5_K_M.gguf | Q5_K_M | 5 | 4.95 GB | large, very low quality loss - recommended |
| gemma-3n-E4B-it-Q5_K_S.gguf | Q5_K_S | 5 | 4.87 GB | large, low quality loss - recommended |
| gemma-3n-E4B-it-Q6_K.gguf | Q6_K | 6 | 5.70 GB | very large, extremely low quality loss |
| gemma-3n-E4B-it-Q8_0.gguf | Q8_0 | 8 | 7.35 GB | very large, extremely low quality loss - not recommended |
| gemma-3n-E4B-it-f16.gguf | f16 | 16 | 13.7 GB |
Quantized with llama.cpp b5830