---

Megrez-3B-Instruct-GGUF

Original Model

Infinigence/Megrez-3B-Instruct

Run with LlamaEdge

• LlamaEdge version: coming soon

• Prompt template

  • Prompt type: megrez

  • Prompt string

    <|role_start|>system<|role_end|>{system_message}<|turn_end|><|role_start|>user<|role_end|>{user_message}<|turn_end|><|role_start|>assistant<|role_end|>
    

• Context size: 32000

• Run as LlamaEdge service

  wasmedge --dir .:. --nn-preload default:GGML:AUTO:Megrez-3B-Instruct-Q5_K_M.gguf \
    llama-api-server.wasm \
    --model-name Megrez-3B-Instruct \
    --prompt-template megrez \
    --ctx-size 32000
  

  For use cases of conversations or article writing, temperature=0.7 is strongly recommended. For use cases of mathematics or logical reasoning, temperature=0.2 is strongly recommended.

• Run as LlamaEdge command app

  wasmedge --dir .:. --nn-preload default:GGML:AUTO:Megrez-3B-Instruct-Q5_K_M.gguf \
    llama-chat.wasm \
    --prompt-template megrez \
    --ctx-size 32000
  

  For use cases of conversations or article writing, temperature=0.7 is strongly recommended. For use cases of mathematics or logical reasoning, temperature=0.2 is strongly recommended.

Quantized GGUF Models

Name	Quant method	Bits	Size	Use case
Megrez-3B-Instruct-Q2_K.gguf	Q2_K	2	1.21 GB	smallest, significant quality loss - not recommended for most purposes
Megrez-3B-Instruct-Q3_K_L.gguf	Q3_K_L	3	1.60 GB	small, substantial quality loss
Megrez-3B-Instruct-Q3_K_M.gguf	Q3_K_M	3	1.50 GB	very small, high quality loss
Megrez-3B-Instruct-Q3_K_S.gguf	Q3_K_S	3	1.38 GB	very small, high quality loss
Megrez-3B-Instruct-Q4_0.gguf	Q4_0	4	1.73 GB	legacy; small, very high quality loss - prefer using Q3_K_M
Megrez-3B-Instruct-Q4_K_M.gguf	Q4_K_M	4	1.81 GB	medium, balanced quality - recommended
Megrez-3B-Instruct-Q4_K_S.gguf	Q4_K_S	4	1.74 GB	small, greater quality loss
Megrez-3B-Instruct-Q5_0.gguf	Q5_0	5	2.05 GB	legacy; medium, balanced quality - prefer using Q4_K_M
Megrez-3B-Instruct-Q5_K_M.gguf	Q5_K_M	5	2.09 GB	large, very low quality loss - recommended
Megrez-3B-Instruct-Q5_K_S.gguf	Q5_K_S	5	2.05 GB	large, low quality loss - recommended
Megrez-3B-Instruct-Q6_K.gguf	Q6_K	6	2.40 GB	very large, extremely low quality loss
Megrez-3B-Instruct-Q8_0.gguf	Q8_0	8	3.10 GB	very large, extremely low quality loss - not recommended
Megrez-3B-Instruct-f16.gguf	f16	16	5.84 GB

Quantized with llama.cpp b4381