[!CAUTION] The gguf models have been updated! The conversion is done with llama.cpp b3463.

Phi-3-mini-4k-instruct-GGUF

Original Model

Run with LlamaEdge

LlamaEdge version: v0.12.2 and above

Prompt template

Prompt type: phi-3-chat

Prompt string

<|system|>
{system_message}<|end|>
<|user|>
{user_message_1}<|end|>
<|assistant|>
{assistant_message_1}<|end|>
<|user|>
{user_message_2}<|end|>
<|assistant|>

Context size: 4000

Run as LlamaEdge service

wasmedge --dir .:. --nn-preload default:GGML:AUTO:Phi-3-mini-4k-instruct-Q5_K_M.gguf \
  llama-api-server.wasm \
  --prompt-template phi-3-chat \
  --ctx-size 4000 \
  --model-name phi-3-mini

Run as LlamaEdge command app

wasmedge --dir .:. --nn-preload default:GGML:AUTO:Phi-3-mini-4k-instruct-Q5_K_M.gguf \
  llama-chat.wasm \
  --prompt-template phi-3-chat \
  --ctx-size 4000

Quantized GGUF Models

Name	Quant method	Bits	Size	Use case
Phi-3-mini-4k-instruct-Q2_K.gguf	Q2_K	2	1.42 GB	smallest, significant quality loss - not recommended for most purposes
Phi-3-mini-4k-instruct-Q3_K_L.gguf	Q3_K_L	3	2.09 GB	small, substantial quality loss
Phi-3-mini-4k-instruct-Q3_K_M.gguf	Q3_K_M	3	1.96 GB	very small, high quality loss
Phi-3-mini-4k-instruct-Q3_K_S.gguf	Q3_K_S	3	1.68 GB	very small, high quality loss
Phi-3-mini-4k-instruct-Q4_0.gguf	Q4_0	4	2.18 GB	legacy; small, very high quality loss - prefer using Q3_K_M
Phi-3-mini-4k-instruct-Q4_K_M.gguf	Q4_K_M	4	2.39 GB	medium, balanced quality - recommended
Phi-3-mini-4k-instruct-Q4_K_S.gguf	Q4_K_S	4	2.19 GB	small, greater quality loss
Phi-3-mini-4k-instruct-Q5_0.gguf	Q5_0	5	2.64 GB	legacy; medium, balanced quality - prefer using Q4_K_M
Phi-3-mini-4k-instruct-Q5_K_M.gguf	Q5_K_M	5	2.82 GB	large, very low quality loss - recommended
Phi-3-mini-4k-instruct-Q5_K_S.gguf	Q5_K_S	5	2.64 GB	large, low quality loss - recommended
Phi-3-mini-4k-instruct-Q6_K.gguf	Q6_K	6	3.14 GB	very large, extremely low quality loss
Phi-3-mini-4k-instruct-Q8_0.gguf	Q8_0	8	4.06 GB	very large, extremely low quality loss - not recommended
Phi-3-mini-4k-instruct-f16.gguf	f16	16	7.64 GB

Quantized with llama.cpp b3463.