metadata

base_model: HuggingFaceTB/SmolLM2-1.7B-Instruct
license: apache-2.0
library_name: transformers
model_creator: HuggingFaceTB
model_name: SmolLM2-1.7B-Instruct
quantized_by: Second State Inc.
language:
  - en

SmolLM2-1.7B-Instruct-GGUF

Original Model

HuggingFaceTB/SmolLM2-1.7B-Instruct

Run with LlamaEdge

LlamaEdge version: v0.14.15 and above

Prompt template

Prompt type: chatml

Prompt string

<|im_start|>system
{system_message}<|im_end|>
<|im_start|>user
{prompt}<|im_end|>
<|im_start|>assistant

Context size: 2048

Run as LlamaEdge service

wasmedge --dir .:. --nn-preload default:GGML:AUTO:SmolLM2-1.7B-Instruct-Q5_K_M.gguf \
  llama-api-server.wasm \
  --model-name SmolLM2-1.7B-Instruct \
  --prompt-template chatml \
  --ctx-size 2048

Run as LlamaEdge command app

wasmedge --dir .:. --nn-preload default:GGML:AUTO:SmolLM2-1.7B-Instruct-Q5_K_M.gguf \
  llama-chat.wasm \
  --prompt-template chatml \
  --ctx-size 2048

Quantized GGUF Models

Name	Quant method	Bits	Size	Use case
SmolLM2-1.7B-Instruct-Q2_K.gguf	Q2_K	2	675 MB	smallest, significant quality loss - not recommended for most purposes
SmolLM2-1.7B-Instruct-Q3_K_L.gguf	Q3_K_L	3	933 MB	small, substantial quality loss
SmolLM2-1.7B-Instruct-Q3_K_M.gguf	Q3_K_M	3	860 MB	very small, high quality loss
SmolLM2-1.7B-Instruct-Q3_K_S.gguf	Q3_K_S	3	777 MB	very small, high quality loss
SmolLM2-1.7B-Instruct-Q4_0.gguf	Q4_0	4	991 MB	legacy; small, very high quality loss - prefer using Q3_K_M
SmolLM2-1.7B-Instruct-Q4_K_M.gguf	Q4_K_M	4	1.06 GB	medium, balanced quality - recommended
SmolLM2-1.7B-Instruct-Q4_K_S.gguf	Q4_K_S	4	999 MB	small, greater quality loss
SmolLM2-1.7B-Instruct-Q5_0.gguf	Q5_0	5	1.19 GB	legacy; medium, balanced quality - prefer using Q4_K_M
SmolLM2-1.7B-Instruct-Q5_K_M.gguf	Q5_K_M	5	1.23 GB	large, very low quality loss - recommended
SmolLM2-1.7B-Instruct-Q5_K_S.gguf	Q5_K_S	5	1.19 GB	large, low quality loss - recommended
SmolLM2-1.7B-Instruct-Q6_K.gguf	Q6_K	6	1.41 GB	very large, extremely low quality loss
SmolLM2-1.7B-Instruct-Q8_0.gguf	Q8_0	8	1.82 GB	very large, extremely low quality loss - not recommended
SmolLM2-1.7B-Instruct-f16.gguf	f16	16	3.42 GB

Quantized with llama.cpp b4120