SmolLM2
A series of small open-source language models released by Hugging Face, characterized by compact parameter counts while retaining a degree of instruction-following capability.
Timeline (last 90 days)
在Chrome 155、Windows 11、RTX 4070环境下，360M参数的SmolLM2模型在WebGPU后端达到62.9 tokens/s，WASM后端仅1.6 tokens/s。
BEINGS的冷启动加载时间分别为7.5秒（SmolLM2 273MB）、13.4秒（Qwen 483MB）、26.8秒（Llama 1238MB），缓存后降至1.0、2.2、3.1秒。
WebGPU上加载q4f16模型文件（SmolLM2为272.7MB），WASM上加载q4模型文件（SmolLM2为387.9MB）。
Candidate small models include SmolLM2 (135M and 360M), Gemma 3 270M, LFM2.5-350M, Qwen2.5-0.5B, and Qwen3-0.6B, all of which have ONNX browser builds.
Miso supports embedding the local small models SmolLM2 or Qwen2.5, and can be replaced with any compatible .gguf model
SmolLM2 is a series of ultra-small parameter language models released by HuggingFace in 2024, available in three sizes—135M, 360M, and 1.7B—designed specifically for edge devices and on-device inference
All Facts (6)
在Chrome 155、Windows 11、RTX 4070环境下，360M参数的SmolLM2模型在WebGPU后端达到62.9 tokens/s，WASM后端仅1.6 tokens/s。
50%UnverifiedBEINGS的冷启动加载时间分别为7.5秒（SmolLM2 273MB）、13.4秒（Qwen 483MB）、26.8秒（Llama 1238MB），缓存后降至1.0、2.2、3.1秒。
50%UnverifiedWebGPU上加载q4f16模型文件（SmolLM2为272.7MB），WASM上加载q4模型文件（SmolLM2为387.9MB）。
50%UnverifiedCandidate small models include SmolLM2 (135M and 360M), Gemma 3 270M, LFM2.5-350M, Qwen2.5-0.5B, and Qwen3-0.6B, all of which have ONNX browser builds.
50%UnverifiedMiso supports embedding the local small models SmolLM2 or Qwen2.5, and can be replaced with any compatible .gguf model
50%UnverifiedSmolLM2 is a series of ultra-small parameter language models released by HuggingFace in 2024, available in three sizes—135M, 360M, and 1.7B—designed specifically for edge devices and on-device inference
50%