SmolLM2 360M
A small language model with 360M parameters released by Hugging Face, part of the SmolLM2 series, aimed at efficient edge inference scenarios.
Timeline (last 90 days)
With no_repeat_ngram_size: 3 enabled alone, Llama 3.2 1B's JSON pass rate was 1/24 (4%) and SmolLM2 360M's was 0/24 (0%), making it the main cause of the JSON output collapse.
Llama 3.2 1B模型在WebGPU下达到55-64 tokens/s，中位回复耗时0.4-0.5秒，比SmolLM2 360M更快。
SmolLM2 360M fine-tuned with LoRA achieves an F0.5 of 0.663 on the grammar correction task, with a 20.1% false-edit rate.
All Facts (3)
With no_repeat_ngram_size: 3 enabled alone, Llama 3.2 1B's JSON pass rate was 1/24 (4%) and SmolLM2 360M's was 0/24 (0%), making it the main cause of the JSON output collapse.
50%UnverifiedLlama 3.2 1B模型在WebGPU下达到55-64 tokens/s，中位回复耗时0.4-0.5秒，比SmolLM2 360M更快。
50%UnverifiedSmolLM2 360M fine-tuned with LoRA achieves an F0.5 of 0.663 on the grammar correction task, with a 20.1% false-edit rate.
50%