TinySwallow-1.5B
Sakana AI's 1.5B Japanese compact model built with TAID distillation.
About
TinySwallow-1.5B is a Japanese compact language model from Sakana AI and the Swallow Team (Tokyo Tech), released in January 2025 under Apache 2.0. It was created with TAID (Temporally Adaptive Interpolated Distillation), a new knowledge-distillation method that interpolates teacher and student logits over training time, using Qwen2.5-32B-Instruct as teacher and Qwen2.5-1.5B-Instruct as student, followed by further pretraining on Japanese text.
The base model carries Apache 2.0, though the instruct variant's card notes it was trained on Gemma-family data, so Gemma Terms of Use also apply there (flagged for compliance). An instruct version, TinySwallow-1.5B-Instruct, is available for chat use, with GGUF and MLC conversions for local deployment.
Specs
Context window
Pricing
Free quota
API access
Weights
Released
Benchmarks
Provider-reported unless a source says otherwise. Benchmarks are marketing-adjacent by nature — treat them as directional.
Use cases
Trust score — 8/10
Scored 8/10 (verified tier, metrics-v1). Evidence: Verified via vendor site/docs/GitHub and reputable coverage, wave 135; Trending score 78/100 from researcher assessment.