ssurface/qwen3-4b-cot-compress-l1
Text Generation • Updated • 17
Qwen3-4B CoT Compression Study
LoRA adapters trained for 5 progressively shorter chain-of-thought styles on GSM8K, plus the eval artifacts behind the Pareto curve.
-
-
ssurface/qwen3-4b-cot-compress-l2
Text Generation • Updated • 5 -
ssurface/qwen3-4b-cot-compress-l3
Text Generation • Updated • 5 -
ssurface/qwen3-4b-cot-compress-l4
Text Generation • Updated • 4 -
ssurface/qwen3-4b-cot-compress-l5
Text Generation • Updated • 6 -
ssurface/qwen3-4b-cot-compress-eval
Updated • 15