tabularisai/Qwen3-0.3B-distil

Full fine-tune (fewer layers) vs Qwen/Qwen3-0.6B. tabularisai/Qwen3-0.3B-distil matches part of Qwen/Qwen3-0.6B (156 of its 311 tensors; the rest of Qwen/Qwen3-0.6B has no counterpart). On the shared part: Fine-tune: 100% of the weights changed, across 14 of 14 layers.

Tensors changed: Embeddings 1/1, Norms 54/57, MLP 42/42, Attention 56/56

On the Hub · Downloads on Model Pulse