tokyotech-llm/Qwen3-Swallow-30B-A3B-SFT-v0.2
Full fine-tune vs tokyotech-llm/Qwen3-Swallow-30B-A3B-CPT-v0.2. Fine-tune: 100% of the weights changed, across 48 of 48 layers.
Tensors changed: Embeddings 1/1, Norms 193/193, MLP 2613/2613, Attention 192/192, lm_head 1/1
2 derivatives checked
- mmnga-o/Qwen3-Swallow-30B-A3B-SFT-v0.2-gguf: Quantization
- tokyotech-llm/Qwen3-Swallow-30B-A3B-RL-v0.2: Full fine-tune