aldair166/qwen3.5-2.8b
Partial fine-tune (fewer layers) vs Qwen/Qwen3.5-4B. aldair166/qwen3.5-2.8b matches part of Qwen/Qwen3.5-4B (243 of its 738 tensors; the rest of Qwen/Qwen3.5-4B has no counterpart). On the shared part: Partial change: 44% of the weights changed (attention, mlp, norms), in 12 of 20 layers.
Tensors changed: Embeddings 0/1, MLP 36/60, Attention 64/120, Norms 35/62
2 derivatives checked
- aldair166/qwen3.5-2.8b-qlora: Adapter
- aldair166/Qwen3.5-2.8B-V0.1: Quantization