d0xin/Swift-1.5-Qwen3.8-Flash-Next-NVFP4-FP8PLE-Rank2-Abliterated

Partial fine-tune vs d0xin/Swift-1.5-Qwen3.8-Flash-Next-NVFP4-FP8PLE. Partial change: 5% of the weights changed (attention, embeddings, mlp, other), in 48 of 48 layers.

Tensors changed: MLP 417/1943, Attention 49/608, Norms 0/302, Embeddings 1/138, other 1/8, lm_head 0/1

On the Hub · Downloads on Model Pulse