trl-lab/qwen3.5-2b-grpo-sqale

Partial fine-tune vs Qwen/Qwen3.5-2B. Partial change: 83% of the weights changed (attention, embeddings, mlp, norms), in 24 of 24 layers.

Tensors changed: Attention 132/268, Norms 37/183, Embeddings 1/5, MLP 72/175, other 0/1

1 derivatives checked

• mradermacher/qwen3.5-2b-grpo-sqale-GGUF: Same weights, converted

On the Hub · Downloads on Model Pulse