codefuse-ai/OpAgent-32B

Full fine-tune vs Qwen/Qwen3-VL-32B-Thinking. Fine-tune: 95% of the weights changed, across 64 of 64 layers.

Tensors changed: Embeddings 3/4, Norms 239/373, MLP 295/316, Attention 364/364, lm_head 0/1

2 derivatives checked

• mradermacher/OpAgent-32B-i1-GGUF: Quantization (fewer layers)
• mradermacher/OpAgent-32B-GGUF: Quantization (fewer layers)

On the Hub · Downloads on Model Pulse