OpenGVLab/InternVL3_5-38B
Partial fine-tune vs OpenGVLab/InternVL3_5-38B-MPO. Partial change: 84% of the weights changed (attention, lm_head, mlp, norms), in 64 of 64 layers.
Tensors changed: Embeddings 0/5, Norms 188/437, Attention 256/391, other 0/90, MLP 197/378, lm_head 1/1
5 derivatives checked
- bartowski/OpenGVLab_InternVL3_5-38B-GGUF: Same weights, converted (fewer layers)
- mradermacher/InternVL3_5-38B-GGUF: Quantization (fewer layers)
- mradermacher/InternVL3_5-38B-i1-GGUF: Quantization (fewer layers)
- cyankiwi/InternVL3_5-38B-AWQ-4bit: Quantization
- OpenGVLab/InternVL3_5-38B-Flash: Full fine-tune