OpenGVLab/InternVL2_5-38B
Full fine-tune (fewer layers) vs OpenGVLab/InternViT-6B-448px-V2_5. OpenGVLab/InternViT-6B-448px-V2_5 matches part of OpenGVLab/InternVL2_5-38B (589 of its 1366 tensors; the rest of OpenGVLab/InternVL2_5-38B has no counterpart). On the shared part: Fine-tune: 100% of the weights changed, across 45 of 45 layers.
Tensors changed: Embeddings 4/4, Norms 180/180, Attention 135/135, other 90/90, MLP 180/180
4 derivatives checked
- mradermacher/InternVL2_5-38B-i1-GGUF: Quantization (fewer layers)
- mradermacher/InternVL2_5-38B-GGUF: Quantization (fewer layers)
- OpenGVLab/InternVL2_5-38B-MPO: Full fine-tune
- Skywork/SkyworkVL-38B: Partial fine-tune
1 other repos hold byte-for-byte identical weight files.