OpenGVLab/InternVL2-8B
Full fine-tune (fewer layers) vs OpenGVLab/InternViT-300M-448px. OpenGVLab/InternViT-300M-448px matches part of OpenGVLab/InternVL2-8B (340 of its 573 tensors; the rest of OpenGVLab/InternVL2-8B has no counterpart). On the shared part: Fine-tune: 100% of the weights changed, across 24 of 24 layers.
Tensors changed: Embeddings 4/4, Attention 96/96, other 48/48, MLP 96/96, Norms 94/96
3 derivatives checked
- OpenGVLab/InternVL2-8B-MPO: Full fine-tune
- MathLLMs/MathCoder-VL-8B: Full fine-tune
- Virtue-AI-HUB/SafeWatch-8B: Full fine-tune
5 other repos hold byte-for-byte identical weight files.