OpenGVLab/InternVL3-78B-Instruct
Full fine-tune vs OpenGVLab/InternVL3-78B-Pretrained. Fine-tune: 100% of the weights changed, across 80 of 80 layers.
Tensors changed: Embeddings 5/5, Norms 340/341, Attention 695/695, other 90/90, MLP 426/426, lm_head 1/1
6 derivatives checked
- OpenGVLab/InternVL3-78B: Full fine-tune
- unsloth/InternVL3-78B-Instruct-GGUF: Same weights, converted (fewer layers)
- mradermacher/InternVL3-78B-Instruct-i1-GGUF: Quantization (fewer layers)
- mradermacher/InternVL3-78B-Instruct-GGUF: Quantization (fewer layers)
- OpenGVLab/InternVL3-78B-hf: Full fine-tune
- AbteeXAILab/lumynax-multimodal-internvl3-78b-instruct: Identical copy
2 other repos hold byte-for-byte identical weight files.