prithivMLmods/Lumian-VLR-7B-Thinking
Partial fine-tune vs Qwen/Qwen2.5-VL-7B-Instruct. Partial change: 85% of the weights changed (attention, lm_head, mlp, norms), in 28 of 32 layers.
Tensors changed: Embeddings 0/2, Norms 53/122, MLP 84/280, Attention 196/324, lm_head 1/1
3 derivatives checked
- mradermacher/Lumian-VLR-7B-Thinking-i1-GGUF: Quantization (fewer layers)
- mradermacher/Lumian-VLR-7B-Thinking-GGUF: Same weights, converted (fewer layers)
- prithivMLmods/Lumian2-VLR-7B-Thinking: LoRA-style