AudioVisual-Caption/ASID-Captioner-7B
Partial fine-tune (fewer layers) vs Qwen/Qwen2.5-Omni-7B. AudioVisual-Caption/ASID-Captioner-7B matches part of Qwen/Qwen2.5-Omni-7B (1346 of its 2448 tensors; the rest of Qwen/Qwen2.5-Omni-7B has no counterpart). On the shared part: Partial change: 86% of the weights changed (attention, embeddings, lm_head, mlp, norms, other), in 28 of 32 layers.
Tensors changed: other 2/7, MLP 88/408, Norms 56/252, Attention 196/676, Embeddings 1/2, lm_head 1/1
2 derivatives checked
- mradermacher/ASID-Captioner-7B-GGUF: Same weights, converted (fewer layers)
- mradermacher/ASID-Captioner-7B-i1-GGUF: Quantization (fewer layers)