DavidAU/Qwen3-53B-A3B-2507-TOTAL-RECALL-v2-MASTER-CODER
Partial fine-tune (fewer layers) vs Qwen/Qwen3-30B-A3B-Instruct-2507. Qwen/Qwen3-30B-A3B-Instruct-2507 matches part of DavidAU/Qwen3-53B-A3B-2507-TOTAL-RECALL-v2-MASTER-CODER (18867 of its 33015 tensors; the rest of DavidAU/Qwen3-53B-A3B-2507-TOTAL-RECALL-v2-MASTER-CODER has no counterpart). On the shared part: Partial change: 11% of the weights changed (attention, lm_head, mlp, norms), in 1 of 48 layers.
Tensors changed: Embeddings 0/1, Norms 5/193, MLP 30/1337, Attention 4/192, lm_head 1/1
2 derivatives checked
- mradermacher/Qwen3-53B-A3B-2507-TOTAL-RECALL-v2-MASTER-CODER-i1-GGUF: Quantization
- mradermacher/Qwen3-53B-A3B-2507-TOTAL-RECALL-v2-MASTER-CODER-GGUF: Quantization
1 other repos hold byte-for-byte identical weight files.