empero-ai/Qwable-9B-Claude-Fable-5
Partial fine-tune vs Qwen/Qwen3.5-9B. Partial change: 84% of the weights changed (attention, lm_head, mlp, norms), in 32 of 32 layers.
Tensors changed: lm_head 1/1, Embeddings 0/4, MLP 96/208, Attention 176/332, Norms 58/215
9 derivatives checked
- empero-ai/Qwable-9B-Claude-Fable-5-GGUF: Same weights, converted
- mradermacher/Qwable-9B-Claude-Fable-5-i1-GGUF: Quantization
- mradermacher/Qwable-9B-Claude-Fable-5-GGUF: Same weights, converted
- Atomic-Germ/Qwable-Claude-Fable-5-Distilled-NPU2: Quantization
- wepiqx/Qwable-9B-Claude-Fable-5-SHQ8-GGUF: Quantization
- usermma/Qwable-9B-Claude-Fable-5-mlx-8Bit: Quantization
- Dzluck/Qwable-9B-Claude-Fable-5-GGUF: Same weights, converted
- pestlee/Qwable-9B-Claude-Fable-5-StraTA: LoRA-style (fewer layers)
- usermma/Qwable-9B-Claude-Fable-5-OBLITERATED: Partial fine-tune (fewer layers)
2 other repos hold byte-for-byte identical weight files.