tvall43/Qwen3.5-22B-A3B-heretic-reap
LoRA-style (fewer layers) vs llmfan46/Qwen3.5-35B-A3B-ultra-uncensored-heretic. tvall43/Qwen3.5-22B-A3B-heretic-reap matches part of llmfan46/Qwen3.5-35B-A3B-ultra-uncensored-heretic (19413 of its 31666 tensors; the rest of llmfan46/Qwen3.5-35B-A3B-ultra-uncensored-heretic has no counterpart). On the shared part: Looks like a merged LoRA (or a fine-tune with frozen norms): only the weight matrices of down_proj, gate_proj, up_proj changed, in 40 of 40 layers; norms, biases and embeddings are untouched.
Tensors changed: lm_head 0/1, Embeddings 0/1, Norms 0/131, Attention 0/280, MLP 1230/1414
3 derivatives checked
- mradermacher/Qwen3.5-22B-A3B-heretic-reap-i1-GGUF: Quantization
- mradermacher/Qwen3.5-22B-A3B-heretic-reap-GGUF: Quantization
- tvall43/Qwen3.5-22B-A3B-heretic-reap-gguf: Same weights, converted