JamieBradfield/qwen3.8-9b-hermes-fc-clean-restraint
LoRA-style (fewer layers) vs empero-ai/Qwen3.8-9B-Distill. JamieBradfield/qwen3.8-9b-hermes-fc-clean-restraint matches part of empero-ai/Qwen3.8-9B-Distill (440 of its 775 tensors; the rest of empero-ai/Qwen3.8-9B-Distill 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, k_proj, o_proj, q_proj, up_proj, v_proj changed, in 31 of 32 layers; norms, biases and embeddings are untouched.
Tensors changed: Norms 0/111, Attention 32/228, MLP 96/99, other 0/1, Embeddings 0/1
2 derivatives checked
- mradermacher/qwen3.8-9b-hermes-fc-clean-restraint-GGUF: Same weights, converted
- JamieBradfield/qwen3.8-9b-hermes-fc-clean-restraint-GGUF: Same weights, converted