Indexnusrefather/Nyx-RP-9B-Instruct-2608-v1
LoRA-style (fewer layers) vs Qwen/Qwen3.5-9B. Indexnusrefather/Nyx-RP-9B-Instruct-2608-v1 matches part of Qwen/Qwen3.5-9B (442 of its 775 tensors; the rest of Qwen/Qwen3.5-9B 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: lm_head 0/1, Embeddings 0/2, MLP 96/99, Attention 32/228, other 0/1, Norms 0/111
4 derivatives checked
- mradermacher/Nyx-RP-9B-Instruct-2608-v1-i1-GGUF: Quantization
- mradermacher/Nyx-RP-9B-Instruct-2608-v1-GGUF: Same weights, converted
- Atomheart-Father/Nyx-RP-9B-Instruct-2608-v1-MLX-6bit: Quantization
- Muyuxiao/Nyx-RP-9B-Instruct-2608-v1-OBLITERATED: Partial fine-tune