derek33125/PA-stage1-DeepSeekQwen7B-300
LoRA-style vs unsloth/DeepSeek-R1-Distill-Qwen-7B. 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 28 of 28 layers; norms, biases and embeddings are untouched.
Tensors changed: Embeddings 0/1, Norms 0/57, MLP 84/84, Attention 112/196, lm_head 0/1
2 derivatives checked
- mradermacher/PA-stage1-DeepSeekQwen7B-300-GGUF: Same weights, converted
- derek33125/PA-stage2-DeepSeekQwen-7B-400: LoRA-style