DQN-Labs/dqnMath-v0.2-3.8B-HF
LoRA-style vs microsoft/Phi-4-mini-reasoning. Looks like a merged LoRA (or a fine-tune with frozen norms): only the weight matrices of down_proj, gate_up_proj, o_proj, qkv_proj changed, in 8 of 32 layers; norms, biases and embeddings are untouched.
Tensors changed: Embeddings 0/1, Norms 0/65, MLP 16/64, Attention 16/64
1 derivatives checked
- Flexan/DQN-Labs-dqnMath-v0.2-3.8B-HF-GGUF: Same weights, converted