Trelis/Llama-3.2-1B-Instruct-MATH-synthetic
LoRA-style vs unsloth/Llama-3.2-1B-Instruct. 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 16 of 16 layers; norms, biases and embeddings are untouched.
Tensors changed: Embeddings 0/1, Norms 0/33, MLP 48/48, Attention 64/64
2 derivatives checked
- featherless-ai-quants/Trelis-Llama-3.2-1B-Instruct-MATH-synthetic-GGUF: Quantization
- Trelis/Llama-3.2-1B-Instruct-MATH: LoRA-style