d3LLM/d3LLM_LLaDA
LoRA-style vs GSAI-ML/LLaDA-8B-Instruct. Looks like a merged LoRA (or a fine-tune with frozen norms): only the weight matrices of k_proj, q_proj, up_proj, v_proj changed, in 32 of 32 layers; norms, biases and embeddings are untouched.
Tensors changed: Attention 96/128, Norms 0/65, Embeddings 0/1, other 0/65, MLP 32/32
2 derivatives checked
- mradermacher/d3LLM_LLaDA-GGUF: Same weights, converted
- mradermacher/d3LLM_LLaDA-i1-GGUF: Quantization