Japan-AI-Consulting/AKT-caption-v0.1-lfm2-350m-beta1
LoRA-style vs unsloth/LFM2-350M. Looks like a merged LoRA (or a fine-tune with frozen norms): only the weight matrices of k_proj, out_proj, q_proj, v_proj, w1, w2, w3 changed, in 16 of 16 layers; norms, biases and embeddings are untouched.
Tensors changed: Embeddings 0/2, other 0/30, MLP 48/48, Norms 0/44, Attention 24/24
1 derivatives checked
- Japan-AI-Consulting/AKT-caption-v0.1-lfm2-350m-gguf: Quantization