kristaller486/sokol-1.5b-02
LoRA-style vs tiiuae/Falcon-H1-1.5B-Deep-Base. 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 66 of 66 layers; norms, biases and embeddings are untouched.
Tensors changed: lm_head 0/1, Embeddings 0/1, Norms 0/199, MLP 198/198, other 0/462, Attention 264/264
2 derivatives checked
- mradermacher/sokol-1.5b-02-i1-GGUF: Quantization
- mradermacher/sokol-1.5b-02-GGUF: Same weights, converted