aisingapore/SEA-LION-v1-7B-IT-Research
LoRA-style vs aisingapore/SEA-LION-v1-7B. Looks like a merged LoRA (or a fine-tune with frozen norms): only the weight matrices of Wqkv, down_proj, out_proj, up_proj changed, in 32 of 32 layers; norms, biases and embeddings are untouched.
Tensors changed: Attention 64/256, MLP 64/128, Norms 0/130, Embeddings 0/2
2 derivatives checked
- mradermacher/sea-lion-7b-instruct-research-i1-GGUF: Quantization
- mradermacher/sea-lion-7b-instruct-research-GGUF: Same weights, converted