Sakalti/SJT-990M

Same weights, converted vs sail/Sailor2-1B. Same weights stored in another dtype (BF16 → F16).

Tensors changed: lm_head 0/1, Embeddings 0/1, Norms 0/97, MLP 0/144, Attention 0/336

1 derivatives checked

• mradermacher/SJT-990M-i1-GGUF: Quantization

1 other repos hold byte-for-byte identical weight files.

On the Hub · Downloads on Model Pulse