unsloth/Qwen3-4B-Instruct-2507-FP8

Same weights, converted vs Qwen/Qwen3-4B-Instruct-2507-FP8. Same weights stored in another dtype (F8_E4M3/F16 → F8_E4M3/BF16).

Tensors changed: Embeddings 0/1, Norms 0/145, MLP 0/216, Attention 0/288

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

On the Hub · Downloads on Model Pulse