marcorez8/acestep-v15-xl-turbo-bf16

Same weights, converted vs ACE-Step/acestep-v15-xl-turbo. Same weights stored in another dtype (F32 → BF16).

Tensors changed: Norms 0/293, Attention 0/327, MLP 0/138, other 0/49, Embeddings 0/22

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

On the Hub · Downloads on Model Pulse