mradermacher/granite-speech-4.1-2b-plus-GGUF

Same weights, converted (fewer layers) vs ibm-granite/granite-speech-4.1-2b-plus. mradermacher/granite-speech-4.1-2b-plus-GGUF matches part of ibm-granite/granite-speech-4.1-2b-plus (362 of its 953 tensors; the rest of ibm-granite/granite-speech-4.1-2b-plus has no counterpart). On the shared part: Same weights stored in another dtype (BF16/I64 → F16/F32).

Tensors changed: Embeddings 0/1, Norms 0/81, MLP 0/120, Attention 0/160

On the Hub · Downloads on Model Pulse