ibm-granite/granite-speech-3.3-8b
Same weights, repacked (fewer layers) vs ibm-granite/granite-3.3-8b-instruct. ibm-granite/granite-3.3-8b-instruct matches part of ibm-granite/granite-speech-3.3-8b (361 of its 953 tensors; the rest of ibm-granite/granite-speech-3.3-8b has no counterpart). On the shared part: Same weights in different files (re-saved, re-sharded or new metadata), plus 591 new tensors.
Tensors changed: Norms 0/81, MLP 0/120, Attention 0/160
3 derivatives checked
- mradermacher/granite-speech-3.3-8b-i1-GGUF: Quantization (fewer layers)
- mradermacher/granite-speech-3.3-8b-GGUF: Same weights, converted (fewer layers)
- cstr/granite-speech-3.3-8b-GGUF: Not comparable yet