HumeAI/tada-3b-ml
Full fine-tune (fewer layers) vs meta-llama/Llama-3.2-3B. meta-llama/Llama-3.2-3B matches part of HumeAI/tada-3b-ml (254 of its 502 tensors; the rest of HumeAI/tada-3b-ml has no counterpart). On the shared part: Fine-tune: 100% of the weights changed, across 28 of 28 layers.
Tensors changed: Embeddings 1/1, Norms 57/57, MLP 84/84, Attention 112/112
1 derivatives checked
- cstr/tada-tts-3b-ml-GGUF: Not comparable yet
1 other repos hold byte-for-byte identical weight files.