SirSahOl/Tev1-0.8B-experimental-chat-mlx-16bit
Partial fine-tune (fewer layers) vs togethercomputer/Tev1-0.8B-experimental. SirSahOl/Tev1-0.8B-experimental-chat-mlx-16bit matches part of togethercomputer/Tev1-0.8B-experimental (302 of its 488 tensors; the rest of togethercomputer/Tev1-0.8B-experimental has no counterpart). On the shared part: Partial change: 0% of the weights changed (norms), in 24 of 24 layers.
Tensors changed: Attention 0/150, Norms 61/79, Embeddings 0/1, MLP 0/72
1 other repos hold byte-for-byte identical weight files.