Files
llama.cpp/conversion
Yaniss 189b67fc2c convert: accept the official Puzzle BF16 checkpoint's tensor naming
The officially distributed BF16 checkpoint (NVIDIA-Nemotron-Labs-3-Puzzle-
75B-A9B-BF16) names the trunk model.* (model.layers.*, model.embeddings,
model.norm_f) where the original release used the NemotronH-style
backbone.*, and spells the router bias e_score_correction_bias instead of
e_score_correction.bias. Normalize both at the top of
NemotronHPuzzleModel.modify_tensors so either checkpoint converts; every
tensor name in the official index (42683 keys, MTP head included) resolves
through the tensor map after normalization.
2026-07-26 11:30:37 +02:00
..
2026-06-29 16:58:51 +08:00
2026-06-23 12:03:31 +02:00
2026-05-15 15:18:12 +02:00