tokenizer component mismatch and w1_weight_scale_2 must match w3_weight_scale_2. Accuracy may be affected issue
#5
by mtcl - opened
Hey, can you take a look at this?
[2026-04-13 14:20:19 TP0] Detected nvfp4 checkpoint. Please note that the format is experimental and subject to change.
[2026-04-13 14:20:20] Fixing v5 tokenizer component mismatch for /models/lukealonso/MiniMax-M2.7-NVFP4: pre_tokenizer ByteLevel -> Sequence, decoder ByteLevel -> ByteLevel
[2026-04-13 14:20:20] Using default HuggingFace chat template with detected content format: openai
Multi-thread loading shards: 0% Completed | 0/26 [00:00<?, ?it/s][2026-04-13 14:20:22] Fixing v5 tokenizer component mismatch for /models/lukealonso/MiniMax-M2.7-NVFP4: pre_tokenizer ByteLevel -> Sequence, decoder ByteLevel -> ByteLevel
Multi-thread loading shards: 100% Completed | 26/26 [00:38<00:00, 1.49s/it]
[2026-04-13 14:20:59 TP0] w1_weight_scale_2 must match w3_weight_scale_2. Accuracy may be affected.
[2026-04-13 14:20:59 TP1] w1_weight_scale_2 must match w3_weight_scale_2. Accuracy may be affected.