i can run thin models for DGX spark
#1
by Shaoqing - opened
I encountered this error: packages/transformers/models/auto/tokenization_auto.py", line 1153, in from_pretrained (APIServer pid=1) raise ValueError( (APIServer pid=1) ValueError: Tokenizer class TokenizersBackend does not exist or is not currently imported. I tried using --tokenizer Qwen/Qwen3-32-Instruct (or your specific model path), but it didn't work.
I've updated tokenizer_config.json — the tokenizer_class was incorrectly set to "TokenizersBackend" instead of "Qwen2Tokenizer". This was a bug from the modelopt export process.
Please re-download the model (or just pull the updated tokenizer_config.json). The --tokenizer workaround should no longer be needed.
If you still hit issues, make sure you're using:
- transformers >= 5.3.0 (from source: pip install git+https://github.com/huggingface/transformers.git) — required for qwen3_5 model type support
- vllm == 0.17.1
- The vision encoder patch described in the README
berkerdooo changed discussion status to closed