Fix/vllm processor cache for text only model#359

Merged

hiyouga merged 3 commits intohiyouga:mainfrom

cyc00518:fix/vllm-processor-cache-for-text-only-model

Jun 17, 2025

Commits on Jun 17, 2025

[utils] handle models that do not provide a processor by using a try/except block
cyc00518
committed
[rollout] Adjusted the vLLM rollout code so disable_mm_preprocessor_cache is only set for multimodal cases, avoiding errors with text-only models
cyc00518
committed
Update tokenizer.py
hiyouga
authored