ai-agents-vocab-size-mismatch-decode-error-nonetype-object-cannot-be-converted-t-8dea978d

THEN Set the model's `vocab_size` to match the actual tokenizer vocabulary length. For vLLM, modify the model file (e.g., `opt.py`) to pass `len(tokenizer)` instead of `config.vocab_size` to the sampler, or directly edit the `config.json` of the cached model to reduce `vocab_size` to the tokenizer's length (e.g., for `facebook/opt-125m`, change from 50272 to 50265).

In vLLM's OPT model file, change:
self.sampler = Sampler(config.vocab_size)
to:
self.sampler = Sampler(len(tokenizer))  # or a fixed correct vocab_size