transformers

b4526a76 - Fix DataCollatorForLanguageModeling ignoring seed=0 (#49349)

Fix DataCollatorForLanguageModeling ignoring seed=0 (#49349) The seed check used truthiness (`if self.seed`), so `seed=0` was treated as "no seed" and masking fell back to the global RNG. Check `self.seed is not None` instead, in both the torch and numpy paths. Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>