transformers

8464284d - chore(typing): Add type checking to `src/transformers/quantizers` (#44412)

chore(typing): Add type checking to `src/transformers/quantizers` (#44412) * chore(typing): add `ty` type checking for src/transformers/generation Add type declarations for mixin host-class attributes on GenerationMixin, class-level annotations for dynamically-set attributes on GenerationConfig, and fix minor typing issues in candidate_generator, watermarking, and stopping_criteria. Create _typing.py Protocol for documentation/reuse. * extend make file * chore(typing): further reduce ty diagnostics (36→18), replace assert with if guards - Replace assert narrowing with if guards (candidate_generator.py) - Add if-guard for model_kwargs None in _prepare_model_inputs and _maybe_initialize_input_ids_for_generation - Add if-guard for encoder_input_ids None checks - Fix save_directory.split() for PathLike by wrapping with str() - Fix tuple[Tensor] reassignment in compute_transition_scores - Add type annotation for self.model in WhisperTimeStampLogitsProcessor - Add type annotation for self.sequence_bias in SequenceBiasLogitsProcessor - Add no-assert rule to AGENTS.md typing strategy * more tweaking * fix get word size * date * cast value * @zucchini-nlp reviews * model_kwargs cannot be None, fix type annotations * removed redundancy * bypass type checking to reduce code complexity * de-vectorize * unified _typing * explain why we pollute inheritance with a protocol * do not import torch for type checking, it breaks on CI logging usage * Add check_types.py to wrap ty type checking * improved ty usage * chore(typing): Add type checking to src/transformers/quantizers * fix merge error * reworked the typing using previous PR best practices * format and torch.neuron fix * revert bad upstream merge * narrow isinstance * cleaner const, also 1.23.99 vs 1.24.0 * clean up old cast * removing leftover * lets just use setattr * tweaked one more optimum version * setattr again * more setattr/getattr patyterns * more cleanups * fmt * cleanups from rebase abd ty 0.20.0 * fix ignores * const for torchao version * simplify call Co-authored-by: vasqu * use is_torch_greater_or_equal * use type guard * remove the getattr() pattenr * reworked the approach * Update src/transformers/quantizers/quantizer_sinq.py Co-authored-by: Marc Sun <57196510+SunMarc@users.noreply.github.com> * Update src/transformers/quantizers/quantizer_quanto.py Co-authored-by: Marc Sun <57196510+SunMarc@users.noreply.github.com> * class is there we can use it --------- Co-authored-by: Marc Sun <57196510+SunMarc@users.noreply.github.com>