transformers

11e03b0b - Add a fine-grained quantizer that hosts one quantizer per module group

Add a fine-grained quantizer that hosts one quantizer per module group FineGrainedConfig maps module patterns to the configs of registered quantization methods. FineGrainedHfQuantizer builds one member quantizer per config. A hook on one tensor goes to the member that owns its module. A hook on the whole model runs on every member, each limited to its own modules. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>