Add a LogitProcessor abstract interface (process(float* logits, int32_t vocab_size)) and wire a processor chain into TextTokenGenerator::generate() between step() and logits_to_token(). No behavior change when no processors registered.
cc @larryliu0820 @mergennachin @cccclai @helunwencser @jackzhxng
Add a LogitProcessor abstract interface (process(float* logits, int32_t vocab_size)) and wire a processor chain into TextTokenGenerator::generate() between step() and logits_to_token(). No behavior change when no processors registered.
cc @larryliu0820 @mergennachin @cccclai @helunwencser @jackzhxng