espnet2.speechlm.model.speechlm.lm.parallel.build_parallel_hf_class

source

espnet2.speechlm.model.speechlm.lm.parallel.build_parallel_hf_class(model_hf_tag)

Dynamically create a parallel LLM class based on HuggingFace architecture.

Creates a subclass of the original HF model with added multimodal support, parallel stream processing, and custom embedding/loss computation.

• Parameters:model_hf_tag – HuggingFace model identifier to determine base architecture
• Returns: ParallelLLM class inheriting from the original HF architecture