#audio-llm

Why We Moved an AudioLLM to Megatron

Scaling a 10B multimodal model beyond FSDP with tensor, pipeline, and context parallelism