HF RL Explorer

Add NVIDIA Canary 1B v2 ONNX speech recognition engine

Add NVIDIA Canary 1B v2 ONNX speech recognition engine: a task in LegoFlow-SWE (Harbor dataset). We need a new ONNX-based engine in transcribe-rs that supports the NVIDIA Canary 1B v2 model. This is an encoder-decoder architecture with autoregressive KV-cache decoding, capable of transcribing…

The task

We need a new ONNX-based engine in transcribe-rs that supports the NVIDIA Canary 1B v2 model. This is an encoder-decoder architecture with autoregressive KV-cache decoding, capable of transcribing speech and translating to English across 27 languages.

Part of Lego-X/LegoFlow-SWE.