canary model family
Tasks
Supported architectures and task heads
The Hugging Face values below are copied from checkpoint metadata at the recorded revision. The TRTMC task contract comes from the exact E2E recipe. Architecture identity selects or describes a source graph; it does not imply that TRTMC reproduces every Hugging Face head with the same model_type. For example, an encoder recipe may consume the base model and intentionally return hidden states instead of the checkpoint's pretraining or classification logits.
HF model_type | HF architecture / pipeline class | TRTMC task contract | Exact recipe profiles |
|---|---|---|---|
| — | nemo.collections.asr.models.aed_multitask_models.EncDecMultiTaskModelSource: model_config.targetMetadata: canary-1b-v2.nemo:model_config.yaml at 87bc52657add | speech_to_text | canary-1b-v2canary-1b-v2-tp4 |
Declared recipes
Each row comes from a model-owned E2E manifest. It declares a test recipe; it is not a live pass receipt for every hardware target.
| Recipe | Exact Hugging Face checkpoint | Task | Build configuration | Declared E2E cases | Manifest |
|---|---|---|---|---|---|
canary-1b-v2 | nvidia/canary-1b-v2 | fp16; single device | canary-1b-v2, canary-1b-v2-asr-probe01, canary-1b-v2-asr-probe02, canary-1b-v2-asr-probe03, canary-1b-v2-asr-probe04, canary-1b-v2-asr-probe05, canary-1b-v2-asr-probe06, canary-1b-v2-asr-probe08 | tests/e2e/models/canary/manifests/canary-1b-v2.json | |
canary-1b-v2-tp4 | nvidia/canary-1b-v2 | family default; TP4 | canary-1b-v2-tp4 | tests/e2e/models/canary/manifests/canary-1b-v2-tp4.json |
Family-owned configuration
This family does not declare a family-owned --set namespace. Use the explicit CLI options shown below and the shared configuration namespaces documented in Configure Runtime Behavior.
Family-specific CLI contracts
Inputs and options below are filtered by the declared E2E task and the methods and configuration fields used by that family's native runtime implementation. The global CLI parser accepts a wider union of flags; flags absent here are not declared for this family.
trtmc transcribe
Transcribe an audio file.
Declared recipes: canary-1b-v2, canary-1b-v2-tp4
trtmc transcribe <bundle.bundle> --audio <input.wav>| Supported input or option | Requirement | Runtime behavior |
|---|---|---|
--audio <PATH> | Required | Input WAV file; repeat for supported offline batches. |
--max-new-tokens <N> | Optional | Set the maximum number of decoder output tokens. |
--beam-size <N> | Optional | Select greedy decoding at 1 or beam search above 1. |
--length-penalty <F> | Optional | Set the beam-search length-normalization exponent. |
--beam-fallback-max-size <N> | Optional | Retry an unterminated decode with larger beams up to N. |
--source-language <TAG> | Optional | Set the source language for a multilingual transcription request. |
--target-language <TAG> | Optional | Set the target language for a multilingual transcription request. |
--task <transcribe|translate> | Optional | Select transcription or speech translation. |
--punctuation | --no-punctuation | Optional | Enable or disable punctuation in the result. |
--timestamps | Optional | Return timestamped transcription segments. |
--max-input-seconds <F> | Optional | Reject input longer than the specified duration. |
--segment-length-seconds <F> | Optional | Split long audio into segments no longer than this duration. |
--segment-min-seconds <F> | Optional | Set the minimum duration for dynamic segmentation windows. |
--segment-overlap-seconds <F> | Optional | Set overlap between adjacent transcription segments. |
--lcs-merge | Optional | Merge overlapping segment tokens with boundary-constrained LCS. |
Code basis
Runtime provider: canary
include/trtmc/pipeline.hsrc/cli/main.cppsrc/runtime/models/canary/canary_request.hsrc/runtime/models/canary/pipeline.cpptests/e2e/models/canary/manifests/canary-1b-v2-tp4.jsontests/e2e/models/canary/manifests/canary-1b-v2.json
Bundle lifecycle commands
These commands apply to every declared recipe in this family; they do not add task inputs or model capabilities.
trtmc build
Build one exact checkpoint into a TensorRT-Model-Connect bundle.
trtmc build <hf-id> --output <bundle.bundle>trtmc inspect
Inspect bundle metadata, runtime identity, and packaged sections.
trtmc inspect <bundle.bundle>See the CLI Reference for all options and limitations.