This model requires you to accept the CC BY 4.0 license terms before download.
Description
Canary-Qwen 2.5B — NVIDIA speech-augmented language model pairing the Canary (FastConformer) audio encoder with a Qwen LLM for English speech recognition plus LLM-style summarization and question-answering over transcribed audio. CC-BY-4.0.
Tags
At a glance
| Kind | audio language model |
|---|---|
| Precision | bfloat16 |
| Container size | 4.7 GiB |
| Weights inside | 4.7 GiB |
| Components | 3 |
| Licence | CC BY 4.0 |
What is inside the container
A .nbx container packs every part of canary-qwen-2.5b — 3 components with weights, graphs and runtime settings — so one file is all the runtime needs.
| Component | Blocks | Layers | Weights |
|---|---|---|---|
| embed_tokens | 0 | — | 594 MiB |
| llm | 0 | — | 2.6 GiB |
| perception | 0 | — | 1.5 GiB |
Default settings
The values the container ships with. They apply when you do not pass your own.
- Temperature
- 0
Memory needed
The weights of canary-qwen-2.5b take 4.7 GiB in bfloat16. That is the floor for the memory the runtime has to place, spread over the devices it finds; activations come on top and grow with the size and length you ask for. A measured peak per device class is not published for this model yet.
Verify what you downloaded
Every download carries its SHA-256, and the file endpoint sends it in the X-NBX-SHA256 header. The container is 4.7 GiB (5,085,714,949 bytes).
$ sha256sum canary-qwen-2.5b.nbx
ea079a90f31d9822d4047985b9e24fa2f72da4db24523255fa5cee0137009624
Questions
What is canary-qwen-2.5b?
A audio language model by nvidia, distributed as a bfloat16 .nbx container and licensed CC BY 4.0.
How big is the download?
4.7 GiB, of which 4.7 GiB are weights.
How do I run it?
Install the runtime from the download page, then use the commands under “CLI Usage” below: neurobrix import nvidia/canary-qwen-2.5b, then neurobrix serve --model canary-qwen-2.5b.
CLI Usage
$ neurobrix import nvidia/canary-qwen-2.5b
$ neurobrix serve --model canary-qwen-2.5b
$ neurobrix run --prompt "your prompt here"
$ neurobrix stop