Skip to content
English

Switch theme

// NVIDIA Parakeet / Canary speech · Speech-to-text

canary-1b-flash: VRAM & local deployment

811M parameters. About 1.0 GB at 4-bit and 2.3 GB at BF16.

released 2025-03-07licence: Commercial OKmemory figures are estimates

01Specs

PublisherNVIDIA · nvidia
Family / seriesNVIDIA Parakeet / Canary speech · canary-flash
TypeSpeech-to-text (automatic-speech-recognition)
Parameters811M Source: counted from official safetensors weights
Released weightsFP32 · 3.2 GBdownload
LicenceCC-BY-4.0 · Commercial OKPermissive licence (Apache-2.0 / MIT / CC-BY / OpenMDW): commercial use allowed.
Released2025-03-07 (Hugging Face repo created)

02How much VRAM it needs (estimate)

Speech models are small; the weight size is close to the total memory needed.

PrecisionWeightsWeights + overheadSmallest GPUMac
BF16 / FP16 (original)1.6 GB 2.3 GB 8GB GPUMac 16GB
FP8 / INT8 / Q8_00.9 GB 1.4 GB 8GB GPUMac 16GB
4-bit (Q4_K_M / AWQ / GPTQ-Int4)0.5 GB 1.0 GB 8GB GPUMac 16GB

Bits per weight: BF16 16, FP8 8.5, 4-bit 4.85. “Fits” = total ≤ 90% of memory; Mac GPU can use ~2/3 of unified memory (≤36GB) or ~3/4 (larger). For MoE, all experts must be in memory, so size by total parameters. Method →

03How to run it

Only runtimes the publisher or the runtime’s official library names for this model.

Which inference engine to choose →

04Official repositories & quantised builds

1 official repo(s) from nvidia.

05Other NVIDIA Parakeet / Canary speech models

All 10 →
ModelParams4-bit totalReleased
canary-180m-flash · same series180M0.6 GB2025-03-11
parakeet-unified-en-0.6b600M0.9 GB2026-04-07
multitalker-parakeet-streaming-0.6b-v1600M0.9 GB2025-10-15
canary-1b-v2979M1.2 GB2025-08-04
canary-qwen-2.5b2.56B2.2 GB2025-06-26
parakeet-tdt-0.6b-v2600M0.9 GB2025-04-15
parakeet-tdt_ctc-110m110M0.6 GB2024-09-17
parakeet-tdt_ctc-0.6b-ja600M0.9 GB2024-05-13
parakeet-tdt_ctc-1.1b1.1B1.2 GB2024-05-07

06Sources

Data checked 2026-06-29. Parameters, context and licence are copied from the official repository; memory figures are our estimates.

Data updated: Self-host open models(UTC+8)