Skip to content
English

Switch theme

// Microsoft VibeVoice / Fara / Harrier · Embedding

harrier-oss-v1-0.6b: VRAM & local deployment

596M parameters. About 0.9 GB at 4-bit and 1.8 GB at BF16.

released 2026-03-30licence: Commercial OKmemory figures are estimates

01Specs

PublisherMicrosoft · microsoft
Family / seriesMicrosoft VibeVoice / Fara / Harrier · harrier-oss
TypeEmbedding (feature-extraction)
Parameters596M Source: counted from official safetensors weights
Released weightsBF16 · 1.2 GBdownload
LicenceMIT · Commercial OKPermissive licence (Apache-2.0 / MIT / CC-BY / OpenMDW): commercial use allowed.
Released2026-03-30 (Hugging Face repo created)

02How much VRAM it needs (estimate)

Embedding models are small; memory grows with batch size and sequence length.

PrecisionWeightsWeights + overheadSmallest GPUMac
BF16 / FP16 (original)1.2 GB 1.8 GB 8GB GPUMac 16GB
FP8 / INT8 / Q8_00.6 GB 1.2 GB 8GB GPUMac 16GB
4-bit (Q4_K_M / AWQ / GPTQ-Int4)0.4 GB 0.9 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 microsoft.

05Other Microsoft VibeVoice / Fara / Harrier models

All 12 →
ModelParams4-bit totalReleased
harrier-oss-v1-270m · same series268M0.7 GB2026-03-30
harrier-oss-v1-27b · same series27B18.5 GB2026-03-30
VibeVoice-ASR-Streaming-1.5B2.81B2.4 GB2026-09-02
VibeVoice-ASR-Streaming-7B8.67B6.3 GB2026-09-02
Fara1.5-4B4.54B3.8 GB2026-07-17
Fara1.5-27B27.4B19.3 GB2026-07-17
Fara1.5-9B9.41B7.0 GB2026-05-12
VibeVoice-ASR8.67B6.3 GB2026-01-21
VibeVoice-Realtime-0.5B1.02B1.2 GB2025-12-04
Fara-7B8.29B6.5 GB2025-10-30

06Sources

Data checked 2026-03-30. Parameters, context and licence are copied from the official repository; memory figures are our estimates.

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