// FunAudioLLM speech · Speech-to-text
Fun-ASR-Nano-2512-hf: VRAM & local deployment
830M parameters. About 1.1 GB at 4-bit and 2.3 GB at BF16.
released 2026-05-24licence: Commercial OKmemory figures are estimates
01Specs
| Publisher | Alibaba Tongyi Lab · FunAudioLLM |
|---|---|
| Family / series | FunAudioLLM speech · Fun-ASR-Nano |
| Type | Speech-to-text (automatic-speech-recognition) |
| Parameters | 830M Source: counted from official safetensors weights |
| Released weights | BF16 · 1.7 GBdownload |
| Licence | Apache-2.0 · Commercial OKPermissive licence (Apache-2.0 / MIT / CC-BY / OpenMDW): commercial use allowed. |
| Released | 2026-05-24 (Hugging Face repo created) |
02How much VRAM it needs (estimate)
Speech models are small; the weight size is close to the total memory needed.
| Precision | Weights | Weights + overhead | Smallest GPU | Mac |
|---|---|---|---|---|
| BF16 / FP16 (original) | 1.7 GB | 2.3 GB | 8GB GPU | Mac 16GB |
| FP8 / INT8 / Q8_0 | 0.9 GB | 1.5 GB | 8GB GPU | Mac 16GB |
| 4-bit (Q4_K_M / AWQ / GPTQ-Int4) | 0.5 GB | 1.1 GB | 8GB GPU | Mac 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.
- vLLM — mentioned in the official model card
- Transformers — mentioned in the official model card
- llama.cpp — official GGUF weights
04Official repositories & quantised builds
3 official repo(s) from FunAudioLLM.
- Fun-ASR-Nano-2512-hf HF-transformers
- Fun-ASR-Nano-2512-GGUF GGUF
- Fun-ASR-Nano-2512 main
05Other FunAudioLLM speech models
All 5 →| Model | Params | 4-bit total | Released |
|---|---|---|---|
| Fun-ASR-MLT-Nano-2512 | — | — | 2025-12-15 |
| Fun-CosyVoice3-0.5B-2512 | 500M | 0.8 GB | 2025-12-11 |
| CosyVoice2-0.5B | 500M | 0.8 GB | 2025-07-29 |
| SenseVoiceSmall | — | — | 2024-07-03 |
06Sources
Data checked 2026-09-10. Parameters, context and licence are copied from the official repository; memory figures are our estimates.