ark asr 3b needs about 4.4 GB of memory at Q4_K (a 3.5 GB file) with an 8K context window, so it fits on a 6 GB GPU. This repository has 6 quants, from 0.0 GB to 7.5 GB (F16).
19‑language automatic speech recognition model in GGUF format
ARK‑ASR‑3B is a 3.75‑billion‑parameter audio‑capable autoregressive transformer. It combines a Whisper‑style audio encoder, an MLP adapter, and a Qwen decoder. The model is available as a single GGUF file for use with CrispASR.
Good at
- Multilingual ASR for 19 languages
- Whisper‑style encoder with arbitrary‑length audio
- Qwen decoder with audio‑token injection
Limits
- No vision support
- Experimental backend with rough edges
- No explicit context length limit but memory usage is high
Best for
- Transcribing long audio clips
- Multilingual speech recognition tasks
- Batch ASR inference with CrispASR
Who it’s for. Developers and researchers needing a multilingual ASR model in GGUF format for local inference. The model is suitable for those who can handle experimental code and have sufficient memory for the larger quantizations.
Quants & memory
| Quant | File | Est. memory | vs VRAM | Result | |
|---|---|---|---|---|---|
| — | 0.0 GB | 0.5 GB | Fits | Get file → | |
| Q3_K | 3.2 GB | 4.0 GB | Fits | Get file → | |
| Q4_K | 3.5 GB | 4.4 GB | Fits | Get file → | |
| Q4_K | 3.5 GB | 4.4 GB | Fits | Get file → | |
| Q8_0 | 4.3 GB | 5.3 GB | Fits | Get file → | |
| F16 | 7.5 GB | 8.9 GB | Too big | Get file → |
For 8 GB VRAM, use Q4_K (file size 3.52 GB, estimated 4.4 GB memory at 8 k context). For 16 GB VRAM, Q8_0 (4.29 GB, 5.3 GB memory) or F16 (7.51 GB, 8.9 GB memory) provide higher quality. The unknown quant is very small but memory usage is not specified.
What GPU do you need for ark asr 3b?
These figures add the file size, an fp16 KV cache at an 8K context window and about 0.5 GB for the runtime. They are estimates, so leave a little headroom, and expect to use less context or a smaller quant if you are right at the limit.
- 4 GB VRAM: the best quant that fits is Q3_K (3.2 GB file, about 4.0 GB in use).
- 6 GB VRAM: the best quant that fits is Q8_0 (4.3 GB file, about 5.3 GB in use).
- 8 GB VRAM: the best quant that fits is Q8_0 (4.3 GB file, about 5.3 GB in use).
- 12 GB VRAM: the best quant that fits is F16 (7.5 GB file, about 8.9 GB in use).
Which ark asr 3b quant should you choose?
Q4_K (3.5 GB) is the middle option in this repository. The usual sweet spot between file size and quality, and the quant most people start with.
If memory is tight, the smallest file is the first one in the table at 0.0 GB, with the quality trade-offs that come with fewer bits.
New to quant names? Read GGUF quantization explained.
Specifications
Architecture from the GGUF header
Source
Frequently asked questions about ark asr 3b
How much VRAM does ark asr 3b need?
About 4.4 GB at Q4_K with an 8K context window. That is the 3.5 GB file, plus the KV cache and roughly 0.5 GB of runtime overhead. Smaller quants need less and longer context needs more.
Can I run ark asr 3b on an 8 GB GPU?
Yes, with Q8_0: the 4.3 GB file needs about 5.3 GB at 8K context. Larger quants in this repository do not fit in 8 GB.
What is the file size of ark asr 3b GGUF?
It depends on the quant: file is 0.0 GB, Q3_K is 3.2 GB, Q4_K is 3.5 GB, Q4_K is 3.5 GB, Q8_0 is 4.3 GB, F16 is 7.5 GB.
Where can I download ark asr 3b GGUF?
The files are on Hugging Face in the cstr/ark-asr-3b-GGUF repository. The download buttons on this page open the file on Hugging Face. ggufmodels.com does not host any model files.
What license does ark asr 3b use?
Hugging Face lists the license as other. Licenses can differ between the quantized files and the original model, so check the repository and the base model before commercial use.
Can I run ARK‑ASR‑3B on 8 GB VRAM?
Yes, the Q4_K quantization requires about 4.4 GB memory at 8 k context, which fits within 8 GB VRAM.
Does the model support vision tasks?
No, the model is designed for automatic speech recognition only.
Is the GGUF conversion stable?
The conversion is experimental; core ASR is validated but the backend has known rough edges.