Voxtral-Mini-4B-Realtime-2602 needs about 3.7 GB of memory at Q4_K_M (a 2.9 GB file) with an 8K context window, so it fits on a 4 GB GPU. This repository has 11 quants, from 1.5 GB (Q2_K) to 4.7 GB (Q8_0).
Voxtral‑Mini‑4B‑Realtime‑2602 – low‑latency, multilingual speech transcription
Voxtral‑Mini‑4B‑Realtime‑2602 is a 4.43 billion‑parameter model for real‑time speech‑to‑text. It supports streaming audio with a delay of 480 ms and can process up to 131072 tokens (~3 h) per session. The model is available in multiple GGUF quantizations.
Good at
- Low‑latency real‑time transcription
- Multilingual speech support
- High accuracy comparable to offline systems
Limits
- No vision support
- Requires vLLM for production use
- License not specified
Best for
- Live captioning of meetings
- Voice‑assistant transcription
- Interactive audio applications
Who it’s for. Developers building real‑time transcription services or voice‑assistant applications who need a lightweight, open‑source solution. The model is suitable for CPU or GPU inference via vLLM or voxtral.cpp.
Quants & memory
| Quant | File | Est. memory | vs VRAM | Result | |
|---|---|---|---|---|---|
| Q2_K | 1.5 GB | 2.1 GB | Fits | Get file → | |
| Q3_K | 1.9 GB | 2.7 GB | Fits | Get file → | |
| Q4_0 | 2.5 GB | 3.3 GB | Fits | Get file → | |
| Q4_K | 2.5 GB | 3.3 GB | Fits | Get file → | |
| Q4_1 | 2.8 GB | 3.6 GB | Fits | Get file → | |
| Q4_K_M | 2.9 GB | 3.7 GB | Fits | Get file → | |
| Q5_0 | 3.1 GB | 3.9 GB | Fits | Get file → | |
| Q5_K | 3.1 GB | 3.9 GB | Fits | Get file → | |
| Q5_1 | 3.3 GB | 4.2 GB | Fits | Get file → | |
| Q6_K | 3.6 GB | 4.6 GB | Fits | Get file → | |
| Q8_0 | 4.7 GB | 5.8 GB | Fits | Get file → |
For 8 GB VRAM, Q4_0 (file size 2.5 GB, estimated 3.3 GB memory) is a good balance. If memory is limited to 4 GB, Q3_K (1.92 GB, 2.7 GB memory) is recommended. For maximum quality within 5 GB, Q5_1 (3.33 GB, 4.2 GB memory) can be used.
What GPU do you need for Voxtral-Mini-4B-Realtime-2602?
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 Q5_K (3.1 GB file, about 3.9 GB in use).
- 6 GB VRAM: the best quant that fits is Q8_0 (4.7 GB file, about 5.8 GB in use).
Which Voxtral-Mini-4B-Realtime-2602 quant should you choose?
Q4_K_M is the usual starting point for Voxtral-Mini-4B-Realtime-2602: it balances file size and quality well. Here it is a 2.9 GB file.
The Q8_0 file is 4.7 GB, about 1.6 times the size of Q4_K_M. It keeps the weights closer to the original, which only matters if you have the memory to spare.
If memory is tight, the smallest file is Q2_K at 1.5 GB, with the quality trade-offs that come with fewer bits. Use it only when memory is very tight. Expect a clear drop in quality, which is usually worse on small models and easier to live with on large ones.
New to quant names? Read GGUF quantization explained.
Specifications
Architecture from the GGUF header
Source
Frequently asked questions about Voxtral-Mini-4B-Realtime-2602
How much VRAM does Voxtral-Mini-4B-Realtime-2602 need?
About 3.7 GB at Q4_K_M with an 8K context window. That is the 2.9 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 Voxtral-Mini-4B-Realtime-2602 on an 8 GB GPU?
Yes, with Q8_0: the 4.7 GB file needs about 5.8 GB at 8K context. Larger quants in this repository do not fit in 8 GB.
What is the file size of Voxtral-Mini-4B-Realtime-2602 GGUF?
It depends on the quant: Q2_K is 1.5 GB, Q3_K is 1.9 GB, Q4_0 is 2.5 GB, Q4_K is 2.5 GB, Q4_1 is 2.8 GB, Q4_K_M is 2.9 GB and 5 more.
Where can I download Voxtral-Mini-4B-Realtime-2602 GGUF?
The files are on Hugging Face in the andrijdavid/Voxtral-Mini-4B-Realtime-2602-GGUF repository. The download buttons on this page open the file on Hugging Face. ggufmodels.com does not host any model files.
Can I run Voxtral‑Mini‑4B‑Realtime‑2602 on 8 GB VRAM?
Yes, the Q4_0 quantization uses about 3.3 GB of VRAM at 8 k context, fitting comfortably in 8 GB.
Does the model support streaming audio?
Yes, it is designed for low‑latency streaming and works with vLLM’s Realtime API.
Is there a GPU requirement?
No GPU is required; the model runs on CPU via voxtral.cpp or on GPU via vLLM.