cohere transcribe arabic 07 2026 needs about 2.2 GB of memory at Q4_K (a 1.5 GB file) with an 8K context window, so it fits on a 4 GB GPU. This repository has 6 quants, from 0.0 GB to 4.1 GB (F16).
Cohere Arabic ASR model, 2.07B parameters, 11% WER on FLEURS Arabic
This model is a 2.07‑billion‑parameter Arabic automatic speech recognition system. It uses a FastConformer encoder and a Transformer decoder. The GGUF weights are available in several quantizations and run on CPU, Metal, CUDA, or Vulkan via CrispASR.
Good at
- Accurate Arabic speech transcription
- Fast inference on modern GPUs
- Supports English and Arabic languages
Limits
- Only en and ar languages supported
- Requires 8k context for memory estimates
- No vision or multimodal support
Best for
- Transcribing Arabic audio clips
- Integrating into voice‑to‑text pipelines
- Evaluating ASR performance on Arabic datasets
Who it’s for. Developers building Arabic speech‑to‑text applications. Researchers testing ASR models on Arabic data. Engineers needing a lightweight, open‑source ASR solution.
Quants & memory
| Quant | File | Est. memory | vs VRAM | Result | |
|---|---|---|---|---|---|
| — | 0.0 GB | 0.5 GB | Fits | Get file → | |
| — | 0.0 GB | 0.5 GB | Fits | Get file → | |
| Q4_K | 1.5 GB | 2.2 GB | Fits | Get file → | |
| Q4_K | 1.5 GB | 2.2 GB | Fits | Get file → | |
| Q8_0 | 2.4 GB | 3.2 GB | Fits | Get file → | |
| F16 | 4.1 GB | 5.1 GB | Fits | Get file → |
For 8 GB VRAM, use the Q4_K‑imatrix quant (file size 1.51 GB, estimated memory 2.2 GB). For 12 GB VRAM, the Q8_0 quant (2.42 GB, 3.2 GB memory) offers higher precision. The F16 quant (4.14 GB, 5.1 GB memory) is best for maximum accuracy when memory allows.
What GPU do you need for cohere transcribe arabic 07 2026?
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 Q8_0 (2.4 GB file, about 3.2 GB in use).
- 6 GB VRAM: the best quant that fits is F16 (4.1 GB file, about 5.1 GB in use).
Which cohere transcribe arabic 07 2026 quant should you choose?
Q4_K (1.5 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
Frequently asked questions about cohere transcribe arabic 07 2026
How much VRAM does cohere transcribe arabic 07 2026 need?
About 2.2 GB at Q4_K with an 8K context window. That is the 1.5 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 cohere transcribe arabic 07 2026 on an 8 GB GPU?
Yes, with F16: the 4.1 GB file needs about 5.1 GB at 8K context. Larger quants in this repository do not fit in 8 GB.
What is the file size of cohere transcribe arabic 07 2026 GGUF?
It depends on the quant: file is 0.0 GB, file is 0.0 GB, Q4_K is 1.5 GB, Q4_K is 1.5 GB, Q8_0 is 2.4 GB, F16 is 4.1 GB.
Where can I download cohere transcribe arabic 07 2026 GGUF?
The files are on Hugging Face in the cstr/cohere-transcribe-arabic-07-2026-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 cohere transcribe arabic 07 2026 use?
Hugging Face lists the license as apache-2.0. 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 this model on 8 GB VRAM?
Yes, the Q4_K‑imatrix quant uses about 2.2 GB of VRAM at 8k context, fitting comfortably in 8 GB.
Does the model support other languages?
No, only English and Arabic are supported by the finetune.
What license applies to the GGUF weights?
The weights are licensed under Apache 2.0, inherited from the source model.