Thaki ASR KO 1.7B needs about 2.5 GB of memory at Q4_K_M (a 1.1 GB file) with an 8K context window, so it fits on a 4 GB GPU. This repository has 12 quants, from 0.8 GB (Q2_K) to 3.4 GB (F16). Longer context needs more: at 32K tokens the KV cache alone adds about 3.8 GB.
Thaki ASR KO 1.7B – Korean speech‑recognition model
Thaki ASR KO 1.7B is a Korean fine‑tune of Qwen3-ASR-1.7B. It targets Korean telephony and customer‑service speech, with strong numeric content handling. The model is released under Apache‑2.0 and supports vision.
Good at
- Korean telephony ASR
- Numeric content recognition
- Vision‑enabled inference
Limits
- No support for languages other than Korean
- Limited to audio ≤20 seconds
- Requires qwen-asr package
Best for
- Korean call‑center transcription
- Price and ID extraction from speech
- Vision‑based Korean speech tasks
Who it’s for. Developers building Korean ASR applications that need accurate telephony and numeric transcription. The model is suitable for production use where Apache‑2.0 licensing is acceptable.
Quants & memory
| Quant | File | Est. memory | vs VRAM | Result | |
|---|---|---|---|---|---|
| Q2_K | 0.8 GB | 2.2 GB | Fits | Get file → | |
| Q3_K_S | 0.9 GB | 2.3 GB | Fits | Get file → | |
| Q3_K_M | 0.9 GB | 2.4 GB | Fits | Get file → | |
| Q3_K_L | 1.0 GB | 2.4 GB | Fits | Get file → | |
| IQ4_XS | 1.0 GB | 2.5 GB | Fits | Get file → | |
| Q4_K_S | 1.1 GB | 2.5 GB | Fits | Get file → | |
| Q4_K_M | 1.1 GB | 2.5 GB | Fits | Get file → | |
| Q5_K_S | 1.2 GB | 2.7 GB | Fits | Get file → | |
| Q5_K_M | 1.3 GB | 2.7 GB | Fits | Get file → | |
| Q6_K | 1.4 GB | 2.9 GB | Fits | Get file → | |
| Q8_0 | 1.8 GB | 3.3 GB | Fits | Get file → | |
| F16 | 3.4 GB | 4.9 GB | Fits | Get file → |
For 8 GB VRAM use Q4_K_M (file size 1.11 GB, estimated 2.5 GB memory). For 4 GB VRAM use Q3_K_M (file size 0.94 GB, estimated 2.4 GB memory). For 2 GB VRAM use Q2_K (file size 0.78 GB, estimated 2.2 GB memory).
What GPU do you need for Thaki ASR KO 1.7B?
These figures add the file size, an fp16 KV cache at an 8K context window (worked out from this model’s own header) 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 (1.8 GB file, about 3.3 GB in use).
- 6 GB VRAM: the best quant that fits is F16 (3.4 GB file, about 4.9 GB in use).
Which Thaki ASR KO 1.7B quant should you choose?
Q4_K_M is the usual starting point for Thaki ASR KO 1.7B: it balances file size and quality well. Here it is a 1.1 GB file.
The Q8_0 file is 1.8 GB, about 1.7 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 0.8 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 Thaki ASR KO 1.7B
How much VRAM does Thaki ASR KO 1.7B need?
About 2.5 GB at Q4_K_M with an 8K context window. That is the 1.1 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 Thaki ASR KO 1.7B on an 8 GB GPU?
Yes, with F16: the 3.4 GB file needs about 4.9 GB at 8K context. Larger quants in this repository do not fit in 8 GB.
What is the file size of Thaki ASR KO 1.7B GGUF?
It depends on the quant: Q2_K is 0.8 GB, Q3_K_S is 0.9 GB, Q3_K_M is 0.9 GB, Q3_K_L is 1.0 GB, IQ4_XS is 1.0 GB, Q4_K_S is 1.1 GB and 6 more.
What context length does Thaki ASR KO 1.7B support?
The GGUF header lists a trained context length of 65,536 tokens (64K). The KV cache grows with the context you use: about 0.9 GB at 8K and 3.8 GB at 32K.
Where can I download Thaki ASR KO 1.7B GGUF?
The files are on Hugging Face in the mradermacher/Thaki-ASR-KO-1.7B-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 Thaki ASR KO 1.7B 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, Q4_K_M or Q4_K_M is suitable for 8 GB VRAM, using about 2.5 GB of memory at 8k context.
Does the model support languages other than Korean?
No, the model is fine‑tuned for Korean and expects Korean input; other languages fall back to the base Qwen3-ASR-1.7B.
What is the maximum audio length supported?
The model is evaluated on audio ≤20 seconds; performance beyond that is not guaranteed.