CantoneseLLM v2.0 30B A3B Thinking Chat Vector Merged needs about 19.9 GB of memory at Q4_K_M (a 18.6 GB file) with an 8K context window, so it fits on a 24 GB GPU. This repository has 24 quants, from 0.1 GB to 25.1 GB (Q6_K). Longer context needs more: at 32K tokens the KV cache alone adds about 3.2 GB.
30.5B‑parameter Cantonese‑knowledgeable instruction model
CantoneseLLM v2.0 30B A3B Thinking Chat Vector Merged is a 30.5B‑parameter mixture‑of‑experts model built from Qwen3‑30B‑A3B‑Base. It has 3.3B active parameters and was trained on 784M Cantonese and Hong Kong text tokens. The model supports instruction following and reasoning in Written Chinese.
Good at
- Cantonese instruction following
- Reasoning in Written Chinese
- Large context length 40960 tokens
Limits
- No Cantonese chain‑of‑thought reasoning
- No vision support
- Requires large VRAM for full precision
Best for
- Chatbots needing Cantonese knowledge
- Research on low‑resource language reasoning
- Fine‑tuning Cantonese instruction models
Who it’s for. Developers building Cantonese‑aware chat applications or researchers studying low‑resource language reasoning. The model is suitable when Cantonese instruction following is required but the chain‑of‑thought language can be Written Chinese.
Quants & memory
| Quant | File | Est. memory | vs VRAM | Result | |
|---|---|---|---|---|---|
| — | 0.1 GB | 1.4 GB | Fits | Get file → | |
| IQ1_S | 6.4 GB | 7.7 GB | Tight | Get file → | |
| IQ1_M | 7.1 GB | 8.4 GB | Too big | Get file → | |
| IQ2_XXS | 8.2 GB | 9.5 GB | Too big | Get file → | |
| IQ2_XS | 9.1 GB | 10.4 GB | Too big | Get file → | |
| IQ2_S | 9.3 GB | 10.6 GB | Too big | Get file → | |
| IQ2_M | 10.2 GB | 11.5 GB | Too big | Get file → | |
| Q2_K_S | 10.5 GB | 11.8 GB | Too big | Get file → | |
| Q2_K | 11.3 GB | 12.6 GB | Too big | Get file → | |
| IQ3_XXS | 11.8 GB | 13.2 GB | Too big | Get file → | |
| IQ3_XS | 12.6 GB | 13.9 GB | Too big | Get file → | |
| Q3_K_S | 13.3 GB | 14.6 GB | Too big | Get file → | |
| IQ3_S | 13.3 GB | 14.6 GB | Too big | Get file → | |
| IQ3_M | 13.5 GB | 14.8 GB | Too big | Get file → | |
| Q3_K_M | 14.7 GB | 16.0 GB | Too big | Get file → | |
| Q3_K_L | 15.9 GB | 17.2 GB | Too big | Get file → | |
| IQ4_XS | 16.4 GB | 17.7 GB | Too big | Get file → | |
| Q4_0 | 17.4 GB | 18.7 GB | Too big | Get file → | |
| Q4_K_S | 17.5 GB | 18.8 GB | Too big | Get file → | |
| Q4_K_M | 18.6 GB | 19.9 GB | Too big | Get file → | |
| Q4_1 | 19.2 GB | 20.5 GB | Too big | Get file → | |
| Q5_K_S | 21.1 GB | 22.4 GB | Too big | Get file → | |
| Q5_K_M | 21.7 GB | 23.0 GB | Too big | Get file → | |
| Q6_K | 25.1 GB | 26.4 GB | Too big | Get file → |
For 8‑GB VRAM use IQ1_S (6.42 GB file, 7.7 GB memory at 8k context). For 12‑GB VRAM use Q2_K (11.26 GB file, 12.6 GB memory). For 24‑GB VRAM use Q4_K_M (18.56 GB file, 19.9 GB memory). Estimates are approximate.
What GPU do you need for CantoneseLLM v2.0 30B A3B Thinking Chat Vector Merged?
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 the main file (0.1 GB file, about 1.4 GB in use).
- 6 GB VRAM: the best quant that fits is the main file (0.1 GB file, about 1.4 GB in use).
- 8 GB VRAM: the best quant that fits is IQ1_S (6.4 GB file, about 7.7 GB in use).
- 12 GB VRAM: the best quant that fits is Q2_K_S (10.5 GB file, about 11.8 GB in use).
- 16 GB VRAM: the best quant that fits is Q3_K_M (14.7 GB file, about 16.0 GB in use).
- 24 GB VRAM: the best quant that fits is Q5_K_M (21.7 GB file, about 23.0 GB in use).
- 32 GB VRAM: the best quant that fits is Q6_K (25.1 GB file, about 26.4 GB in use).
Which CantoneseLLM v2.0 30B A3B Thinking Chat Vector Merged quant should you choose?
Q4_K_M is the usual starting point for CantoneseLLM v2.0 30B A3B Thinking Chat Vector Merged: it balances file size and quality well. Here it is a 18.6 GB file.
If memory is tight, the smallest file is the first one in the table at 0.1 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 CantoneseLLM v2.0 30B A3B Thinking Chat Vector Merged
How much VRAM does CantoneseLLM v2.0 30B A3B Thinking Chat Vector Merged need?
About 19.9 GB at Q4_K_M with an 8K context window. That is the 18.6 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 CantoneseLLM v2.0 30B A3B Thinking Chat Vector Merged on an 8 GB GPU?
Yes, with IQ1_S: the 6.4 GB file needs about 7.7 GB at 8K context. Larger quants in this repository do not fit in 8 GB.
What is the file size of CantoneseLLM v2.0 30B A3B Thinking Chat Vector Merged GGUF?
It depends on the quant: file is 0.1 GB, IQ1_S is 6.4 GB, IQ1_M is 7.1 GB, IQ2_XXS is 8.2 GB, IQ2_XS is 9.1 GB, IQ2_S is 9.3 GB and 18 more.
What context length does CantoneseLLM v2.0 30B A3B Thinking Chat Vector Merged support?
The GGUF header lists a trained context length of 40,960 tokens (40K). The KV cache grows with the context you use: about 0.8 GB at 8K and 3.2 GB at 32K.
Where can I download CantoneseLLM v2.0 30B A3B Thinking Chat Vector Merged GGUF?
The files are on Hugging Face in the mradermacher/CantoneseLLM-v2.0-30B-A3B-Thinking-Chat-Vector-Merged-i1-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 CantoneseLLM v2.0 30B A3B Thinking Chat Vector Merged 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, using the IQ1_S quant which requires about 7.7 GB memory at 8k context.
Does this model support Cantonese chain‑of‑thought reasoning?
No, the chain‑of‑thought is in Written Chinese; only the final answer is in Cantonese.
What license applies to this model?
The model is released under the apache‑2.0 license.