Qwen3.6 35B A3B DFlash needs about 0.9 GB of memory at Q4_K_M (a 0.2 GB file) with an 8K context window, so it fits on a 4 GB GPU. This repository has 6 quants, from 0.2 GB (IQ4_XS) to 0.8 GB (F16). Longer context needs more: at 32K tokens the KV cache alone adds about 0.8 GB.
Qwen3.6 35B A3B DFlash draft model for speculative decoding
Qwen3.6 35B A3B DFlash is a draft model built on the Qwen family with a dflash architecture. It is designed to work with a compatible Qwen3.6-35B-A3B target model and supports reasoning and tools abilities. The model is released under the apache-2.0 license.
Good at
- Long‑context reasoning
- Speculative decoding
- Tools integration
Limits
- Requires a compatible target model
- No standalone inference
- Limited to 262144 context tokens
Best for
- High‑throughput text generation
- Parallel token proposal
- Draft‑based inference pipelines
Who it’s for. Developers building inference servers that support DFlash speculative decoding. It is suitable for users who have a compatible Qwen3.6-35B-A3B target model and a recent llama.cpp build with DFlash support.
Quants & memory
| Quant | File | Est. memory | vs VRAM | Result | |
|---|---|---|---|---|---|
| IQ4_XS | 0.2 GB | 0.9 GB | Fits | Get file → | |
| Q4_K_M | 0.2 GB | 0.9 GB | Fits | Get file → | |
| Q5_K_M | 0.3 GB | 1.0 GB | Fits | Get file → | |
| Q6_K | 0.3 GB | 1.0 GB | Fits | Get file → | |
| Q8_0 | 0.4 GB | 1.1 GB | Fits | Get file → | |
| F16 | 0.8 GB | 1.5 GB | Fits | Get file → |
For 8 GB VRAM, the IQ4_XS or Q4_K_M quantizations are recommended, each estimating about 0.9 GB memory at 8 k context. If 12 GB VRAM is available, Q5_K_M or Q6_K provide higher quality with an estimated 1 GB memory usage. For maximum quality on 16 GB VRAM, Q8_0 or F16 can be used, estimating 1.1 GB and 1.5 GB respectively.
What GPU do you need for Qwen3.6 35B A3B DFlash?
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 F16 (0.8 GB file, about 1.5 GB in use).
Which Qwen3.6 35B A3B DFlash quant should you choose?
Q4_K_M is the usual starting point for Qwen3.6 35B A3B DFlash: it balances file size and quality well. Here it is a 0.2 GB file.
The Q8_0 file is 0.4 GB, about 1.8 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 IQ4_XS at 0.2 GB, with the quality trade-offs that come with fewer bits. The usual sweet spot between file size and quality, and the quant most people start with.
New to quant names? Read GGUF quantization explained.
Specifications
Architecture from the GGUF header
Source
Frequently asked questions about Qwen3.6 35B A3B DFlash
How much VRAM does Qwen3.6 35B A3B DFlash need?
About 0.9 GB at Q4_K_M with an 8K context window. That is the 0.2 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 Qwen3.6 35B A3B DFlash on an 8 GB GPU?
Yes, with F16: the 0.8 GB file needs about 1.5 GB at 8K context. Larger quants in this repository do not fit in 8 GB.
What is the file size of Qwen3.6 35B A3B DFlash GGUF?
It depends on the quant: IQ4_XS is 0.2 GB, Q4_K_M is 0.2 GB, Q5_K_M is 0.3 GB, Q6_K is 0.3 GB, Q8_0 is 0.4 GB, F16 is 0.8 GB.
What context length does Qwen3.6 35B A3B DFlash support?
The GGUF header lists a trained context length of 262,144 tokens (256K). The KV cache grows with the context you use: about 0.2 GB at 8K and 0.8 GB at 32K.
Where can I download Qwen3.6 35B A3B DFlash GGUF?
The files are on Hugging Face in the williamliao/Qwen3.6-35B-A3B-DFlash-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 Qwen3.6 35B A3B DFlash 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 Qwen3.6 35B A3B DFlash on 8 GB VRAM?
Yes, using IQ4_XS or Q4_K_M quantizations, which estimate 0.9 GB memory at 8 k context.
Does this model support vision tasks?
No, the model does not include vision capabilities.
What is required to use this draft model?
A compatible Qwen3.6-35B-A3B target model and a recent llama.cpp build with DFlash support.