DeepSeek V4 Flash DSpark draft needs about 12.7 GB of memory at BF16 (a 10.9 GB file) with an 8K context window, so it fits on a 16 GB GPU.
DeepSeek V4 Flash DSpark draft – a 19.8B parameter GGUF model for block‑parallel drafting
DeepSeek V4 Flash DSpark draft is a 19.8B parameter model based on the DeepSeek‑V4‑Flash architecture. It is distributed as a single BF16 GGUF file of 10.9 GB. The model supports a 5‑token block draft and shares embeddings with the target DeepSeek‑V4‑Flash model.
Good at
- Block‑parallel drafting
- Low‑latency inference with BF16
- Compatibility with llama.cpp
Limits
- Requires 12.7 GB VRAM at 8 k context (approx.)
- No vision or multimodal support
- Only BF16 quant available
Best for
- Integrating with existing DeepSeek‑V4‑Flash GGUFs
- Testing DSpark speculative decoding
- Research on block‑parallel drafting
Who it’s for. Developers and researchers who need a lightweight drafter for the DeepSeek‑V4‑Flash model. It is ideal for testing speculative decoding in llama.cpp environments and for exploring block‑parallel drafting techniques.
Quants & memory
| Quant | File | Est. memory | vs VRAM | Result | |
|---|---|---|---|---|---|
| BF16 | 10.9 GB | 12.7 GB | Too big | Get file → |
The only available quant is BF16, which requires approximately 12.7 GB VRAM for an 8 k context. If your GPU has 8 GB VRAM, this model will not run. For lower VRAM budgets, a different model or quant is needed.
What GPU do you need for DeepSeek V4 Flash DSpark draft?
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: no quant fits at an 8K context.
- 6 GB VRAM: no quant fits at an 8K context.
- 8 GB VRAM: no quant fits at an 8K context.
- 12 GB VRAM: no quant fits at an 8K context.
- 16 GB VRAM: the best quant that fits is BF16 (10.9 GB file, about 12.7 GB in use).
Which DeepSeek V4 Flash DSpark draft quant should you choose?
BF16 (10.9 GB) is the middle option in this repository. The unquantized model. It gives original quality and needs the most memory, so it is mainly for conversion or reference.
New to quant names? Read GGUF quantization explained.
Specifications
Architecture from the GGUF header
Source
Frequently asked questions about DeepSeek V4 Flash DSpark draft
How much VRAM does DeepSeek V4 Flash DSpark draft need?
About 12.7 GB at BF16 with an 8K context window. That is the 10.9 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 DeepSeek V4 Flash DSpark draft on an 8 GB GPU?
Not at an 8K context. The smallest file needs about 12.7 GB, so you would need a larger GPU or CPU offloading.
What is the file size of DeepSeek V4 Flash DSpark draft GGUF?
It depends on the quant: BF16 is 10.9 GB.
Where can I download DeepSeek V4 Flash DSpark draft GGUF?
The files are on Hugging Face in the YanissAmz/DeepSeek-V4-Flash-DSpark-draft-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 DeepSeek V4 Flash DSpark draft use?
Hugging Face lists the license as mit. 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?
No. The BF16 quant requires about 12.7 GB VRAM at 8 k context, which exceeds 8 GB.
Does this model support vision tasks?
No. The model has no vision or multimodal capabilities.
What is the context length supported?
The model uses a 5‑token block draft; the underlying DeepSeek‑V4‑Flash supports up to one million tokens.