Kimi K3 DSpark needs about 2.0 GB of memory at Q4_K_M (a 1.3 GB file) with an 8K context window, so it fits on a 4 GB GPU. This repository has 7 quants, from 0.8 GB (Q2_K) to 4.5 GB (BF16). Longer context needs more: at 32K tokens the KV cache alone adds about 0.7 GB.
Kimi K3 DSpark – a 2.25B parameter long‑context speculator
Kimi K3 DSpark is a 2.25B parameter model with a draft architecture based on DFlash. It supports context lengths up to 1 million tokens and is optimized for speculative decoding. The model is available in multiple GGUF quantizations.
Good at
- Long‑context inference
- Speculative decoding
- Embedding generation
Limits
- No vision support
- License not specified
- Requires 8 k context for memory estimates
Best for
- Chat applications needing long context
- Code generation with context awareness
- Embedding tasks for semantic search
Who it’s for. Developers building chat or code assistants that require extended context and efficient inference. The model is suitable for users who can run GGUF files with llama.cpp or BeeLlama.cpp.
Quants & memory
| Quant | File | Est. memory | vs VRAM | Result | |
|---|---|---|---|---|---|
| Q2_K | 0.8 GB | 1.5 GB | Fits | Get file → | |
| Q3_K_M | 1.1 GB | 1.7 GB | Fits | Get file → | |
| Q4_K_M | 1.3 GB | 2.0 GB | Fits | Get file → | |
| Q5_K_M | 1.6 GB | 2.3 GB | Fits | Get file → | |
| Q6_K | 1.9 GB | 2.5 GB | Fits | Get file → | |
| Q8_0 | 2.4 GB | 3.1 GB | Fits | Get file → | |
| BF16 | 4.5 GB | 5.2 GB | Fits | Get file → |
For 8 GB VRAM, Q4_K (1.33 GB) or Q5_K_M (1.58 GB) are suitable, estimating 2 GB or 2.3 GB memory at 8 k context. For 16 GB VRAM, Q6_K (1.85 GB) or Q8_0 (2.4 GB) provide higher quality with 2.5 GB or 3.1 GB memory usage. BF16 offers the highest fidelity but needs 4.51 GB and 5.2 GB memory.
What GPU do you need for Kimi K3 DSpark?
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 (2.4 GB file, about 3.1 GB in use).
- 6 GB VRAM: the best quant that fits is BF16 (4.5 GB file, about 5.2 GB in use).
Which Kimi K3 DSpark quant should you choose?
Q4_K_M is the usual starting point for Kimi K3 DSpark: it balances file size and quality well. Here it is a 1.3 GB file.
The Q8_0 file is 2.4 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 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 Kimi K3 DSpark
How much VRAM does Kimi K3 DSpark need?
About 2.0 GB at Q4_K_M with an 8K context window. That is the 1.3 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 Kimi K3 DSpark on an 8 GB GPU?
Yes, with BF16: the 4.5 GB file needs about 5.2 GB at 8K context. Larger quants in this repository do not fit in 8 GB.
What is the file size of Kimi K3 DSpark GGUF?
It depends on the quant: Q2_K is 0.8 GB, Q3_K_M is 1.1 GB, Q4_K_M is 1.3 GB, Q5_K_M is 1.6 GB, Q6_K is 1.9 GB, Q8_0 is 2.4 GB and 1 more.
What context length does Kimi K3 DSpark support?
The GGUF header lists a trained context length of 1,048,576 tokens (1024K). The KV cache grows with the context you use: about 0.2 GB at 8K and 0.7 GB at 32K.
Where can I download Kimi K3 DSpark GGUF?
The files are on Hugging Face in the Anbeeld/Kimi-K3-DSpark-GGUF repository. The download buttons on this page open the file on Hugging Face. ggufmodels.com does not host any model files.
Can I run Kimi K3 DSpark on 8 GB VRAM?
Yes, using Q4_K or Q5_K_M quantizations will fit within 8 GB VRAM, with memory usage around 2 GB at 8 k context.
Does the model support vision tasks?
No, the model has no vision capabilities.
What context length does it support?
It supports context lengths up to 1 million tokens.