LFM2.5 VL 3B DSpark needs about 1.1 GB of memory at F16 (a 0.6 GB file) with an 8K context window, so it fits on a 4 GB GPU. Longer context needs more: at 32K tokens the KV cache alone adds about 0.3 GB.
LFM2.5 VL 3B DSpark – a speculative‑decoding draft for vision‑language tasks
LFM2.5 VL 3B DSpark is a draft sidecar that speeds up the LiquidAI/LFM2.5‑VL‑3B model. It contains 279.5M draft parameters and runs with 16‑bit precision. The draft is paired with the target model for exact greedy decoding.
Good at
- Fast inference on vision‑language tasks
- Exact output under greedy decoding
- Compatible with llama.cpp
Limits
- Requires paired LFM2.5‑VL‑3B target
- No standalone vision encoder
- Limited to 8k context for memory estimate
Best for
- Rapid VQA prototyping
- Image captioning with low latency
- Edge deployment on Apple silicon
Who it’s for. Developers building vision‑language applications who need faster inference without sacrificing output quality. It is suitable for research and production environments that support llama.cpp and have the target model available.
Quants & memory
| Quant | File | Est. memory | vs VRAM | Result | |
|---|---|---|---|---|---|
| F16 | 0.6 GB | 1.1 GB | Fits | Get file → |
The only available quant is F16, 0.57 GB file size. It requires about 1.1 GB of VRAM at 8k context. For 8 GB VRAM you can run it comfortably; for 4 GB VRAM it is too large.
What GPU do you need for LFM2.5 VL 3B 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 F16 (0.6 GB file, about 1.1 GB in use).
Which LFM2.5 VL 3B DSpark quant should you choose?
F16 (0.6 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 LFM2.5 VL 3B DSpark
How much VRAM does LFM2.5 VL 3B DSpark need?
About 1.1 GB at F16 with an 8K context window. That is the 0.6 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 LFM2.5 VL 3B DSpark on an 8 GB GPU?
Yes, with F16: the 0.6 GB file needs about 1.1 GB at 8K context. Larger quants in this repository do not fit in 8 GB.
What is the file size of LFM2.5 VL 3B DSpark GGUF?
It depends on the quant: F16 is 0.6 GB.
What context length does LFM2.5 VL 3B DSpark support?
The GGUF header lists a trained context length of 128,000 tokens (125K). The KV cache grows with the context you use: about 0.1 GB at 8K and 0.3 GB at 32K.
Where can I download LFM2.5 VL 3B DSpark GGUF?
The files are on Hugging Face in the LiquidAI/LFM2.5-VL-3B-DSpark-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 LFM2.5 VL 3B DSpark use?
Hugging Face lists the license as other. 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 LFM2.5 VL 3B DSpark on 8 GB VRAM?
Yes, the F16 quant uses about 1.1 GB of VRAM at 8k context, leaving room for the target model and overhead.
Does this model support vision tasks?
Yes, it includes a vision encoder and is designed for vision‑language tasks such as VQA and image captioning.
Do I need a separate target model?
Yes, the draft must be paired with the LFM2.5‑VL‑3B target file to function.