Clef-Flash needs about 8.3 GB of memory at Q4_K_M (a 6.7 GB file) with an 8K context window, so it fits on a 12 GB GPU. This repository has 3 quants, from 6.7 GB (Q4_K_M) to 18.2 GB (BF16). Longer context needs more: at 32K tokens the KV cache alone adds about 4.3 GB.
Clef-Flash: 9B multimodal decision model
Clef-Flash is a 9B multimodal model that outputs probabilities for each allowed option in a decision task. It accepts text, JSON, images, or video as state and returns a single forward pass result. The model has no free‑form text generation.
Good at
- Decision making with multimodal input
- Fast single‑pass probability outputs
- No output parsing needed
Limits
- No free‑form text generation
- Requires 262144 context tokens
- No vision support in llama.cpp without patch
Best for
- Automated decision systems
- Jev/SystemOne API integration
- Batch decision scoring
Who it’s for. Developers building decision‑making services that need fast, deterministic probability outputs from multimodal data. The model is suitable for backend services that require a single forward pass per request.
Quants & memory
| Quant | File | Est. memory | vs VRAM | Result | |
|---|---|---|---|---|---|
| Q4_K_M | 6.7 GB | 8.3 GB | Too big | Get file → | |
| Q8_0 | 9.7 GB | 11.2 GB | Too big | Get file → | |
| BF16 | 18.2 GB | 19.7 GB | Too big | Get file → |
For 8 GB VRAM, use Q4_K_M (6.73 GB, ~8.3 GB memory at 8k context). For 12 GB VRAM, Q8_0 (9.66 GB, ~11.2 GB memory). BF16 requires 18.16 GB and ~19.7 GB memory, suitable only for larger GPUs.
What GPU do you need for Clef-Flash?
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: 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: the best quant that fits is Q8_0 (9.7 GB file, about 11.2 GB in use).
- 16 GB VRAM: the best quant that fits is Q8_0 (9.7 GB file, about 11.2 GB in use).
- 24 GB VRAM: the best quant that fits is BF16 (18.2 GB file, about 19.7 GB in use).
Which Clef-Flash quant should you choose?
Q4_K_M is the usual starting point for Clef-Flash: it balances file size and quality well. Here it is a 6.7 GB file.
The Q8_0 file is 9.7 GB, about 1.4 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.
New to quant names? Read GGUF quantization explained.
Specifications
Architecture from the GGUF header
Source
Frequently asked questions about Clef-Flash
How much VRAM does Clef-Flash need?
About 8.3 GB at Q4_K_M with an 8K context window. That is the 6.7 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 Clef-Flash on an 8 GB GPU?
Not at an 8K context. The smallest file needs about 8.3 GB, so you would need a larger GPU or CPU offloading.
What is the file size of Clef-Flash GGUF?
It depends on the quant: Q4_K_M is 6.7 GB, Q8_0 is 9.7 GB, BF16 is 18.2 GB.
What context length does Clef-Flash support?
The GGUF header lists a trained context length of 262,144 tokens (256K). The KV cache grows with the context you use: about 1.1 GB at 8K and 4.3 GB at 32K.
Where can I download Clef-Flash GGUF?
The files are on Hugging Face in the ggml-org/Clef-Flash-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 Clef-Flash 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 Clef-Flash on 8 GB VRAM?
Yes, using the Q4_K_M quantized file (6.73 GB) which requires about 8.3 GB of VRAM at 8k context.
Does Clef-Flash support image input?
Yes, it accepts images and video, but llama.cpp requires a specific pull request to enable vision support.