This is a decision model: use llama-server's /v1/systemone endpoint with -ub 4096 -b 4096. Follow Run it for supported settings and app compatibility.
Start here. Download Clef-Artemis-IQ4_XS.gguf (14.82 GB). 95.6% same answers as BF16; the smallest variants change some confident answers. Details.
Typed-decision accuracy on the card's selection panel, with the measured nearby-size alternatives and BF16 original. These comparisons informed file selection and do not establish performance on unseen inputs. Hardware, runtime and other evaluation panels are listed under How it was tested.
These comparisons informed file selection; they do not establish performance on unseen inputs.
Earlier file names still work for now; new downloads should use the names below.
llama-server -hf turintech/Clef-Artemis-GGUF:IQ4_XS --port 8080 -c 16384 -ub 4096 -b 4096
Clef is Cloudflare's 27B decision model. The model accepts text (and optionally images) with a set of typed questions, and it returns a probability for every allowed answer in a single pass, instead of generating text. See the original model card.
This is an independent TurinTech conversion, not affiliated with or endorsed by Cloudflare. The files are standard
GGUF and run on stock llama.cpp. For images, use the vision projector from
ggml-org/Clef-GGUF (see below). Changes from the original: the weights
are converted to GGUF and quantized for llama.cpp. See NOTICE.
| Variant | Size | Saved | Same answer | Acc. Δ | Conf. changes | Choose for |
|---|---|---|---|---|---|---|
| BF16 original | 54.1 GB | reference | reference | |||
| IQ4_XS | 14.82 GB | 73% | 95.6% | -0.1 | 0 | Default. |
| IQ3_M | 14.31 GB | 74% | 95.9% | +0.7 | 0 | Smaller alternative. |
| Q3_K_M | 12.72 GB | 76% | 93.4% | +0.3 | 0 | Tight memory. |
| IQ3_XXS | 11.53 GB | 79% | 91.0% | +1.1 | 2 | Tight memory. |
| Q2_K | 10.49 GB | 81% | 89.5% | +0.9 | 4 (+1 image) | Smallest. |
"Saved" is the file-size reduction versus the BF16 original. "Acc. Δ" is the accuracy change in percentage points; the other columns are defined below. Variant links download the exact GGUF; full filenames are in the commands and checksums. File labels are size classes, explained under "File names, sizes and SHA-256 checksums" at the end of the card.
IQ4_XS: Default used by this card's commands; see the similar-quality Artemis IQ3_M alternative. IQ3_M: Similar measured quality to Artemis IQ4_XS, with a smaller download. Q2_K: Smallest. Peak 11.3 GiB for text, so it should fit a 12 GB GPU for text.
"Same answer as the original" is how often a file picks the same option as the BF16 original on 2,000 test decisions. "Accuracy change" is the difference in accuracy against the original on the same cases, in points. "Confident answers changed" counts answers the original was more than 80% sure of that the file changed.
The usual public files, on the same tests: Q8_0 (ggml-org, 28.7 GB) gives the same answer 98.8% of the time, Q4_K_M (ggml-org, 19.2 GB) 96.9% and IQ3_M (bartowski, 13.78 GB) 87.6%. None of them changed a confident answer; of ours, the two smallest did (2 and 4 of about 2,000).
echo "My Stripe integration keeps failing. Please help ASAP." | ollama run turintech/clef-artemis:14.8gb --ask "Does this message express urgency?"
This verified tag selects Clef-Artemis-IQ4_XS.gguf. It is a decision model: use /v1/systemone, not a chat prompt. Ollama 0.40.4 was tested; include "model": "turintech/clef-artemis:14.8gb" in the decision request to port 11434. The command runs the documented decision example; the explicit size tag selects the recommended file and is not an immutable revision.
Needs llama.cpp from 9 October 2026 or newer. Tested at commit 1e6f04a, and every file also on release b11572
(12 October 2026), which gave identical answers on a 100-decision check.
llama-server -m Clef-Artemis-IQ4_XS.gguf --port 8080 -c 16384 -ub 4096 -b 4096
Tested on an Apple M3 Max (36 GB) with llama.cpp Metal (b11617): every file here loads and answers with the command above.
Keep -ub 4096 -b 4096. At the default of 512, /v1/systemone rejects inputs longer than 512 tokens. That is true
of every Clef GGUF, public ones included. -c 16384 stops the server from reserving memory for the full training
context.
curl http://localhost:8080/v1/systemone -H 'Content-Type: application/json' -d '{
"state": "My Stripe integration keeps failing. Please help ASAP.",
"questions": {"urgency": {"type": "noul", "instructions": "Does this message express urgency?"}}
}'
The request and response follow the Jev/SystemOne format described on the original model card.
Questions, problems or a model to consider for a smaller version? Reply in our feedback thread.
Add the vision projector from ggml-org/Clef-GGUF and raise the batch size, since several large images can exceed 4,096 tokens:
hf download ggml-org/Clef-GGUF mmproj-Clef-BF16.gguf --local-dir .
llama-server -m Clef-Artemis-IQ4_XS.gguf --port 8080 --mmproj mmproj-Clef-BF16.gguf -c 16384 -ub 8192 -b 8192
Send images as data URLs in a top-level images array of the /v1/systemone request. Our image results use the BF16
projector; the smaller Q8_0 projector from the same repo was not tested with these files.
This is a decision model. Run it on llama-server's /v1/systemone endpoint and keep -ub 4096 -b 4096, as in
Run it. The llama.cpp, Ollama and LM Studio buttons under "Use this model" start it as a chat model
without those settings, which is not a supported way to run it.
Download Clef-Artemis-IQ4_XS.gguf (14.82 GB, peak GPU memory 15.0 GiB for text on an NVIDIA A100)
and start it with the settings below, then send requests as shown in Run it. It gives the same answer as
the original on 95.6% of 2,000 test decisions and changed no answer the original was more than 80% sure of. Go smaller
only if memory is tight: the two smallest files changed 2 and 4 such answers.
These files are per-tensor mixes, and their labels are size classes (see the file names table at the end). Use the exact filename or the :IQ4_XS tag; without a tag, app buttons may select a different file.
Cloudflare/clef is 54.1 GB and needs 49.9 GiB of GPU memory, so an 80 GB data-centre card. The default file here is 14.82 GB, 73% smaller, and needs 15.0 GiB, so it should fit a single 16 GB GPU for text (measured on an NVIDIA A100). It gives the same answer as the original on 95.6% of 2,000 test decisions, has measured accuracy (72.0% against 72.1%), and changed no answer the original was more than 80% sure of.
The separate train-split results are labelled held-out in the tables. Their independence from calibration and variant selection has not been verified for every file.
Test results informed candidate selection, so they do not independently measure performance on new inputs. Read them alongside the separate train-split results and public-file comparisons. Agreement measures whether a compressed model keeps the original's answers, including its mistakes. The typed-decisions accuracy figures compare answers with teacher-generated labels.
Every file was compared with the BF16 original and with the public builds, all at identical llama.cpp settings, on:
Measured with llama.cpp 1e6f04a at the settings above, every file on one NVIDIA A100 80 GB. Public files:
ggml-org/Clef-GGUF (d2aa75c) and bartowski/Cloudflare_clef-GGUF (9d3b28a).
| Check | BF16 | Artemis IQ4_XS | Artemis IQ3_M | Artemis Q3_K_M | Artemis IQ3_XXS | Artemis Q2_K | Public Q8_0 | Public Q4_K_M | Public IQ3_M |
|---|---|---|---|---|---|---|---|---|---|
| Typed-decision accuracy | 72.1 | 72.0 | 72.8 | 72.4 | 73.2 | 73.0 | 72.6 | 72.2 | 72.6 |
| Same answer as BF16 | 95.6 | 95.9 | 93.4 | 91.0 | 89.5 | 98.8 | 96.9 | 87.6 | |
| Held-out accuracy | 74.9 | 73.7 | 74.2 | 73.7 | 74.3 | 74.8 | 74.7 | 74.6 | 73.4 |
| Held-out: same answer as BF16 | 95.4 | 95.6 | 92.2 | 91.8 | 88.3 | 98.1 | 96.2 | 87.1 | |
| Decision Index sample accuracy | 86.7 | 86.3 | 87.7 | 87.3 | 86.7 | 87.0 | 86.7 | 86.3 | 86.0 |
| Decision Index sample: same answer as BF16 | 97.0 | 97.0 | 96.0 | 95.3 | 93.7 | 99.0 | 98.0 | 94.3 | |
| Image accuracy | 73.9 | 73.6 | 73.9 | 73.9 | 73.6 | 72.7 | 73.6 | 74.5 | 70.9 |
| Images: same answer as BF16 | 94.5 | 94.5 | 94.5 | 90.3 | 90.9 | 98.5 | 97.6 | 92.1 | |
| Changed answers where BF16 was over 80% sure | 0 | 0 | 0 | 2 | 4 | 0 | 0 | 0 |
The confident-change row counts text answers; on images the Artemis Q2_K file also changed 1. For every file, shuffling the answer options changed no answers and a repeat run gave identical answers. Clef answers a request's questions jointly, so sending them one per request changes 12.8% of the original's own decisions; every number here uses the same request shape. Small differences in this sample should not be treated as general improvements.
The same files on two more machines, text checks only, with llama.cpp 1e6f04a (CUDA) and
-c 16384 -ub 4096 -b 4096 --parallel 2. Each file is compared with the original run on the same machine. Peak GPU memory is the server
process's own allocation as nvidia-smi reports it.
The GB10 has 128 GB of memory shared by the CPU and GPU. The original ran fully on the GPU with a peak of 49.6 GiB.
| File | Same answer as BF16 | Accuracy change | Confident answers changed | Held-out: same answer | Peak GPU memory |
|---|---|---|---|---|---|
| Clef-Artemis-IQ4_XS | 95.6% | +0.1 | 0 | 95.3% | 14.7 GiB |
| Clef-Artemis-IQ3_M | 95.7% | +0.8 | 0 | 95.3% | 14.5 GiB |
| Clef-Artemis-Q3_K_M | 93.6% | +0.3 | 0 | 92.4% | 13.0 GiB |
| Clef-Artemis-IQ3_XXS | 91.0% | +1.1 | 2 | 91.7% | 12.1 GiB |
| Clef-Artemis-Q2_K | 89.5% | +0.5 | 5 | 88.6% | 11.1 GiB |
| Public Q8_0 (ggml-org) | 99.0% | +0.2 | 0 | 97.9% | 27.1 GiB |
| Public Q4_K_M (ggml-org) | 96.9% | +0.2 | 0 | 95.9% | 18.8 GiB |
| Public IQ3_M (bartowski) | 87.4% | +0.7 | 1 | 87.6% | 13.9 GiB |
Compared with the A100, same answer moves by at most 0.2 points (held-out 0.5), accuracy change by at most half a point, and peak memory is 0.2 to 0.3 GiB lower. The Artemis Q2_K file changed 5 confident answers here against 4 on the A100, and the public IQ3_M changed 1 against 0.
The original does not fit in 32 GB, so it ran with 35 of its 65 layers on the GPU and the rest in system memory. Every other file ran fully on the GPU.
| File | Same answer as BF16 | Accuracy change | Confident answers changed | Held-out: same answer | Peak GPU memory |
|---|---|---|---|---|---|
| Clef-Artemis-IQ4_XS | 96.0% | +0.3 | 0 | 94.2% | 15.1 GiB |
| Clef-Artemis-IQ3_M | 95.7% | +0.4 | 0 | 96.0% | 14.9 GiB |
| Clef-Artemis-Q3_K_M | 93.2% | +0.6 | 0 | 91.1% | 13.4 GiB |
| Clef-Artemis-IQ3_XXS | 90.8% | +1.1 | 2 | 91.3% | 12.5 GiB |
| Clef-Artemis-Q2_K | 89.1% | +0.8 | 5 | 88.4% | 11.5 GiB |
| Public Q8_0 (ggml-org) | 98.9% | +0.1 | 0 | 98.3% | 27.5 GiB |
| Public Q4_K_M (ggml-org) | 97.0% | +0.4 | 0 | 96.7% | 19.2 GiB |
| Public IQ3_M (bartowski) | 87.5% | +0.7 | 0 | 87.4% | 14.3 GiB |
Compared with the A100, same answer moves by at most 0.4 points (held-out 1.2), accuracy change by at most half a point, and peak memory is 0.1 to 0.2 GiB higher. The Artemis Q2_K file changed 5 confident answers here against 4 on the A100.
Peak GPU memory of llama-server on an NVIDIA A100 80 GB with the settings above: text (-ub 4096 -b 4096) and with
the BF16 vision projector loaded (-ub 8192 -b 8192).
| File | Size | Peak GPU memory, text | Peak GPU memory, with images |
|---|---|---|---|
| Original, BF16 | 54.1 GB | 49.9 GiB | 52.3 GiB |
| Clef-Artemis-IQ4_XS | 14.82 GB | 15.0 GiB | 17.4 GiB |
| Clef-Artemis-IQ3_M | 14.31 GB | 14.7 GiB | 17.1 GiB |
| Clef-Artemis-Q3_K_M | 12.72 GB | 13.2 GiB | 15.6 GiB |
| Clef-Artemis-IQ3_XXS | 11.53 GB | 12.3 GiB | 14.7 GiB |
| Clef-Artemis-Q2_K | 10.49 GB | 11.3 GiB | 13.8 GiB |
| Public Q4_K_M (ggml-org) | 19.2 GB | 19.1 GiB | 21.5 GiB |
| Public IQ3_M (bartowski) | 13.78 GB | 14.2 GiB | 16.6 GiB |
So for text, the 14.8 GB file should fit a 16 GB GPU with about 1 GiB to spare; with images it needs more.
File names end in a standard llama.cpp quant label, so llama.cpp, Ollama and LM Studio can pick a file by tag. Each file is a per-tensor mix of quant types, so the label is a size class: the smallest standard llama.cpp quant whose typical bits per weight are at least the file's. The general.file_type field inside the files reads Q4_K_M, the base type given to llama-quantize; the table shows the tensor types each file holds.
| File | Earlier name | Bits per weight | Main tensor types (share of bytes) |
|---|---|---|---|
Clef-Artemis-IQ4_XS.gguf | Clef-Artemis-14.8GB.gguf | 4.38 | IQ4_XS 61%, Q4_K 28%, Q6_K 6%, Q3_K 4% |
Clef-Artemis-IQ3_M.gguf | Clef-Artemis-14.3GB.gguf | 4.23 | IQ4_XS 60%, Q4_K 24%, Q6_K 7%, IQ3_XXS 3% |
Clef-Artemis-Q3_K_M.gguf | Clef-Artemis-12.7GB.gguf | 3.76 | IQ3_XXS 38%, Q4_K 27%, IQ4_XS 24%, Q6_K 8% |
Clef-Artemis-IQ3_XXS.gguf | Clef-Artemis-11.5GB.gguf | 3.41 | Q4_K 30%, IQ4_XS 26%, IQ2_S 16%, IQ2_XS 14% |
Clef-Artemis-Q2_K.gguf | Clef-Artemis-10.5GB.gguf | 3.10 | IQ4_XS 65%, IQ2_XXS 28%, IQ1_M 5% |
The earlier names still download the same bytes, with the same checksums, for a transition period.
| File | Bytes | sha256 |
|---|---|---|
Clef-Artemis-IQ4_XS.gguf | 14,820,180,032 | 1b181894acbf7dd9aaf680463eb8359f35be610b284a5c28f936efa84f1ac55f |
Clef-Artemis-IQ3_M.gguf | 14,313,361,472 | b7ddd85efabff7d2bde9398a30e33eecca416104a922bec4de6602417a7b66da |
Clef-Artemis-Q3_K_M.gguf | 12,719,484,992 | 1d1bb4c68af0094c63fc6fda996b31764b8efe3a420ff90ddc17705bd8d64613 |
Clef-Artemis-IQ3_XXS.gguf | 11,530,211,392 | 5c1de1321fe8005a78d49128b31e620006c43e4fe9db10f22f37af81202cb0b3 |
Clef-Artemis-Q2_K.gguf | 10,488,680,512 | c48104084f6d09c9c84b0cb2d8df443aee3e3443e7e25aeac9ee03033ff0dd7c |
We also make fine-tuned and further optimised versions on request: tuned on specific decisions and data, built for the target hardware (CPU servers, laptops, Apple Silicon, edge devices), or compressed from a private model. Book a call with TurinTech.
Apache-2.0 (see LICENSE and NOTICE), the same as Cloudflare/clef, which is post-trained from
Qwen/Qwen3.8-27B.
When serving these files to others, for example through an API, an inference provider or OpenRouter, we'd appreciate a mention of TurinTech (turintech.ai) as the source of the quantization.
These files keep improving.