turintech/pplx-decider-27B-Artemis-GGUF

TurinTech models, built with Artemis

Perplexity's pplx-decider 27B for typed decisions about text or images on a 24 GB GPU. It answers typed questions with a probability for each answer, in one pass.

• 15.28 GB download for the default file, pplx-decider-27B-Artemis-Q4_K_S.gguf.
• Peak GPU memory 16.6 GiB for text and 18.9 GiB with images on an NVIDIA A100 80 GB; the original needs 50.8 GiB.
• Same answer as the original on 96.8% of 2,000 test decisions, against 95.7% for a stock Q4_K_M we made (16.55 GB); 0 confident answers changed. These decisions also chose the file.

Run it

This is a decision model: send typed questions to llama-server's /v1/systemone endpoint, not to chat. Needs llama.cpp from 9 October 2026 or newer. Tested at commit 1e6f04a, and both files also on release b11572 (12 October 2026), which gave identical answers on a 100-decision check.

llama-server -hf turintech/pplx-decider-27B-Artemis-GGUF:Q4_K_S --port 8080 -c 16384 -ub 4096 -b 4096 --no-mmproj

Request example, images, Ollama and LM Studio

In Hugging Face's generated "Use this model" commands, :BF16 and :Q8_0 name the vision projector files, not model weights. Use :Q4_K_S or the exact filename below.

llama-server -m pplx-decider-27B-Artemis-Q4_K_S.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: the numbers on this card were measured with them, and the batch size moves the probabilities slightly. In our test on an NVIDIA RTX 5090, pplx-decider-27B-Artemis-Q4_K_S.gguf gave 0.876 at the default of 512 and 0.897 at 4096 on a 923-token input. Inputs longer than the batch size are accepted: with llama.cpp 1e6f04a and b11609, a 4,983-token input was answered at both settings. -c 16384 stops the server from reserving memory for the full training context.

For concurrent clients, an optional configuration is -np 8 -c 32768 --cache-ram 0: eight request slots with 4,096 tokens each and the host prompt cache disabled. In our NVIDIA RTX 5090 test with llama.cpp fcac190, this used about 1.5 GiB more GPU memory than the command above. Measure the configuration on the intended workload before choosing it.

llama-server -m pplx-decider-27B-Artemis-Q4_K_S.gguf --port 8080 -np 8 -c 32768 -ub 4096 -b 4096 --cache-ram 0

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?"}}
}'

Questions, problems or a model to consider for a smaller version? Reply in our feedback thread.

Vision inputs

Perplexity does not publish a llama.cpp vision projector, so this repository includes one converted from the original weights (mmproj-pplx-decider-27B-BF16.gguf). Load it with --mmproj and raise the batch size, since several large images can exceed 4,096 tokens:

hf download turintech/pplx-decider-27B-Artemis-GGUF mmproj-pplx-decider-27B-BF16.gguf --local-dir .
llama-server -m pplx-decider-27B-Artemis-Q4_K_S.gguf --port 8080 --mmproj mmproj-pplx-decider-27B-BF16.gguf -c 16384 -ub 8192 -b 8192

Send images as data URLs in a top-level images array of the /v1/systemone request. A smaller Q8_0 projector (mmproj-pplx-decider-27B-Q8_0.gguf, 0.63 GB) is also included; with the original model it gives the same answer as the BF16 projector on 98.8% of our image questions. Our image results use the BF16 projector.

Recommended Ollama package

echo "My Stripe integration keeps failing. Please help ASAP." | ollama run turintech/pplx-decider-27b-artemis:15.3gb --ask "Does this message express urgency?"

This verified tag selects pplx-decider-27B-Artemis-Q4_K_S.gguf. Ollama 0.40.4 was tested; include "model": "turintech/pplx-decider-27b-artemis:15.3gb" in the decision request to port 11434. The explicit size tag selects the recommended file and is not an immutable revision.

LM Studio

The model is on the LM Studio Hub as turintech/pplx-decider-27b-artemis. That entry loads it as a decision model with the settings above: a 16,384-token context and batch sizes of 4,096. Keep the @q4_k_s label when you download it, because without one LM Studio can pick a different file for your hardware:

lms get turintech/pplx-decider-27b-artemis@q4_k_s --gguf
lms load turintech/pplx-decider-27b-artemis

Then, with LM Studio's server running (port 1234 by default), send decision requests to /v1/systemone and name the model. The chat window and /v1/chat/completions do not run it as a decision model.

curl http://localhost:1234/v1/systemone -H 'Content-Type: application/json' -d '{
  "model": "turintech/pplx-decider-27b-artemis",
  "state": "I was charged twice for the same order.",
  "questions": {"angry": {"type": "noul", "instructions": "Is the customer angry?"}}
}'

Tested on 25 October 2026 with LM Studio 0.4.26 (headless llmster 0.0.25-1) and its CPU llama.cpp engine 2.64.0 on Linux x86-64. pplx-decider-27B-Artemis-Q4_K_S.gguf was downloaded from Hugging Face, its SHA-256 matched, it loaded through a local copy of this Hub definition and /v1/systemone answered a yes/no and a choice question. That is a compatibility check: the numbers on this card come from llama-server.

Only text input was tested in LM Studio.

Choose this model

• Best for typed decisions on text or images with a 24 GB GPU.
• Consider another file when you need the closest match to the original: the stock Q8_0 we made (28.6 GB) gives the same answer 99.2% of the time. No public GGUF of this model runs on stock llama.cpp.

Files

Every file we publish for this model. Sizes are download sizes, not memory.

• With llama.cpp or a Hugging Face pull, add the file's quant label as the tag: -hf turintech/pplx-decider-27B-Artemis-GGUF:Q4_K_S.
• The Ollama tag runs our package, which holds the same file: ollama run turintech/pplx-decider-27b-artemis:<tag>.
• LM Studio's Hub entry turintech/pplx-decider-27b-artemis offers every model file above, without the vision projector.

Without a tag, -hf and hf.co fetch pplx-decider-27B-Artemis-IQ3_M.gguf, not the default file, so always add the tag.

How the files compare

"Saved" is the file-size reduction versus the BF16 original. "Acc. Δ" is the accuracy change against the original on the same cases, 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.

"Same answer as the original" is how often a file picks the same option as the BF16 original on 2,000 test decisions. "Confident answers changed" counts answers the original was more than 80% sure of that the file changed.

As of 13 October 2026, no other public GGUF of this model runs on stock llama.cpp: Livesport's Q5_K_M and Q8_0 use their own runtime and do not load in stock llama-server. So the comparison is with plain Q8_0 and Q4_K_M files we made from the original with stock llama-quantize: the Q8_0 (28.6 GB) gives the same answer 99.2% of the time, the Q4_K_M (16.55 GB) 95.7%. Test and held-out results and file sizes are in the tables. The stock Q4_K_M is closer on the benchmark sample (98.0% against 96.0%).

Results

Our default file against a stock Q4_K_M we made with llama-quantize (no public GGUF of this model runs on stock llama.cpp) and the original, on 2,000 typed decisions from the test split. These decisions were also used to choose the file.

The stock Q4_K_M is larger and closer to the original on the benchmark sample. How it was tested.

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. The plain Q4_K_M comparator was made locally with stock llama-quantize; it is not a public third-party release.

Evaluation details and limitations

How it was tested

Compatibility notes

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 and Ollama buttons under "Use this model" start it as a chat model without those settings, which is not a supported way to run it. The LM Studio button imports the file without those settings; use the Hub entry in LM Studio instead.

Recommendation notes

Download pplx-decider-27B-Artemis-Q4_K_S.gguf (15.28 GB, peak GPU memory 16.6 GiB on an NVIDIA A100, so it should fit a 24 GB GPU) and start it with the settings below, then send requests as shown in Run it. It gives the same answer as the original on 96.8% of 2,000 test decisions and changed no answer the original was more than 80% sure of on the A100 (1 of 2,000 on an RTX 5090). The Artemis IQ3_M file needs 15.9 GiB, which is very tight on a 16 GB GPU (not tested on one).

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 :Q4_K_S tag; without a tag, app buttons may select a different file.

Size and quality trade-offs

perplexity-ai/pplx-decider-v1.1-27b is 53.8 GB and needs 50.8 GiB of GPU memory, so an 80 GB data-centre card. The default file here is 15.28 GB, 72% smaller, and needs 16.6 GiB, so it should fit a 24 GB GPU; the Artemis IQ3_M file needs 15.9 GiB (both measured on an NVIDIA A100). The default gives the same answer as the original on 96.8% of 2,000 test decisions, has measured accuracy (70.8% against 70.5%), and changed no answer the original was more than 80% sure of, in text or images (on the A100; on an RTX 5090, 1 of 2,000).

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 plain Q8_0 and Q4_K_M quantizations, all at identical llama.cpp settings, on:

• 2,000 typed decisions (questions with a fixed answer type: yes/no, pick one, a score) from the test split of LocalLLaMA/typed-decisions: 400 cases with about five questions each,
• 1,000 further decisions from the dataset's train split, reported separately below,
• a 300-item stratified sample of the public Decision Index benchmarks. This is a sample for comparing builds, not a Decision Index score,
• 330 image questions from CV-Bench, RealWorldQA and BLINK, with the same vision projector for every file,
• option shuffling, the longest inputs, many-option questions and repeat runs.

Measured with llama.cpp 1e6f04a at the settings above, every file on one NVIDIA A100 80 GB, images with the BF16 projector.

The original itself changes 8.4% of its answers when the options are shuffled; the compressed files stay close to that. Repeat runs gave identical answers for every file. Small differences in this sample should not be treated as general improvements.

Results on other hardware

Same files on other hardware

The same files on one NVIDIA RTX 5090 32 GB (llama.cpp 1e6f04a, CUDA, -c 16384 -ub 4096 -b 4096 --parallel 2), text checks only. The original does not fit 32 GB and ran with 35 of 65 layers on the GPU.

Within 1.5 points of the A100 on every check. On this GPU the Artemis Q4_K_S file changed 1 answer the original was more than 80% sure of (at 84%), and our two files gave identical answers on 99.6% and 99.2% of a repeat run (100% on the A100).

Memory

Peak GPU memory of llama-server on an NVIDIA A100 80 GB: text (-ub 4096 -b 4096) and with the BF16 vision projector (-ub 8192 -b 8192).

Agreement with the original

The agreement numbers from the tables above, in one fixed layout with the protocol behind each table.

Protocols and standard tables

• Same answer as BF16: the share of decisions where the file picks the same answer as the BF16 original, mistakes included. Higher is closer.
• Mean KL to BF16: KL divergence between the original's and the file's answer probabilities, as the protocol describes. Lower is closer.
• Accuracy: the share of decisions that match the panel's labels. Accuracy change: the file's accuracy minus the original's, in points.
• Confident answers changed: answers the original gave with more than 80% probability that the file changed. Lower is closer.

n/a means the card's tables or our run records do not have that value.

Protocol a100-1e6f04a-20261012

NVIDIA A100 80 GB, 12 October 2026.

• Reference: TurinTech's BF16 GGUF conversion of perplexity-ai/pplx-decider-v1.1-27b
• Panel test: LocalLLaMA/typed-decisions test split, typed decisions; dataset: LocalLLaMA/typed-decisions; split: test; cases: 400; decisions: 2,000; text SHA-256: 4f294f218ea1da27f3efef936359389c62ea4d3973a41457732990f1d31b647c
• Panel heldout: LocalLLaMA/typed-decisions train split, reported as held-out; dataset: LocalLLaMA/typed-decisions; split: train; cases: 200; decisions: 1,000; text SHA-256: 46a58d63edfd86e23229c78afe8b72307bb4ca9fb0e8df180cabb3c67ec9dcd5
• Panel decision_index: 300-item stratified sample of the public Decision Index benchmarks (not a Decision Index score); dataset: MMLU, ARC-Easy, ARC-Challenge, WinoGrande, HellaSwag, ANLI, BANKING77, CLINC150, MMLU-Pro, When2Call; split: per source benchmark; cases: 300; decisions: 300; text SHA-256: 9cb4301cc7d7d96dde9663263c53020912ad628201e5609ab2d65bf16ebb652e
• Panel images: 330 image questions from CV-Bench, RealWorldQA and BLINK; dataset: CV-Bench, RealWorldQA, BLINK; split: val (BLINK); cases: 330; decisions: 330; text SHA-256: 46d96d8d9556847a4fd2aecfa930fc47accc15995b0d1f0d83928dee413cf5b1
• Tool: llama-server /v1/systemone, which returns a probability for every allowed answer; all of a case's questions in one request
• llama.cpp build: b11540 (commit 1e6f04a75e91da4bb66b5248f3fd4bdeff169d41)
• Command: llama-server -m <file.gguf> -c 16384 -ub 4096 -b 4096 --parallel 2 -ngl 99 with our decision suite as the client; image questions with -ub 8192 -b 8192 --mmproj mmproj-pplx-decider-27B-BF16.gguf
• Hardware: one NVIDIA A100 80 GB per file
• Settings: 16,384-token context, batch 4,096, two slots, all layers on the GPU
• Measured by: TurinTech measured every file in these tables with the same build and settings. The comparison files are plain Q8_0 and Q4_K_M quantizations that TurinTech made from the same BF16 with stock llama-quantize and did not publish, because no public GGUF of this model loads in stock llama.cpp
• Selection: the test results informed the choice of files, so they are not an independent test
• Source on this card: the table column "Acc. Δ"; the table rows "Same answer as BF16", "Typed-decision accuracy", "Changed answers where BF16 was over 80% sure", "Held-out: same answer as BF16", "Held-out accuracy", "Decision Index sample: same answer as BF16", "Decision Index sample accuracy", "Images: same answer as BF16", "Image accuracy", "KL to BF16 (lower is closer)"
• Comparison files: TurinTech (unpublished stock quantization), Plain Q8_0; TurinTech (unpublished stock quantization), Plain Q4_K_M

Protocol a100-1e6f04a-20261012, panel test (LocalLLaMA/typed-decisions test split, typed decisions):

Protocol a100-1e6f04a-20261012, panel heldout (LocalLLaMA/typed-decisions train split, reported as held-out):

Protocol a100-1e6f04a-20261012, panel decision_index (300-item stratified sample of the public Decision Index benchmarks (not a Decision Index score)):

Protocol a100-1e6f04a-20261012, panel images (330 image questions from CV-Bench, RealWorldQA and BLINK):

Protocol rtx5090-1e6f04a-20261012

One NVIDIA RTX 5090, 12 October 2026.

• Reference: the same BF16 GGUF conversion, SHA-256 b4f462ea76d66ba2500c0b521964cc763c44796f2bfd352d3a2002a122b56b04; it does not fit 32 GB and ran with 35 of 65 layers on the GPU (-ngl 35 -lm none)
• Panel test: LocalLLaMA/typed-decisions test split, typed decisions; dataset: LocalLLaMA/typed-decisions; split: test; cases: 400; decisions: 2,000; text SHA-256: 4f294f218ea1da27f3efef936359389c62ea4d3973a41457732990f1d31b647c
• Panel heldout: LocalLLaMA/typed-decisions train split, reported as held-out; dataset: LocalLLaMA/typed-decisions; split: train; cases: 200; decisions: 1,000; text SHA-256: 46a58d63edfd86e23229c78afe8b72307bb4ca9fb0e8df180cabb3c67ec9dcd5
• Tool: llama-server /v1/systemone, which returns a probability for every allowed answer; all of a case's questions in one request
• llama.cpp build: b11540 (commit 1e6f04a75e91da4bb66b5248f3fd4bdeff169d41), CUDA
• Command: llama-server -m <file.gguf> -c 16384 -ub 4096 -b 4096 --parallel 2 --fit off -t 128 -ngl 99 (BF16: -ngl 35 -lm none) with two client workers, one file at a time
• Hardware: one NVIDIA RTX 5090 32 GB
• Settings: 16,384-token context, batch 4,096, two slots, text checks only
• Measured by: TurinTech measured every file in these tables with the same build and settings; the plain Q4_K_M is TurinTech's own unpublished stock quantization of the same BF16
• Selection: the test results informed the choice of files, so they are not an independent test
• Source on this card: the table columns "Same answer as BF16", "Accuracy change", "Confident answers changed", "Held-out: same answer"
• Comparison files: TurinTech (unpublished stock quantization), Plain Q4_K_M, SHA-256 240624d15a99e51d73cc1943efade044d61fe029c4e99b74db897d9844e3ebe3

Protocol rtx5090-1e6f04a-20261012, panel test (LocalLLaMA/typed-decisions test split, typed decisions):

Protocol rtx5090-1e6f04a-20261012, panel heldout (LocalLLaMA/typed-decisions train split, reported as held-out):

File names, sizes and SHA-256 checksums

File names

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.

The earlier names were removed from the repository on 26 October 2026; the files under the names above hold the same bytes.

File checksums

Custom versions

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.

About and licence

pplx-decider is Perplexity'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 Perplexity. The files are standard GGUF and run on stock llama.cpp. For images, use the vision projector in this repository (see below). Changes from the original: the weights are converted to GGUF and quantized for llama.cpp. See NOTICE.

Apache-2.0 (see LICENSE and NOTICE), the same as perplexity-ai/pplx-decider-v1.1-27b, which is post-trained from Qwen/Qwen3.8-27B. The original's NOTICE is reproduced in ours.

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.