Instructions to use plantwaterer/muria with libraries, inference providers, notebooks, and local apps. Follow these links to get started.
- Libraries
- Transformers
How to use plantwaterer/muria with Transformers:
# Use a pipeline as a high-level helper from transformers import pipeline pipe = pipeline("text-generation", model="plantwaterer/muria") messages = [ { "role": "user", "content": [ {"type": "image", "url": "https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/p-blog/candy.JPG"}, {"type": "text", "text": "What animal is on the candy?"} ] }, ] pipe(text=messages)# pip install -U transformers accelerate # Load model directly from transformers import AutoProcessor, AutoModelForMultimodalLM processor = AutoProcessor.from_pretrained("plantwaterer/muria") model = AutoModelForMultimodalLM.from_pretrained("plantwaterer/muria", device_map="auto") messages = [ { "role": "user", "content": [ {"type": "image", "url": "https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/p-blog/candy.JPG"}, {"type": "text", "text": "What animal is on the candy?"} ] }, ] inputs = processor.apply_chat_template( messages, add_generation_prompt=True, tokenize=True, return_dict=True, return_tensors="pt", ).to(model.device) outputs = model.generate(**inputs, max_new_tokens=256) print(processor.decode(outputs[0][inputs["input_ids"].shape[-1]:])) - Notebooks
- Google Colab
- Kaggle
- Local Apps Settings
- vLLM
How to use plantwaterer/muria with vLLM:
Install from pip and serve model
# Install vLLM from pip: pip install vllm # Start the vLLM server: vllm serve "plantwaterer/muria" # Call the server using curl (OpenAI-compatible API): curl -X POST "http://localhost:8000/v1/chat/completions" \ -H "Content-Type: application/json" \ --data '{ "model": "plantwaterer/muria", "messages": [ { "role": "user", "content": "What is the capital of France?" } ] }'Use Docker
docker model run hf.co/plantwaterer/muria
- SGLang
How to use plantwaterer/muria with SGLang:
Install from pip and serve model
# Install SGLang from pip: pip install sglang # Start the SGLang server: python3 -m sglang.launch_server \ --model-path "plantwaterer/muria" \ --host 0.0.0.0 \ --port 30000 # Call the server using curl (OpenAI-compatible API): curl -X POST "http://localhost:30000/v1/chat/completions" \ -H "Content-Type: application/json" \ --data '{ "model": "plantwaterer/muria", "messages": [ { "role": "user", "content": "What is the capital of France?" } ] }'Use Docker images
docker run --gpus all \ --shm-size 32g \ -p 30000:30000 \ -v ~/.cache/huggingface:/root/.cache/huggingface \ --env "HF_TOKEN=<secret>" \ --ipc=host \ lmsysorg/sglang:latest \ python3 -m sglang.launch_server \ --model-path "plantwaterer/muria" \ --host 0.0.0.0 \ --port 30000 # Call the server using curl (OpenAI-compatible API): curl -X POST "http://localhost:30000/v1/chat/completions" \ -H "Content-Type: application/json" \ --data '{ "model": "plantwaterer/muria", "messages": [ { "role": "user", "content": "What is the capital of France?" } ] }' - Docker Model Runner
How to use plantwaterer/muria with Docker Model Runner:
docker model run hf.co/plantwaterer/muria
muria
Licence: the weights and the code in this repository are released under Apache-2.0.
muria is a fine-tune of google/gemma-4-12B-it that answers multiple-choice
decision questions with a probability for each option. It includes a calibrated last-layer head (vbll_head.safetensors,
vbll_head.json) whose output is the answer. For ordinary multiple-choice questions that output is the softmax of the option
letters' logits at the answer position, divided by one fitted temperature (T = 0.970); no posterior is involved. For yes / no
questions, a Bayesian last-layer residual on the model's hidden state (a Laplace posterior, integrated over) adjusts the two
logits.
Two parts of the model are not optional:
- Attention. In the full-attention layers, each prompt's tokens attend to every token of the same prompt; the sliding-window
layers stay causal (window 1,024). transformers' stock attention would run the full-attention layers causally and give different
answers, so the model is served with the
jev_memattention function infullft_mem.py. It registers with transformers' attention interface and runs on PyTorch SDPA (memory-efficient and flash kernels). The engine refuses any other attention. - The head. The letter logits on their own are not the model's answer.
The weights (model-*.safetensors, vbll_head.safetensors) are hosted here on the Hugging Face Hub; the GitHub mirror,
treewaterer/muria, holds the code and this card only.
Usage
The repository includes a Decision Index engine, dikit_engine.py, with
garas_vbll_engine.py, vbll_core.py, fullft_mem.py, garas_engine.py and garas_core.py. It needs the Decision Index kit:
pip install "git+https://github.com/apolinario/decision-index@9eb2dbe2a358004c8782c66e40a83ac07b953fec"
REPO=$(python -c "from huggingface_hub import snapshot_download; print(snapshot_download('plantwaterer/muria'))")
PYTHONPATH="$REPO" python -m decision_index run --engine dikit_engine:DikitEngine --option model="$REPO" --compact --out runs/muria
In Python:
import sys
from huggingface_hub import snapshot_download
repo = snapshot_download("plantwaterer/muria")
sys.path.insert(0, repo)
from dikit_engine import DikitEngine # needs the decision_index kit installed
engine = DikitEngine(model=repo) # bf16 on "cuda"; the head runs on the CPU
response, _ = engine(
"The forecast gives a 70% chance of heavy rain this afternoon. The match is outdoors, on grass.",
{"plan": {"type": "choice", "instructions": "What should the organisers do?",
"criteria": {"play": "Play as planned", "indoors": "Move it indoors", "postpone": "Postpone it"}},
"rain": {"type": "noul", "instructions": "Is heavy rain likely?"}})
print(response["answers"])
The inference procedure is as follows:
- Prompts: one prompt per question, with the model's chat template and the options lettered A-Z a-z. A choice question's
options are shown in one fixed non-identity order derived from a hash of the request. Yes / no questions use the options
YesandNo; when the request gives their definitions (itscriteria), they are shown asYes: <definition>/No: <definition>. - Read-out: one forward pass per prompt. The final hidden state and the option letters' logits (fp32, with the model's logit soft-cap) go through the head, on the CPU in float64, which gives the probabilities.
- Batching: the prompts of one request run in packed rows of up to 32,768 tokens, each prompt with its own positions, so a prompt only attends to itself.
- More than 52 options: handled in two rounds.
- No truncation: a request that does not fit 32,768 tokens is reported as unsupported.
Requirements: one GPU (the bf16 weights are 24 GB, and the packed rows add activations for up to 32,768 tokens). Tested on 1 x H100 80GB and 1 x RTX PRO 6000 (96 GB) with torch 2.14.1 (CUDA 13.0), transformers 5.12.1, huggingface_hub, safetensors and numpy. bf16 results differ slightly between GPU types.
Evaluation
Decision Index 0.3, public panel (37 benchmarks). The run used this repository's weights, head and engine, with the kit's runner and scorer at commit 9eb2dbe2, in bf16. The full public suite ran as 8 row shards on single GPUs (shards 0, 1, 2 and 7 on H100 80GB, 3 to 6 on RTX PRO 6000), merged without changes. This is our own run of the public kit, not a leaderboard entry. The run files are at plantwaterer/muria-decision-index.
| public index | knowledge | language | retrieval | tools | arts | |
|---|---|---|---|---|---|---|
| board rule (the 3 test items listed below counted as wrong) | 51.40 | 31.41 | 56.62 | 58.90 | 70.82 | 39.06 |
| as scored | 51.40 | 31.41 | 56.62 | 58.90 | 70.82 | 39.06 |
140,620 requests were run and 140,178 scored after the official exclusions; all of them were answered (0 unsupported, 0 errors). Nothing was truncated. The two rows are equal at two decimals: one of the 3 listed items was answered correctly, and counting it as wrong moves ANLI's score from 0.7171 to 0.7168.
Latency (our approximation of the board's check, which uses a private sample): the kit runner's per-request wall time on the four RTX PRO 6000 shards of the run above, one request at a time, 70,309 answered requests:
| median | mean | 80th percentile |
|---|---|---|
| 48 ms | 209 ms | 169 ms |
Overlap with evaluation data
No test split of any benchmark was used to build the training data. The head's fit and calibration data include train-split rows of CLINC150, GSM8K, Amazon ESCI, BANKING77 and WinoGrande, and units from DocNLI's training split, some of which contain premise-hypothesis pairs from ANLI (see the note at the end of this card); the training data include Lichess puzzles (a different dataset from ChessBench's).
An audit of the training text, the head's data included, against the Decision Index public suite found 3 ANLI test items that are
near-duplicates of ANLI pairs inside DocNLI units in the head's fit data: 12:ANLI:ANLI:test_r2:5460dddd-0280-47bd-9148-1422b6729a4d,
12:ANLI:ANLI:test_r2:f6a3b2f4-863e-41af-bde1-762da7822ebd and 12:ANLI:ANLI:test_r3:d469aabc-dd2a-4bd7-b840-ab39764b6625. They are
counted as wrong in the board-rule number above, as the board does for test items found in a model's training data (see the
Contamination section of the Decision Index methodology).
Licence
The weights and the head are released under Apache-2.0, and the base model, google/gemma-4-12B-it, is Apache-2.0. The training data come from the sources below, under the licences and terms listed. Some of these sources carry use conditions that apply to the data itself; they are repeated here: US Census PUMS (never use the data to identify any person or household), NVD (not endorsed or certified), California DMHC (modified; not official government data), MITRE CVE / CWE (copyright notices below), XNLI (CC BY-NC 4.0, non-commercial; its translations are part of AmericasNLI), and, in the head's fit and calibration data only, ANLI (CC BY-NC 4.0, non-commercial; its premise-hypothesis pairs are part of DocNLI) and wikiHow (CC BY-NC-SA 3.0, non-commercial; via goal-step-wikihow). The note at the end of this card gives the counts.
Data attribution
muria was trained on decision questions built from the public data sources below, plus synthetic questions generated by the author (no third-party data). Thanks to everyone who builds and maintains them. Records were reformatted, and many were modified (fields selected, text clipped, identifiers masked), so none of the material should be read as an official copy of its source.
Open data and public records
- GDELT Project: event data. Data from The GDELT Project, https://www.gdeltproject.org/.
- Wikipedia: article text from the English Wikipedia dump of 2023-11-01, via
wikimedia/wikipedia
20231101.en, by Wikipedia contributors. Licensed CC BY-SA 3.0/4.0 and GFDL. https://en.wikipedia.org/ - Wikipedia Articles for Deletion: deletion debates by English Wikipedia contributors (CC BY-SA 3.0/4.0, GFDL), via the AfD corpus of Mayfield & Black (2019), https://github.com/emayfield/AFD_Decision_Corpus (the repository's code is GPL-3.0; the corpus files are distributed separately).
- Wikidata: structured data (CC0 1.0), and Requests-for-deletion discussions by the Wikidata contributors (CC BY-SA 4.0). https://www.wikidata.org/
- Swiss Judgment Prediction XL: rcds/swiss_judgment_prediction_xl (Rasiah et al., 2023), CC BY-SA 4.0. Decisions of the Swiss Federal Supreme Court, anonymised by the court.
- NHTSA Office of Defects Investigation investigations and complaints (public domain), https://static.nhtsa.gov/odi/ffdd/
- OSHA Severe Injury Reports and enforcement data, US Department of Labor (public domain), https://www.osha.gov/severe-injury-reports
- NTSB aviation accident database (public domain), https://data.ntsb.gov/avdata (Added 2026-10-22: the OSHA and NTSB records above were in this model's training and head data but missing from this list.)
- openFDA enforcement reports and device recalls, CC0 1.0, https://open.fda.gov/
- ClinicalTrials.gov (US National Library of Medicine), via the ClinicalTrials.gov API v2; data processed by ClinicalTrials.gov on 2026-10-01. Modified: selected registration fields only, free text clipped, trial ids masked, study-status sentences removed. Some ClinicalTrials.gov data may be subject to third-party copyright outside the United States.
- ClinVar (NCBI): Landrum et al., Nucleic Acids Research 2018, PMID 29165669.
- Orphanet inheritance data, CC BY 4.0, https://www.orpha.net/
- CIViC (civicdb.org), CC0 1.0: Griffith et al., Nature Genetics 2017, PMID 28138153.
- California Department of Managed Health Care, "DMHC IMR Data, 2001 - Current", CC BY, https://data.ca.gov/dataset/independent-medical-review-imr-determinations-trend. Modified; not official government data.
- NOAA / NWS Storm Prediction Center text products (public domain), via the Iowa Environmental Mesonet, Iowa State University.
- LIGO / Virgo / KAGRA public alerts and the GWOSC event catalogue, CC BY 4.0, by the LIGO Scientific, Virgo and KAGRA Collaborations. This model uses data obtained from the Gravitational Wave Open Science Center (gwosc.org), a service of the LIGO Scientific Collaboration, the Virgo Collaboration and KAGRA.
- USGS PAGER and ComCat (public domain). Population exposure is from ORNL LandScan (CC BY 4.0).
- National Vulnerability Database (NIST, public domain) and CVE records (CVE Terms of Use; Copyright The MITRE Corporation). CWE: Copyright 2006-2026, The MITRE Corporation; CWE is a trademark of The MITRE Corporation. This product uses the NVD API but is not endorsed or certified by the NVD.
- American Community Survey PUMS, US Census Bureau (CC0 1.0), with task definitions from folktables (Ding et al., 2021; MIT). Never to be used to identify any person or household.
- OpenReview reviews and scores, CC BY 4.0 (titles and abstracts: CC0 1.0), https://openreview.net/
- Lichess open database, CC0 1.0, https://database.lichess.org/
- Amazon ESCI shopping queries, amazon-science/esci-data (Reddy et al., 2022), Apache-2.0. Copyright Amazon.com, Inc. or its affiliates.
- SWE-rebench: nebius/SWE-rebench-openhands-trajectories (Trofimova et al., 2025), CC BY 4.0.
- Open-Jev: ZefanCai/Open-Jev, CC0 1.0.
- LLM Bayesian Teaching Data: Qiu, L. (2025), Zenodo, doi:10.5281/zenodo.17677329, CC BY 4.0.
Research datasets
| dataset | reference | licence |
|---|---|---|
| BANKING77 | Casanueva et al., 2020 | CC BY 4.0 |
| CLINC150 | Larson et al., 2019 | CC BY 3.0 |
| GoEmotions | Demszky et al., 2020 | Apache-2.0 |
| GSM8K | Cobbe et al., 2021 | MIT |
| HelpSteer2 | Wang et al., 2024 | CC BY 4.0 |
| Measuring Hate Speech | Kennedy et al., 2020 | CC BY 4.0 |
| IBM Debater ArgQ-Rank-30kArgs | Gretz et al., 2019 | CC BY 3.0 |
| BBQ | Parrish et al., 2022 | CC BY 4.0 |
| CaSiNo | Chawla et al., 2021 | CC BY 4.0 |
| Circa | Louis et al., 2020 | CC BY 4.0 (dataset card; its README links CC BY-SA 4.0) |
| CodeContests | Li et al., 2022 | CC BY 4.0 |
| CosmosQA | Huang et al., 2019 | CC BY 4.0 |
| CommonsenseQA | Talmor et al., 2019 | MIT |
| e-CARE | Du et al., 2022 | MIT |
| Social IQa | Sap et al., 2019 | CC BY 4.0 |
| WinoGrande | Sakaguchi et al., 2019 | CC BY (code Apache-2.0) |
| MedMCQA | Pal et al., 2022 | Apache-2.0 |
| PubMedQA | Jin et al., 2019 | MIT |
| Stanford Politeness Corpus, via ConvoKit | Danescu-Niculescu-Mizil et al., 2013 | CC BY 4.0 |
| Scruples | Lourie et al., 2021 | Apache-2.0 |
| TabFact | Chen et al., 2020 | CC BY 4.0 |
| QASPER | Dasigi et al., 2021 | CC BY 4.0 |
Through tasksource (Sileo, 2024), each dataset under its own licence as recorded on its dataset page:
Community synthetic datasets
- tasksource/procedural-typed-decisions (tasksource, Sileo, 2024), Apache-2.0
- frontier-infra/jebadiah-synth-v2, Apache-2.0
- kaivoss/system-one-270m-data, Apache-2.0
- vagmi/jevlite_dataset, CC BY-SA 4.0
- mghafiri/decision-model-scenarios, MIT
Non-commercial data in the head
The head's fit data (16,384 decision questions) and calibration data (4,086) include two sources whose text is non-commercial. The model's fine-tuning data contain neither.
- DocNLI package (BSD-3-Clause, Yin et al., 2021), via tasksource/doc-nli: 70 units in the fit data and 17 in the calibration data. 10 of them (8 fit, 2 calibration) contain a complete premise-hypothesis pair from upstream ANLI (CC BY-NC 4.0, Nie et al., 2020), and 23 (21 fit, 2 calibration) contain ANLI text, from ANLI rounds 2 and 3. Some of these units contain near-duplicates of 3 ANLI test items on the Decision Index suite; the items are listed under Overlap with evaluation data and counted as wrong in the board-rule score.
- wikiHow, via tasksource/goal-step-wikihow (MIT, the package): 81 units in the fit data and 9 in the calibration data. Their text is from wikiHow (CC BY-NC-SA 3.0).
No dataset text is redistributed.
- Downloads last month
- 216