Sieve-9B-v2

Sieve-9B-v2 is a decision model from the Sieve family. You give it a piece of text or JSON (the state) and typed questions, and it returns a calibrated probability for every option of every question. It reads the state once, scores the options directly, and never generates text.

It is a LoRA adapter (rank 32) and a pointer head on Qwen/Qwen3.5-9B (revision c2022362), and it stores one calibration temperature per question category. The code is at github.com/sthanika-ai/Sieve.

It was first published as Sieve-9B-Plus. Its training and evaluation records (training_config.json, training_metrics.json, train.log, result.json) and the Decision Index run files keep that name.

How it works

• Backbone. Only the text part of Qwen3.5-9B is used; the model cannot produce text.
• Readout. A pointer head scores the end of each option against the decision point. A calibrated softmax gives the probabilities.
• Isolation. Each question runs as its own row from the state's cache, so questions never see each other.

type	options	returns
choice	the caller's keys, optionally described (up to 255)	a probability per key
noul	yes / no	P(yes)
score	ordered levels	a probability per level

Use

pip install "sieve-decisions[cuda,serve] @ git+https://github.com/sthanika-ai/Sieve"
pip install --no-build-isolation causal-conv1d==1.7.0   # recommended: the fused kernel our runs used

from sieve import load_sieve, decide

m = load_sieve("sthanika-ai/Sieve-9B-v2")
decide(m, {"subject": "Duplicate charge on invoice 4411", "body": "Billed twice. Refund today or we cancel."},
       {"department": {"type": "choice", "instructions": "Which team handles this?",
                       "criteria": {"billing": "invoices, payments, refunds", "technical": "bugs and outages"}},
        "churn_risk": {"type": "noul", "instructions": "Does the customer threaten to cancel?"}})

• Loading. Load it with sieve 0.2.0 or later; the adapter alone is not a text-generation model. The backbone is downloaded at its pinned revision on first use.
• GPU memory. About 20 GB in bf16. Longer states need more.
• Token limits. States and questions up to 16,384 tokens each, as in training. sieve 0.2.1 or later uses these limits by default; 0.2.0 stops at 8,192 tokens. Longer requests are refused, never truncated. The Decision Index engine declares larger limits (a 65,536-token state and a 32,768-token question). In the Decision Index run below, 241 requests had a question longer than 16,384 tokens (at most 29,944, from ToolRet and BRIGHT) and were answered in full; no state was longer than 16,384.
• Server. sieve-serve --model sthanika-ai/Sieve-9B-v2 --graphs.
• causal-conv1d. Without it, transformers falls back to a slower PyTorch convolution, whose probabilities can differ slightly. Install it to reproduce our numbers.
• Temperature per category. head.pt stores a global T = 1.109 and 23 temperatures (1.000 to 1.338), fitted by NLL on the calibration split (43,176 questions). A question's category is its type, its number of options (2, 3-4, 5-9, 10-25, 26+) and the kind of state (empty, text or JSON). A question takes the temperature of its most specific category in the table (choice:3-4:text, then choice:3-4, then choice), else the global T. Temperatures never change an answer, and temperature=1.0 gives raw probabilities. sieve versions before 0.2.0 ignore the table and use the global T.

Results

Decision Index 0.3

Sieve-9B-v2 scores 53.35 (raw index 64.39, breadth 51.95) on the public 0.3 suite of the Decision Index, on our own run with the official kit at 62d2f51 (0.3) and its scorer.

• Coverage. All 140,178 scoreable requests over 43 benchmarks were answered, with no truncation, no unsupported request and no error. The index averages 37 of the benchmarks.
• How the run was made. One complete run of the 0.3 suite on two H100 80GB GPUs, in four shards.
• Results. sthanika-ai/Sieve-9B-v2-decision-index-results.
• Not on the leaderboard yet. The score is self-reported. On the 0.3 board the ranking uses the Full score, where this public suite counts 20% and the maintainers' private tests count 80%.

Decision Index 0.3	Knowledge & Reasoning	Language Understanding	Retrieval & Classification	Tools & Automation	Arts & Human Taste
53.35	38.3	55.9	64.2	69.8	34.0

The index and area scores are chance-corrected: 0 is random guessing and 100 is perfect. Below, the score is in each benchmark's own metric and the skill is chance-corrected.

All 43 benchmarks

benchmark	metric	score	skill (×100)	requests
ACOS	per-review F1	0.175	14.8	1,565
Amazon ESCI	macro-F1	0.499	37.2	5,000
ANLI	macro-F1	0.651	47.7	3,200
API-Bank	accuracy	0.858	85.6	508
BANKING77	macro-F1	0.860	85.8	3,080
BBH fixed-option tasks	accuracy	0.703	57.0	5,507
BFCL	case exact accuracy	0.958	94.3	1,694
BPoMP	accuracy	0.882	77.0	5,000
BRIGHT	nDCG@10	0.451	37.9	220
cfcolor	accuracy	0.623	26.4	5,000
ChessBench	accuracy	0.128	5.1	5,000
CLadder	accuracy	0.669	33.9	5,000
CLINC150+OOS	macro-F1	0.912	91.1	5,500
ContractNLI	macro-F1	0.734	61.6	123
CRUXEval	accuracy	0.589	34.9	570
FinEntity	macro-F1	0.889	83.8	979
GPQA Diamond	accuracy	0.464	28.6	196
GSM8K	accuracy	0.687	62.4	2,638
Habermas Machine	accuracy	0.403	13.4	1,676
HellaSwag	accuracy	0.967	95.6	10,042
HLE	accuracy	0.100	0.0	501
Home appliance simulator	case exact accuracy	0.250	25.0	88
HoVer claim verification	accuracy	0.802	60.5	4,000
Humicroedit	accuracy	0.614	22.8	2,628
iSarcasmEval	Sarcasm F1 · track A, English	0.549	42.1	4,600
MMLU-Pro	accuracy	0.594	54.3	12,032
MuSR	accuracy	0.560	30.0	752
New Yorker caption matching	accuracy	0.659	57.4	528
NLI4CT	macro-F1	0.792	59.6	5,500
PhishNChips phishing decisions	accuracy	0.841	68.1	2,000
POP909-CL	accuracy	0.084	7.1	2,000
RAGTruth response-level hallucination	F1 on hallucinated class	0.802	59.0	2,700
SATA-Bench	case exact accuracy	0.279	26.9	1,650
ToolRet	nDCG@10	0.673	62.3	685
VAST	macro-F1	0.554	33.2	3,006
When2Call MCQ	accuracy	0.801	73.5	3,652
WinoGrande	accuracy	0.908	81.7	1,267
ARC-Challenge (not in the index)	accuracy	0.954	93.9	1,172
ARC-Easy (not in the index)	accuracy	0.981	97.5	2,376
MMLU (not in the index)	accuracy	0.787	71.6	14,033
RouterBench (not in the index)	selected quality (quality objective)	0.795	51.9	10,000
SGD/SGD-X (not in the index)	macro-F1	0.438	6.5	2,500
SimpleBench (not in the index)	accuracy	0.100	0.0	10

Calibration split

The calibration split chose the checkpoint and fitted the temperatures.

split	questions	accuracy	NLL	Brier	ECE
calibration, per-category temperatures (shipped)	43,176	0.8775	0.3071	0.1676	0.0068
calibration, one global T = 1.109	43,176	0.8775	0.3077	0.1678	0.0074
calibration, raw (T = 1)	43,176	0.8775	0.3091	0.1677	0.0093

In a 2-fold check inside the calibration split (fit on one half, score the other), the per-category table and the single T scored the same: NLL 0.3079 vs 0.3078, ECE 0.0087 vs 0.0084.

Latency

• Per Decision Index request. Measured on one H100 80GB GPU, single process, eager and in-process, over a random 3,000-request sample of the public 0.3 suite: median 43 ms, mean 138 ms, 80th percentile 58 ms, 95th percentile 144 ms. The answers on that sample matched the full run exactly.

Training

• Recipe.
  • LoRA r=32, α=64, dropout 0.05 on the attention, MLP and linear-attention projections (12 modules), with the pointer head trained from scratch.
  • AdamW, lr 5e-5 for the adapter and the head, OneCycle with 10% warm-up, 1 epoch, 32 records per step, bf16.
  • States and questions up to 16,384 tokens each; longer records are dropped, never truncated (none were).
  • 5,183 steps, 5.6 hours on two H100 80GB GPUs.
  • Permuted choice options and "none of the above" augmentation, with a cross-entropy loss.
• Selection. The checkpoint with the highest accuracy on a fixed 2,000-question calibration sample, with NLL breaking ties. This rule was fixed before training, and it chose step 4,500. The benchmarks were never used to choose.
• Temperatures. Fitted after selection on the whole calibration split: the global T, then one T per category with at least 200 calibration questions.
• Data. 165,832 training records and 30,722 calibration records.
  • Sources: public datasets and generated decision records.
  • Not published. The training files are not part of this release; the SHA-256 of each file is in training_config.json.

Files

file	contents
adapter_model.safetensors, adapter_config.json	the LoRA adapter (rank 32, PEFT format)
head.pt	the pointer head and the calibrated temperatures (global and per category)
tokenizer.json, tokenizer_config.json	Qwen3.5-9B's tokenizer, unchanged
training_config.json, training_metrics.json, train.log	the recipe, the run and its log
result.json	the calibration, Decision Index and latency numbers above, with the calibration metrics per category
provenance.json	SHA-256 of every file and of the training data, and library versions

License

Apache-2.0, see LICENSE.