Decision model
KnowLine-2B-Gen1
1 question per request
- Accuracy, all 2,198 questions
- 36.8%34.8% to 38.8%, 95% interval; 809 right
- Answers kept at 99% confidence
- No answers at this thresholdnone of the 2,198 questions at 99% or more
- Cost, $ per 1,000 questionsper 1M questions
- No list priceBasis and price under Cost
- Latency
- 0.062 smedian request, 1 question per request
How we called it
- Kind
- Decision model
- Access
- KnowLine server (knowline_server.py over SGLang 0.5.21, PelaAI's release code), self-hosted, BF16
- Paid as
- Our own hardware
- Model ID
PelaAI/KnowLine-2B-Gen1- Parameters
- 2B dense, as the maker states it (counted from the released files: 2.27B).language model (tied embeddings) + multi-token prediction layer + 0.33B vision encoder; source: counted from the checkpoint's safetensors, PelaAI/KnowLine-2B-Gen1@53245d4, huggingface.co/PelaAI/KnowLine-2B-Gen1; read 11 November 2026
- Hardware
- 1 × DGX Spark, GB10 (128 GB unified memory); BF16
- Note
- PelaAI's released KnowLine-2B-Gen1 (PelaAI/KnowLine-2B-Gen1 at 53245d4), a merged LoRA fine-tune of Qwen3.5-2B with its readout temperature (T = 1.32) folded into the weights. It ran on our hardware (one DGX Spark), one request at a time, through the card's serve_knowline.sh (SGLang 0.5.21 in BF16 and the repository's knowline_server.py, which reads a request of up to 3,000 tokens twice), with vision on: its vision tower is the base model's, not trained on images. Its card lists English and Chinese.
- Confidence from
- Option score
- Run
- 1079, 13 November 2026
Settings
- Reasoning
- off
- Request format
- 1 question per request
- Requests at once
- 1
- Temperature
- not recorded
Run record: harness seed, harness commit and prompt version
- Harness seed
- 0
- Harness commit
39a6b9a3f536- Prompt version
254597b8be16
Results
| Questions | Accuracy | 95% interval | Right |
|---|---|---|---|
| All questions | 36.8% | 34.8% to 38.8% | 809 of 2,198 |
| Reading | 54.3% | 50.4% to 58.0% | 356 of 656 |
| Grammar | 29.4% | 27.2% to 31.7% | 453 of 1,542 |
Comparison group: μξ. Compare with another setup.
Accuracy by topic
English
The Text tab's 547 kept reading questions translated into English, asked with this setup's Text tab settings. Every setup: English reading; against Turkish: English and Turkish.
- Right answers
- 390 of 547, first English run (1057)
- Accuracy
- 71.3%
- Non-hallucination with “I don't know”
- 22.0%: 38 “I don't know” of 173 not right (run 1059)
- Answered with “I don't know”
- 93.1%: 509 of 547
- Error rate with “I don't know”
- 26.5%: 135 of 509 answers
- Cost, $ per 1,000 questionsper 1M questions, first English run
- No list price
- Latency, median request, first English run
- 0.062 s; 1 request at once
- English run
- 1057
- Machine, English runs
- DGX Spark
- Date
- 12 November 2026
- Request format
- 1 question per request
- Reasoning
- off
- Prompt version
47f5776e71db- Translations, SHA-256
bf0d47cd385a93050c65906977e199ef7e45d5848dbbdef2bc8d124f4d5abf68
Confidence
- Answers kept at 99% or more
- No answers at this threshold; 2,198 skipped, the model less than 99% sure
- Calibration gap
- 8.4 points
- AUROC, all questions
- 0.66 0.64 to 0.68
- AUROC, reading
- 0.73 0.69 to 0.77
- AUROC, grammar
- 0.58 0.55 to 0.61
- Brier score
- 0.221
- Accuracy of its surest answers
- most confident fifth: 60.9% right; most confident half: 46.2% right; most confident four fifths: 40.7% right
- Typical score of a right answer
- 47.9%
- Typical score of a wrong answer
- 36.8%
Repeats
We did not repeat this setup.
Latency
- Latency, median request
- 0.062 s (this run)
- Latency, 90th percentile
- 0.085 s
- Requests at once
- 1
- Throughput
- 914.2 questions per minute of request time
- Request time
- 2.4 minutes, over the requests whose answers were kept
Latency: the time from sending one request to receiving its answer, measured from our computer, network and any retries within the request included. Request time: the time at least one request was in flight, over the requests whose answers were kept. It is not the run's elapsed time: it leaves out requests that failed and any pause before a run was resumed.
Cost
- Cost per 1,000 questionsper 1M questions
- No list price
- Basis
- No dollar estimate; see hardware and time
- Input tokens per question
- 768
- Output tokens per question
- 1