A model is still running

A model is still running or downloading. Leaving will stop the current operation. Stay on this page to let it finish.

Skip to content
LocalAI.toolsBeta

Join the waitlist

Get an email when the new apps and model runs launch.

No account needed.

Laya Typed Decisions — Q4

Ask structured questions about an English text state, such as choosing a category, assigning a score, or answering yes/no. Returns option probabilities rather than free-form replies; decisions require human review.

Browser inferencePowered by WebAI · Private · Unlimited runs
About this model

Ask structured questions about an English text state, such as choosing a category, assigning a score, or answering yes/no. Returns option probabilities rather than free-form replies; decisions require human review.

License
Apache-2.0
Model released
Unknown
Source repository created
Sep 18, 2026
Model weights · Q4
448.6 MB

Best for

  • Classify and score English agent traces, customer-service cases, invoices, and security incidents.
  • Return application-provided choices, ordered ratings, and yes/no probabilities for human-reviewed workflow automation.
  • Prototype deterministic structured-decision interfaces in a browser.

Keep in mind

  • This open model is not TypeSafe AI's proprietary Jev model and is not affiliated with TypeSafe AI.
  • The source reports 0.766 accuracy on its four synthetic workflow families, but performance outside those English domains may fall back to the weaker general checkpoint behavior.
  • The source reports ECE 0.213 and warns that the checkpoint remains over-confident; recalibrate probabilities on held-out application data before thresholding them.
  • The Q4 publisher reports six-of-six argmax parity cases, not task accuracy; conversion parity does not prove that a decision is correct.
  • The source model's option-count calibration includes a value below the SDK's safe range for 11+ choices; this adapter clamps it to 0.5 and treats affected confidence as uncalibrated.
  • The model is English-only and choice questions should remain below roughly 20 options because they share a fixed question-head token budget.
  • Long states are truncated to fit the 1,024-token context; option text is truncated within a fixed 256-token question-head budget.
  • It cannot generate explanations, and its outputs must not be the sole basis for safety-critical or consequential decisions.
  • The Q4 first download is approximately 452 MB including tokenizer assets and may not persist in every browser cache profile.

Processed on your deviceInference runs in this browser. Input/output sharing requires consent for each run. Basic run logs include device information and approximate connection location. Model and runtime files download from external hosts. Saved downloads may be cleared by your browser.

Sources & technical details
Original model
convaiinnovations/laya (typed-decisions checkpoint)
Release verification
Unknown
Release evidence
Exact original checkpoint public release date has not been verified. Repository creation, training, paper, artifact conversion and update dates are not release evidence.
Source updated
Sep 20, 2026
Browser model files
m1rhan/laya-typed-decisions-ONNX
Files updated
Sep 22, 2026
Artifact revision
7388ca40228f08fd72fb3d2fdbd29153e9bc63d4

Input

TEXT

Sample result

Laya Typed Decisions — Q4
Advanced settings 2
Maximum encoded tokens for each question/state row. Lower values may process less text but truncate relevant context sooner. Higher values retain more context, within the model limit, and can need more memory and time.
Question-and-options token budget before state tokens are appended. Lower values may truncate questions or choices. Higher values retain more question context but leave less room for the state within the fixed model context.

Output

Your result will appear here

Add input or try the sample, then run the model.

Benchmark evidence

Saved worker evidence is separate from benchmarks run on your device. It does not guarantee speed or quality on every browser.

Laya Typed Decisions — Q4

2 configurations tested
Passed2
Failed0
TestedSep 25, 2026
RuntimeTransformers.js 4.2.0

Quality criteria

Every agent-observability, customer-service, invoice, and security case must return finite normalized non-degenerate distributions and the checked-in expected choice/yes-no result or ordered-score interval.

PrecisionDeviceResult
Q4GPU · WebGPUAcceleratedPassedModerate
Q4CPU · WASMPassedVery slow
Test setup & sample details
Quality gate
Every agent-observability, customer-service, invoice, and security case must return finite normalized non-degenerate distributions and the checked-in expected choice/yes-no result or ordered-score interval.
Tested
2026-09-25T09:51:15.016Z
Environment
Isolated headless Google Chrome on macOS; Transformers.js 4.2.0
Runtime
Transformers.js 4.2.0
Artifact revision
7388ca40228f08fd72fb3d2fdbd29153e9bc63d4
Fixture
{
  "file": "laya-typed-decisions.json",
  "url": "https://f005.backblazeb2.com/file/hubters-webai-public/assets/benchmarks/laya-typed-decisions.json",
  "size": 2242,
  "sha1": "cee490dc2efc280f3d781dd6c7ec47c6d23ba737",
  "source": "WebAI-authored regression cases representing the source model's four intended workflow families.",
  "license": "CC0-1.0",
  "licenseUrl": "https://creativecommons.org/publicdomain/zero/1.0/",
  "preprocessing": "Pinned ModernBERT tokenizer; each question/options prefix uses Laya's marker format and the shared state is truncated to the remaining context."
}
Rerun when
artifact revision change; runtime version change; fixture or quality-gate change; browser/backend behavior change
Per-case quality & failure details
[
  {
    "precision": "q4",
    "device": "wasm",
    "status": "pass",
    "quality": {
      "grade": "typed-decision-pass",
      "cases": [
        {
          "domain": "agent-observability",
          "type": "choice",
          "pass": true
        },
        {
          "domain": "agent-observability",
          "type": "noul",
          "pass": true
        },
        {
          "domain": "customer-service",
          "type": "choice",
          "pass": true
        },
        {
          "domain": "customer-service",
          "type": "noul",
          "pass": true
        },
        {
          "domain": "invoice-processing",
          "type": "choice",
          "pass": true
        },
        {
          "domain": "invoice-processing",
          "type": "score",
          "pass": true
        },
        {
          "domain": "security-incidents",
          "type": "choice",
          "pass": true
        },
        {
          "domain": "security-incidents",
          "type": "noul",
          "pass": true
        }
      ]
    },
    "runtime": {
      "loadMs": 53499.4,
      "inferenceMs": 9939.8,
      "totalMs": 64657.9,
      "grade": "very-slow",
      "questionsPerSecond": 0.8048
    }
  },
  {
    "precision": "q4",
    "device": "webgpu",
    "status": "pass",
    "quality": {
      "grade": "typed-decision-pass",
      "cases": [
        {
          "domain": "agent-observability",
          "type": "choice",
          "pass": true
        },
        {
          "domain": "agent-observability",
          "type": "noul",
          "pass": true
        },
        {
          "domain": "customer-service",
          "type": "choice",
          "pass": true
        },
        {
          "domain": "customer-service",
          "type": "noul",
          "pass": true
        },
        {
          "domain": "invoice-processing",
          "type": "choice",
          "pass": true
        },
        {
          "domain": "invoice-processing",
          "type": "score",
          "pass": true
        },
        {
          "domain": "security-incidents",
          "type": "choice",
          "pass": true
        },
        {
          "domain": "security-incidents",
          "type": "noul",
          "pass": true
        }
      ]
    },
    "runtime": {
      "loadMs": 53566.3,
      "inferenceMs": 938.7,
      "totalMs": 55818.6,
      "grade": "good",
      "questionsPerSecond": 8.5224
    }
  }
]

Open Jev Laya Multilingual — Typed Decisions

4 configurations tested
Passed4
Failed0
TestedSep 25, 2026
RuntimeTransformers.js 4.2.0

Quality criteria

Every English, Chinese, and Spanish case must return finite normalized non-degenerate distributions and the checked-in expected choice/yes-no result or ordered-score interval.

PrecisionDeviceResult
FP32GPU · WebGPUAcceleratedPassedModerate
FP16GPU · WebGPUAcceleratedPassedModerate
FP32CPU · WASMPassedModerate
FP16CPU · WASMPassedModerate
Test setup & sample details
Quality gate
Every English, Chinese, and Spanish case must return finite normalized non-degenerate distributions and the checked-in expected choice/yes-no result or ordered-score interval.
Tested
2026-09-25T08:11:04.378Z
Environment
Isolated headless Google Chrome on macOS; Transformers.js 4.2.0
Runtime
Transformers.js 4.2.0
Artifact revision
643219b2b3a112ff1c40fb1143f050aa871f76f7
Fixture
{
  "file": "open-jev-laya-multilingual.json",
  "url": "https://f005.backblazeb2.com/file/hubters-webai-public/assets/benchmarks/open-jev-laya-multilingual.json",
  "size": 1368,
  "sha1": "ff4a5039dbc0c37a52717f0c82420ff45e0a2803",
  "source": "WebAI-authored multilingual regression cases with unambiguous gold labels.",
  "license": "CC0-1.0",
  "licenseUrl": "https://creativecommons.org/publicdomain/zero/1.0/",
  "preprocessing": "Pinned mmBERT tokenizer; each question/options prefix uses Laya's marker format and the shared state is truncated to the remaining context."
}
Rerun when
artifact revision change; runtime version change; fixture or quality-gate change; browser/backend behavior change
Per-case quality & failure details
[
  {
    "precision": "fp32",
    "device": "wasm",
    "status": "pass",
    "quality": {
      "grade": "multilingual-typed-decision-pass",
      "cases": [
        {
          "language": "en",
          "type": "choice",
          "pass": true
        },
        {
          "language": "en",
          "type": "noul",
          "pass": true
        },
        {
          "language": "zh",
          "type": "choice",
          "pass": true
        },
        {
          "language": "zh",
          "type": "noul",
          "pass": true
        },
        {
          "language": "es",
          "type": "score",
          "pass": true
        },
        {
          "language": "es",
          "type": "noul",
          "pass": true
        }
      ]
    },
    "runtime": {
      "loadMs": 141147.9,
      "inferenceMs": 1615.8,
      "totalMs": 144071.1,
      "grade": "good",
      "questionsPerSecond": 3.7133
    }
  },
  {
    "precision": "fp32",
    "device": "webgpu",
    "status": "pass",
    "quality": {
      "grade": "multilingual-typed-decision-pass",
      "cases": [
        {
          "language": "en",
          "type": "choice",
          "pass": true
        },
        {
          "language": "en",
          "type": "noul",
          "pass": true
        },
        {
          "language": "zh",
          "type": "choice",
          "pass": true
        },
        {
          "language": "zh",
          "type": "noul",
          "pass": true
        },
        {
          "language": "es",
          "type": "score",
          "pass": true
        },
        {
          "language": "es",
          "type": "noul",
          "pass": true
        }
      ]
    },
    "runtime": {
      "loadMs": 132489.8,
      "inferenceMs": 524.2,
      "totalMs": 134322.4,
      "grade": "excellent",
      "questionsPerSecond": 11.446
    }
  },
  {
    "precision": "fp16",
    "device": "wasm",
    "status": "pass",
    "quality": {
      "grade": "multilingual-typed-decision-pass",
      "cases": [
        {
          "language": "en",
          "type": "choice",
          "pass": true
        },
        {
          "language": "en",
          "type": "noul",
          "pass": true
        },
        {
          "language": "zh",
          "type": "choice",
          "pass": true
        },
        {
          "language": "zh",
          "type": "noul",
          "pass": true
        },
        {
          "language": "es",
          "type": "score",
          "pass": true
        },
        {
          "language": "es",
          "type": "noul",
          "pass": true
        }
      ]
    },
    "runtime": {
      "loadMs": 72630.7,
      "inferenceMs": 1860.8,
      "totalMs": 75768.9,
      "grade": "good",
      "questionsPerSecond": 3.2244
    }
  },
  {
    "precision": "fp16",
    "device": "webgpu",
    "status": "pass",
    "quality": {
      "grade": "multilingual-typed-decision-pass",
      "cases": [
        {
          "language": "en",
          "type": "choice",
          "pass": true
        },
        {
          "language": "en",
          "type": "noul",
          "pass": true
        },
        {
          "language": "zh",
          "type": "choice",
          "pass": true
        },
        {
          "language": "zh",
          "type": "noul",
          "pass": true
        },
        {
          "language": "es",
          "type": "score",
          "pass": true
        },
        {
          "language": "es",
          "type": "noul",
          "pass": true
        }
      ]
    },
    "runtime": {
      "loadMs": 79046.5,
      "inferenceMs": 1648.2,
      "totalMs": 81778.8,
      "grade": "good",
      "questionsPerSecond": 3.6403
    }
  }
]

Something not quite right?

Tell us what happened with Laya. Loading, results, or the page itself—we’re here for all of it.

Attachments Optional

Up to 3 files · 50 MB each

Browser, OS, CPU/GPU, memory, screen, storage, and network details are included automatically. Model inputs, results, and browsing history are not collected.