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.

BGE Reranker

Rank passages by relevance to your query, on your device.

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

Reorder English or Chinese passages by how relevant they are to your search query. Returns relevance scores and the passages in ranked order; it does not generate an answer.

License
MIT
Model released
Unknown
Source repository created
Sep 11, 2023
Model weights · FP32
1.1 GB

Best for

  • Rerank a small candidate set produced by semantic or lexical retrieval.
  • Improve the ordering of Chinese or English search and RAG passages.

Keep in mind

  • Every query-passage pair requires model inference, making this slower than embedding retrieval for large corpora.
  • Even the smallest tested weight file is about 279 MB and was not retained by Cache API in the isolated Chrome audit profile, so affected browser sessions redownload it.
  • Scores are relative ranking signals, not calibrated probabilities or factuality judgments.
  • Long passages are truncated and relevant content beyond the token limit can be missed.
  • Ranking can reflect training-data bias and degrade on unsupported languages, specialized domains, malformed text, or adversarial content.
  • Do not use rankings as the sole basis for high-impact decisions.

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
BAAI/bge-reranker-base
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
Jun 24, 2024
Browser model files
Xenova/bge-reranker-base
Files updated
Jun 30, 2025
Artifact revision
280bcc27a84e0b898c251e06fddb25171bd9b101

Input

QUERY → RANKED PASSAGES

Sample result

BGE Reranker Base — Cross-Encoder Ranking

Settings

Limits how many ranked predictions or passages are returned. The listed range is a frontend selection policy, not a claimed model or runtime maximum; the adapter may accept values outside it. Lower values show fewer top results. Higher values include weaker matches and more output. Automatic/null returns all results when offered; it does not mean the model is more accurate.
Advanced settings 2
Limits the input token budget, including any special tokens. 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.
Converts ranking logits to bounded scores. On applies the worker’s score normalization for easier display. Off returns raw scores when supported. Normalized scores are not calibrated probabilities and should not be compared across models.

Output

Find the most relevant passage

Add a query and candidate passages above. We’ll sort them from most to least relevant.

Benchmark evidence

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

BGE Reranker Base — Cross-Encoder Ranking

16 configurations tested
Passed12
Failed4
TestedSep 14, 2026
RuntimeTransformers.js 4.2.0

Quality criteria

Finite scores, deterministic descending order, and the France passage must outrank the unrelated whale passage by at least 0.25 sigmoid probability.

PrecisionDeviceResult
Q4GPU · WebGPUAcceleratedPassedFast
INT8GPU · WebGPUAcceleratedPassedFast
UINT8GPU · WebGPUAcceleratedPassedFast
Q8GPU · WebGPUAcceleratedPassedFast
FP32GPU · WebGPUAcceleratedPassedFast
BNB4GPU · WebGPUAcceleratedPassedModerate
FP32CPU · WASMPassedFast
UINT8CPU · WASMPassedFast
Q8CPU · WASMPassedFast
INT8CPU · WASMPassedFast
BNB4CPU · WASMPassedFast
Q4CPU · WASMPassedFast
FP16CPU · WASMFailed—
FP16GPU · WebGPUFailed—
Q4F16CPU · WASMFailed—
Q4F16GPU · WebGPUFailed—
Test setup & sample details
Quality gate
Finite scores, deterministic descending order, and the France passage must outrank the unrelated whale passage by at least 0.25 sigmoid probability.
Tested
2026-09-14T20:55:17.075Z
Environment
Isolated headless Chrome on macOS; Transformers.js 4.2.0
Runtime
Transformers.js 4.2.0
Artifact revision
280bcc27a84e0b898c251e06fddb25171bd9b101
Fixture
{
  "file": "bge-m3.json",
  "url": "https://f005.backblazeb2.com/file/hubters-webai-public/assets/benchmarks/bge-m3.json",
  "size": 500,
  "sha1": "75a7770d06fc312ddeb39987712a8b4bbe6f1e08",
  "source": "Original WebAI semantic retrieval fixture.",
  "license": "MIT",
  "licenseUrl": "https://opensource.org/license/mit/",
  "preprocessing": "First text is the query; remaining texts are candidate documents."
}
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": "ranking-pass",
      "margin": 0.9997
    },
    "runtime": {
      "loadMs": 38416.5,
      "inferenceMs": 207.90000000596046,
      "totalMs": 38783.59999999404,
      "pairsPerSecond": 9.62,
      "grade": "good"
    }
  },
  {
    "precision": "fp32",
    "device": "webgpu",
    "status": "pass",
    "quality": {
      "grade": "ranking-pass",
      "margin": 0.9997
    },
    "runtime": {
      "loadMs": 37458.90000000596,
      "inferenceMs": 437.2000000178814,
      "totalMs": 38002.09999999404,
      "pairsPerSecond": 4.575,
      "grade": "slow"
    }
  },
  {
    "precision": "fp16",
    "device": "wasm",
    "status": "fail",
    "quality": "Not embedded",
    "runtime": "Not embedded",
    "failure": "Error: Can't create a session. ERROR_CODE: 1, ERROR_MESSAGE: /mnt/vss/_work/1/s/onnxruntime/core/graph/graph_utils.cc:30 int onnxruntime::graph_utils::GetIndexFromName(const Node &, const std::string &, bool) itr != node_args.end() was false. Attempting to get index by a name which does not exist:InsertedPrecisionFreeCast_/roberta/encoder/layer.11/output/LayerNorm/Constant_output_0for node: /roberta/embeddings/LayerNorm/Mul/SimplifiedLayerNormFusion/\n"
  },
  {
    "precision": "fp16",
    "device": "webgpu",
    "status": "fail",
    "quality": "Not embedded",
    "runtime": "Not embedded",
    "failure": "quality gate failed: top=1, margin=-0.24457595644260421"
  },
  {
    "precision": "int8",
    "device": "wasm",
    "status": "pass",
    "quality": {
      "grade": "ranking-pass",
      "margin": 0.9999
    },
    "runtime": {
      "loadMs": 24754.59999999404,
      "inferenceMs": 222.7000000178814,
      "totalMs": 25157.90000000596,
      "pairsPerSecond": 8.981,
      "grade": "good"
    }
  },
  {
    "precision": "int8",
    "device": "webgpu",
    "status": "pass",
    "quality": {
      "grade": "ranking-pass",
      "margin": 0.9998
    },
    "runtime": {
      "loadMs": 27378.40000000596,
      "inferenceMs": 306.40000000596046,
      "totalMs": 27784.69999998808,
      "pairsPerSecond": 6.527,
      "grade": "slow"
    }
  },
  {
    "precision": "uint8",
    "device": "wasm",
    "status": "pass",
    "quality": {
      "grade": "ranking-pass",
      "margin": 0.9998
    },
    "runtime": {
      "loadMs": 17714.59999999404,
      "inferenceMs": 220.19999998807907,
      "totalMs": 18042,
      "pairsPerSecond": 9.083,
      "grade": "good"
    }
  },
  {
    "precision": "uint8",
    "device": "webgpu",
    "status": "pass",
    "quality": {
      "grade": "ranking-pass",
      "margin": 0.9998
    },
    "runtime": {
      "loadMs": 14620.90000000596,
      "inferenceMs": 308.2999999821186,
      "totalMs": 15067.699999988079,
      "pairsPerSecond": 6.487,
      "grade": "slow"
    }
  },
  {
    "precision": "q8",
    "device": "wasm",
    "status": "pass",
    "quality": {
      "grade": "ranking-pass",
      "margin": 0.9996
    },
    "runtime": {
      "loadMs": 11681.600000023842,
      "inferenceMs": 221.69999998807907,
      "totalMs": 12031.300000011921,
      "pairsPerSecond": 9.021,
      "grade": "good"
    }
  },
  {
    "precision": "q8",
    "device": "webgpu",
    "status": "pass",
    "quality": {
      "grade": "ranking-pass",
      "margin": 0.9996
    },
    "runtime": {
      "loadMs": 11473.300000011921,
      "inferenceMs": 370.19999998807907,
      "totalMs": 11951.59999999404,
      "pairsPerSecond": 5.402,
      "grade": "slow"
    }
  },
  {
    "precision": "q4",
    "device": "wasm",
    "status": "pass",
    "quality": {
      "grade": "ranking-pass",
      "margin": 0.9998
    },
    "runtime": {
      "loadMs": 35621,
      "inferenceMs": 322.59999999403954,
      "totalMs": 36118.39999997616,
      "pairsPerSecond": 6.2,
      "grade": "slow"
    }
  },
  {
    "precision": "q4",
    "device": "webgpu",
    "status": "pass",
    "quality": {
      "grade": "ranking-pass",
      "margin": 0.9998
    },
    "runtime": {
      "loadMs": 30102.5,
      "inferenceMs": 239.19999998807907,
      "totalMs": 30552.90000000596,
      "pairsPerSecond": 8.361,
      "grade": "good"
    }
  },
  {
    "precision": "q4f16",
    "device": "wasm",
    "status": "fail",
    "quality": "Not embedded",
    "runtime": "Not embedded",
    "failure": "Error: Can't create a session. ERROR_CODE: 1, ERROR_MESSAGE: /mnt/vss/_work/1/s/onnxruntime/core/graph/graph_utils.cc:30 int onnxruntime::graph_utils::GetIndexFromName(const Node &, const std::string &, bool) itr != node_args.end() was false. Attempting to get index by a name which does not exist:InsertedPrecisionFreeCast_/roberta/embeddings/LayerNorm/Constant_output_0for node: /roberta/embeddings/LayerNorm/Mul/SimplifiedLayerNormFusion/\n"
  },
  {
    "precision": "q4f16",
    "device": "webgpu",
    "status": "fail",
    "quality": "Not embedded",
    "runtime": "Not embedded",
    "failure": "quality gate failed: top=1, margin=-0.24198654101080747"
  },
  {
    "precision": "bnb4",
    "device": "wasm",
    "status": "pass",
    "quality": {
      "grade": "ranking-pass",
      "margin": 0.9996
    },
    "runtime": {
      "loadMs": 30218.5,
      "inferenceMs": 261.09999999403954,
      "totalMs": 30566.80000001192,
      "pairsPerSecond": 7.66,
      "grade": "slow"
    }
  },
  {
    "precision": "bnb4",
    "device": "webgpu",
    "status": "pass",
    "quality": {
      "grade": "ranking-pass",
      "margin": 0.9996
    },
    "runtime": {
      "loadMs": 27562.19999998808,
      "inferenceMs": 516,
      "totalMs": 28174.29999998212,
      "pairsPerSecond": 3.876,
      "grade": "slow"
    }
  }
]

BGE Reranker Large — Cross-Encoder Ranking

16 configurations tested
Passed10
Failed6
TestedSep 15, 2026
RuntimeTransformers.js 4.2.0

Quality criteria

Finite nondegenerate scores, valid unique document indices and descending order; France passage ranks first and exceeds unrelated whale passage by >=0.25 sigmoid score.

PrecisionDeviceResult
FP16GPU · WebGPUAcceleratedPassedFast
Q4F16GPU · WebGPUAcceleratedPassedFast
INT8GPU · WebGPUAcceleratedPassedModerate
UINT8GPU · WebGPUAcceleratedPassedModerate
Q8GPU · WebGPUAcceleratedPassedModerate
UINT8CPU · WASMPassedModerate
Q8CPU · WASMPassedModerate
INT8CPU · WASMPassedModerate
BNB4CPU · WASMPassedModerate
Q4CPU · WASMPassedModerate
FP32CPU · WASMFailed—
FP32GPU · WebGPUFailed—
FP16CPU · WASMFailed—
Q4GPU · WebGPUFailed—
Q4F16CPU · WASMFailed—
BNB4GPU · WebGPUFailed—
Test setup & sample details
Quality gate
Finite nondegenerate scores, valid unique document indices and descending order; France passage ranks first and exceeds unrelated whale passage by >=0.25 sigmoid score.
Tested
2026-09-15T21:08:05.156Z
Environment
Isolated headless Chrome 151 on macOS; Transformers.js 4.2.0
Runtime
Transformers.js 4.2.0
Artifact revision
3c4ff3c9420fb24ea62acd31e3884e09c8827f2a
Fixture
{
  "file": "bge-m3.json",
  "url": "https://f005.backblazeb2.com/file/hubters-webai-public/assets/benchmarks/bge-m3.json",
  "size": 500,
  "sha1": "75a7770d06fc312ddeb39987712a8b4bbe6f1e08",
  "source": "Original WebAI semantic retrieval fixture, reused unchanged.",
  "license": "MIT",
  "licenseUrl": "https://opensource.org/license/mit/",
  "preprocessing": "First text is the query; remaining texts are candidate documents; paired tokenizer truncates at max_length."
}
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": "fail",
    "quality": "Not embedded",
    "runtime": {
      "grade": "failed"
    },
    "failure": "TimeoutError: page.waitForFunction: Timeout 240000ms exceeded."
  },
  {
    "precision": "fp32",
    "device": "webgpu",
    "status": "fail",
    "quality": "Not embedded",
    "runtime": {
      "grade": "failed"
    },
    "failure": "TimeoutError: page.waitForFunction: Timeout 240000ms exceeded."
  },
  {
    "precision": "fp16",
    "device": "wasm",
    "status": "fail",
    "quality": "Not embedded",
    "runtime": {
      "loadMs": null,
      "inferenceMs": null,
      "totalMs": 76154.40000000596,
      "grade": "failed"
    },
    "failure": "Error: Can't create a session. ERROR_CODE: 1, ERROR_MESSAGE: /mnt/vss/_work/1/s/onnxruntime/core/graph/graph_utils.cc:30 int onnxruntime::graph_utils::GetIndexFromName(const Node &, const std::string &, bool) itr != node_args.end() was false. Attempting to get index by a name which does not exist:InsertedPrecisionFreeCast_/roberta/embeddings/LayerNorm/Constant_output_0for node: /roberta/embeddings/LayerNorm/Mul/SimplifiedLayerNormFusion/\n"
  },
  {
    "precision": "fp16",
    "device": "webgpu",
    "status": "pass",
    "quality": {
      "finite": true,
      "descending": true,
      "correctTop": true,
      "margin": 0.4691893627387198,
      "grade": "ranking-pass"
    },
    "runtime": {
      "loadMs": 71767.09999999404,
      "inferenceMs": 181.19999998807907,
      "totalMs": 73131.30000001192,
      "grade": "good",
      "pairsPerSecond": 11.03752759454513
    }
  },
  {
    "precision": "int8",
    "device": "wasm",
    "status": "pass",
    "quality": {
      "finite": true,
      "descending": true,
      "correctTop": true,
      "margin": 0.9997021481159925,
      "grade": "ranking-pass"
    },
    "runtime": {
      "loadMs": 52792.59999999404,
      "inferenceMs": 736.5999999940395,
      "totalMs": 54616.80000001192,
      "grade": "slow",
      "pairsPerSecond": 2.7151778441707624
    }
  },
  {
    "precision": "int8",
    "device": "webgpu",
    "status": "pass",
    "quality": {
      "finite": true,
      "descending": true,
      "correctTop": true,
      "margin": 0.9997051861730285,
      "grade": "ranking-pass"
    },
    "runtime": {
      "loadMs": 49560.80000001192,
      "inferenceMs": 845.9000000059605,
      "totalMs": 51599.89999997616,
      "grade": "slow",
      "pairsPerSecond": 2.364345667319905
    }
  },
  {
    "precision": "uint8",
    "device": "wasm",
    "status": "pass",
    "quality": {
      "finite": true,
      "descending": true,
      "correctTop": true,
      "margin": 0.9994827259927187,
      "grade": "ranking-pass"
    },
    "runtime": {
      "loadMs": 45542.59999999404,
      "inferenceMs": 731.1999999880791,
      "totalMs": 47417.59999999404,
      "grade": "slow",
      "pairsPerSecond": 2.7352297593443744
    }
  },
  {
    "precision": "uint8",
    "device": "webgpu",
    "status": "pass",
    "quality": {
      "finite": true,
      "descending": true,
      "correctTop": true,
      "margin": 0.999471478114171,
      "grade": "ranking-pass"
    },
    "runtime": {
      "loadMs": 54627.79999998212,
      "inferenceMs": 847.5999999940395,
      "totalMs": 56620.30000001192,
      "grade": "slow",
      "pairsPerSecond": 2.359603586614045
    }
  },
  {
    "precision": "q8",
    "device": "wasm",
    "status": "pass",
    "quality": {
      "finite": true,
      "descending": true,
      "correctTop": true,
      "margin": 0.9993575115699435,
      "grade": "ranking-pass"
    },
    "runtime": {
      "loadMs": 53207.60000002384,
      "inferenceMs": 733.7999999821186,
      "totalMs": 55044.09999999404,
      "grade": "slow",
      "pairsPerSecond": 2.7255382938794446
    }
  },
  {
    "precision": "q8",
    "device": "webgpu",
    "status": "pass",
    "quality": {
      "finite": true,
      "descending": true,
      "correctTop": true,
      "margin": 0.9996373090202044,
      "grade": "ranking-pass"
    },
    "runtime": {
      "loadMs": 44939.69999998808,
      "inferenceMs": 1118.199999988079,
      "totalMs": 47194.5,
      "grade": "very-slow",
      "pairsPerSecond": 1.788588803453158
    }
  },
  {
    "precision": "q4",
    "device": "wasm",
    "status": "pass",
    "quality": {
      "finite": true,
      "descending": true,
      "correctTop": true,
      "margin": 0.9995467905840061,
      "grade": "ranking-pass"
    },
    "runtime": {
      "loadMs": 97864.40000000596,
      "inferenceMs": 1045.4000000059605,
      "totalMs": 100099,
      "grade": "very-slow",
      "pairsPerSecond": 1.913143294421845
    }
  },
  {
    "precision": "q4",
    "device": "webgpu",
    "status": "fail",
    "quality": "Not embedded",
    "runtime": {
      "loadMs": null,
      "inferenceMs": null,
      "totalMs": 104540.90000000596,
      "grade": "failed"
    },
    "failure": "Error: operation does not support unaligned accesses"
  },
  {
    "precision": "q4f16",
    "device": "wasm",
    "status": "fail",
    "quality": "Not embedded",
    "runtime": {
      "loadMs": null,
      "inferenceMs": null,
      "totalMs": 61940.89999997616,
      "grade": "failed"
    },
    "failure": "Error: Can't create a session. ERROR_CODE: 1, ERROR_MESSAGE: /mnt/vss/_work/1/s/onnxruntime/core/graph/graph_utils.cc:30 int onnxruntime::graph_utils::GetIndexFromName(const Node &, const std::string &, bool) itr != node_args.end() was false. Attempting to get index by a name which does not exist:InsertedPrecisionFreeCast_/roberta/embeddings/LayerNorm/Constant_output_0for node: /roberta/embeddings/LayerNorm/Mul/SimplifiedLayerNormFusion/\n"
  },
  {
    "precision": "q4f16",
    "device": "webgpu",
    "status": "pass",
    "quality": {
      "finite": true,
      "descending": true,
      "correctTop": true,
      "margin": 0.46908440642910454,
      "grade": "ranking-pass"
    },
    "runtime": {
      "loadMs": 57640.39999997616,
      "inferenceMs": 456.2999999821186,
      "totalMs": 59271.30000001192,
      "grade": "slow",
      "pairsPerSecond": 4.383081306329992
    }
  },
  {
    "precision": "bnb4",
    "device": "wasm",
    "status": "pass",
    "quality": {
      "finite": true,
      "descending": true,
      "correctTop": true,
      "margin": 0.9990787634865452,
      "grade": "ranking-pass"
    },
    "runtime": {
      "loadMs": 85690.40000000596,
      "inferenceMs": 868.5,
      "totalMs": 87704.79999998212,
      "grade": "slow",
      "pairsPerSecond": 2.3028209556706964
    }
  },
  {
    "precision": "bnb4",
    "device": "webgpu",
    "status": "fail",
    "quality": "Not embedded",
    "runtime": {
      "loadMs": null,
      "inferenceMs": null,
      "totalMs": 88240.30000001192,
      "grade": "failed"
    },
    "failure": "Error: operation does not support unaligned accesses"
  }
]

Something not quite right?

Tell us what happened with BGE Reranker. 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.