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.

Back to models

Mixedbread Edge ColBERT v0 17M

Encode English queries and passages as 48-number vectors for each retained token.

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

Encode English queries and passages as 48-number vectors for each retained token.

Created by
Mixedbread
License
Apache-2.0
Model released
Oct 16, 2025
Source repository created
Oct 13, 2025
Model weights · INT8
17.2 MB

Best for

  • Late-interaction text retrieval using MaxSim.
  • Token-level query and passage comparison.

Keep in mind

  • Measured on three small English retrieval fixtures. Evaluate on your own corpus before relying on rankings.
  • Queries are truncated to 48 tokens and documents to 512 tokens, including special tokens.
  • Retrieval scores measure relevance and can reflect training-data biases. Scores do not establish factual accuracy.
  • The selected runtime returns complete token matrices after inference. Incremental streaming is unavailable.

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
mixedbread-ai/mxbai-edge-colbert-v0-17m
Release verification
Verified
Release evidence
Mixedbread's dated release announcement explicitly introduces both checkpoints.
Source updated
Aug 19, 2026
Browser model files
mixedbread-ai/mxbai-edge-colbert-v0-17m
Files updated
Aug 19, 2026
Artifact revision
592c6417c1c6687572043408ed1ae5196bce16b1

Input

TEXT

Sample result

Mixedbread Edge ColBERT v0 17M

Settings

Choose the encoder prefix and token limit. Uses the matching query or document token policy.

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.

Mixedbread Edge ColBERT v0 17M

4 configurations tested
Passed4
Failed0
TestedNov 14, 2026
RuntimeONNX Runtime Web 1.24.3

Quality criteria

Three held-out ranking fixtures, exact token IDs and agreement with original native FP32 token vectors on browser.

PrecisionDeviceResult
INT8GPU · WebGPUAcceleratedPassedFast
FP32GPU · WebGPUAcceleratedPassedFast
FP32CPU · WASMPassedVery fast
INT8CPU · WASMPassedVery fast
Test setup & sample details
Quality gate
Three held-out ranking fixtures, exact token IDs and agreement with original native FP32 token vectors on browser.
Tested
2026-11-14T12:31:45.149Z
Environment
Apple M3 Max on macOS 27.2. Managed native or isolated Chrome module workers.
Runtime
ONNX Runtime Web 1.24.3
Artifact revision
592c6417c1c6687572043408ed1ae5196bce16b1
Fixture
{
  "file": "model-samples/mxbai-edge-colbert-v0-astronomy-query.json",
  "size": 163,
  "sha1": "040b125553ac4ebfb935837e3627846506d3fc44",
  "sha256": "0790ccf44917ad2edb477b1eac3d6d7c2e3aeb2e87b74aa25a8b4f3251485cbe",
  "url": "https://assets.localai.tools/samples/v1/files/0790ccf44917ad2edb477b1eac3d6d7c2e3aeb2e87b74aa25a8b4f3251485cbe/mxbai-edge-colbert-v0-astronomy-query.json",
  "durationSeconds": 0
}
Rerun when
artifact revision; runtime version; retrieval fixture; tokenizer or projection semantics
Per-case quality & failure details
[
  {
    "precision": "fp32",
    "device": "wasm",
    "status": "pass",
    "quality": {
      "passed": true,
      "tokenParity": true,
      "winner": 1,
      "expected": 1,
      "minimumTokenCosine": 0.9999999222746616,
      "scores": [
        11.569321213834568,
        11.758439276108566,
        11.7098902410522,
        11.722881123059208
      ],
      "grade": "retrieval-pass"
    },
    "runtime": {
      "durationMs": 31.399999976158142,
      "inferenceMs": 31.399999976158142,
      "loadMs": 2310.699999988079,
      "totalMs": 2342.099999964237,
      "embeddingsPerSecond": 31.847133782143146,
      "grade": "interactive",
      "version": "1.24.3",
      "device": "wasm",
      "precision": "fp32",
      "fallback": false,
      "artifactRevision": "592c6417c1c6687572043408ed1ae5196bce16b1",
      "acceleration": "cpu"
    }
  },
  {
    "precision": "fp32",
    "device": "webgpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "tokenParity": true,
      "winner": 1,
      "expected": 1,
      "minimumTokenCosine": 0.9999999222747175,
      "scores": [
        11.569320979816498,
        11.75843905899234,
        11.709890053989646,
        11.722880868636704
      ],
      "grade": "retrieval-pass"
    },
    "runtime": {
      "durationMs": 373.5999999642372,
      "inferenceMs": 373.5999999642372,
      "loadMs": 2218.100000023842,
      "totalMs": 2591.699999988079,
      "embeddingsPerSecond": 2.6766595291641457,
      "grade": "slow",
      "version": "1.24.3",
      "device": "webgpu",
      "precision": "fp32",
      "fallback": false,
      "artifactRevision": "592c6417c1c6687572043408ed1ae5196bce16b1",
      "acceleration": "partial",
      "gpuSubmissionObserved": true,
      "accelerationDetails": "WebGPU graph execution observed. Strict GPU-only ORT initialization rejects CPU-assigned nodes."
    }
  },
  {
    "precision": "int8",
    "device": "wasm",
    "status": "pass",
    "quality": {
      "passed": true,
      "tokenParity": true,
      "winner": 1,
      "expected": 1,
      "minimumTokenCosine": 0.9878741632290672,
      "scores": [
        11.545614789828376,
        11.780920859510198,
        11.735356228734117,
        11.7317340829946
      ],
      "grade": "retrieval-pass"
    },
    "runtime": {
      "durationMs": 34.60000002384186,
      "inferenceMs": 34.60000002384186,
      "loadMs": 975.0999999642372,
      "totalMs": 1009.6999999880791,
      "embeddingsPerSecond": 28.901734084130894,
      "grade": "interactive",
      "version": "1.24.3",
      "device": "wasm",
      "precision": "int8",
      "fallback": false,
      "artifactRevision": "592c6417c1c6687572043408ed1ae5196bce16b1",
      "acceleration": "cpu"
    }
  },
  {
    "precision": "int8",
    "device": "webgpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "tokenParity": true,
      "winner": 1,
      "expected": 1,
      "minimumTokenCosine": 0.9878741641161962,
      "scores": [
        11.5456147872575,
        11.784995014137278,
        11.735356221114206,
        11.731734083030734
      ],
      "grade": "retrieval-pass"
    },
    "runtime": {
      "durationMs": 136.5,
      "inferenceMs": 136.5,
      "loadMs": 1124,
      "totalMs": 1260.5,
      "embeddingsPerSecond": 7.326007326007326,
      "grade": "interactive",
      "version": "1.24.3",
      "device": "webgpu",
      "precision": "int8",
      "fallback": false,
      "artifactRevision": "592c6417c1c6687572043408ed1ae5196bce16b1",
      "acceleration": "partial",
      "gpuSubmissionObserved": true,
      "accelerationDetails": "WebGPU graph execution observed. Strict GPU-only ORT initialization rejects CPU-assigned nodes."
    }
  }
]

Mixedbread Edge ColBERT v0 32M

4 configurations tested
Passed4
Failed0
TestedNov 14, 2026
RuntimeONNX Runtime Web 1.24.3

Quality criteria

Three held-out ranking fixtures, exact token IDs and agreement with original native FP32 token vectors on browser.

PrecisionDeviceResult
FP32GPU · WebGPUAcceleratedPassedVery fast
INT8GPU · WebGPUAcceleratedPassedFast
FP32CPU · WASMPassedVery fast
INT8CPU · WASMPassedVery fast
Test setup & sample details
Quality gate
Three held-out ranking fixtures, exact token IDs and agreement with original native FP32 token vectors on browser.
Tested
2026-11-14T12:31:56.215Z
Environment
Apple M3 Max on macOS 27.2. Managed native or isolated Chrome module workers.
Runtime
ONNX Runtime Web 1.24.3
Artifact revision
bb13a29ec9b1e7edd4ba8f7a0776c48b55cbad66
Fixture
{
  "file": "model-samples/mxbai-edge-colbert-v0-astronomy-query.json",
  "size": 163,
  "sha1": "040b125553ac4ebfb935837e3627846506d3fc44",
  "sha256": "0790ccf44917ad2edb477b1eac3d6d7c2e3aeb2e87b74aa25a8b4f3251485cbe",
  "url": "https://assets.localai.tools/samples/v1/files/0790ccf44917ad2edb477b1eac3d6d7c2e3aeb2e87b74aa25a8b4f3251485cbe/mxbai-edge-colbert-v0-astronomy-query.json",
  "durationSeconds": 0
}
Rerun when
artifact revision; runtime version; retrieval fixture; tokenizer or projection semantics
Per-case quality & failure details
[
  {
    "precision": "fp32",
    "device": "wasm",
    "status": "pass",
    "quality": {
      "passed": true,
      "tokenParity": true,
      "winner": 1,
      "expected": 1,
      "minimumTokenCosine": 0.9999999104413182,
      "scores": [
        11.208053612063589,
        11.530794767794067,
        11.410383859382843,
        11.475588106651083
      ],
      "grade": "retrieval-pass"
    },
    "runtime": {
      "durationMs": 39.700000047683716,
      "inferenceMs": 39.700000047683716,
      "loadMs": 1649.800000011921,
      "totalMs": 1689.5000000596046,
      "embeddingsPerSecond": 25.18891684631987,
      "grade": "interactive",
      "version": "1.24.3",
      "device": "wasm",
      "precision": "fp32",
      "fallback": false,
      "artifactRevision": "bb13a29ec9b1e7edd4ba8f7a0776c48b55cbad66",
      "acceleration": "cpu"
    }
  },
  {
    "precision": "fp32",
    "device": "webgpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "tokenParity": true,
      "winner": 1,
      "expected": 1,
      "minimumTokenCosine": 0.9999999104413265,
      "scores": [
        11.208053347183395,
        11.530794344719762,
        11.41038352845283,
        11.475587785241157
      ],
      "grade": "retrieval-pass"
    },
    "runtime": {
      "durationMs": 94.39999997615814,
      "inferenceMs": 94.39999997615814,
      "loadMs": 2321.899999976158,
      "totalMs": 2416.2999999523163,
      "embeddingsPerSecond": 10.593220341658496,
      "grade": "interactive",
      "version": "1.24.3",
      "device": "webgpu",
      "precision": "fp32",
      "fallback": false,
      "artifactRevision": "bb13a29ec9b1e7edd4ba8f7a0776c48b55cbad66",
      "acceleration": "partial",
      "gpuSubmissionObserved": true,
      "accelerationDetails": "WebGPU graph execution observed. Strict GPU-only ORT initialization rejects CPU-assigned nodes."
    }
  },
  {
    "precision": "int8",
    "device": "wasm",
    "status": "pass",
    "quality": {
      "passed": true,
      "tokenParity": true,
      "winner": 1,
      "expected": 1,
      "minimumTokenCosine": 0.9552493048890526,
      "scores": [
        11.15346514974274,
        11.473826120479387,
        11.381143182156283,
        11.426124448761167
      ],
      "grade": "retrieval-pass"
    },
    "runtime": {
      "durationMs": 41.200000047683716,
      "inferenceMs": 41.200000047683716,
      "loadMs": 2345,
      "totalMs": 2386.2000000476837,
      "embeddingsPerSecond": 24.271844632102628,
      "grade": "interactive",
      "version": "1.24.3",
      "device": "wasm",
      "precision": "int8",
      "fallback": false,
      "artifactRevision": "bb13a29ec9b1e7edd4ba8f7a0776c48b55cbad66",
      "acceleration": "cpu"
    }
  },
  {
    "precision": "int8",
    "device": "webgpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "tokenParity": true,
      "winner": 1,
      "expected": 1,
      "minimumTokenCosine": 0.955249305606278,
      "scores": [
        11.153465159319337,
        11.476000952012422,
        11.38114318822641,
        11.426879162483457
      ],
      "grade": "retrieval-pass"
    },
    "runtime": {
      "durationMs": 144.5999999642372,
      "inferenceMs": 144.5999999642372,
      "loadMs": 1449.5,
      "totalMs": 1594.0999999642372,
      "embeddingsPerSecond": 6.915629323978715,
      "grade": "interactive",
      "version": "1.24.3",
      "device": "webgpu",
      "precision": "int8",
      "fallback": false,
      "artifactRevision": "bb13a29ec9b1e7edd4ba8f7a0776c48b55cbad66",
      "acceleration": "partial",
      "gpuSubmissionObserved": true,
      "accelerationDetails": "WebGPU graph execution observed. Strict GPU-only ORT initialization rejects CPU-assigned nodes."
    }
  }
]

Something not quite right?

Tell us what happened with Mixedbread Edge ColBERT v0. 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.