Canonical inputs
Use documented dictionary arguments and semantically equivalent content variants.
Tool-calling compatibility
Contract tests for tool-calling chat templates.
Inspect rendering contracts across Python Transformers and Transformers.js. No model weights, inference, or application server required.
Tokenizer onlyNo model weights or inference
Browser-nativeRuns through Transformers.js Jinja
ReproducibleRevision and template hashes recorded
Evidence firstEvery finding includes its observation
20target repositories
18templates loaded
9required-contract failures
0model weights loaded
Live inspector
Test a public Hub tokenizer
Assets stay in your browser cache. Public repositories work without a token; gated repositories cannot be inspected from this static Space.
Loading tokenizer assets…
0%
Only tokenizer and template files are requested.
Reproducible baseline
Twenty model families, one contract
Python reference results from 19 August 2026. Select a row to inspect minimized evidence, resolved revisions, and every individual check.
| Model | Result | Required | Proposed | Dict args | Parallel | Response | Content | Prefix |
|---|
No models match this filter.
Method
Small tests. Explicit contracts.
A failure means a documented or maintainer-confirmed input contract broke. Warnings identify portability or training-pipeline risk without declaring a template invalid.
Preservation
Verify function names, arguments, parallel calls, and tool responses survive rendering.
Cross-engine
Run in JavaScript and flag Python-only Jinja constructs before deployment.
Revision evidence
Record the exact template digest and resolved repository revision for every observation.