Review context token budget

Measure a microtheory-inspired review context against BitNet b1.58 2B4T’s 4,096-token window. Nothing is uploaded or shortened. This checker does not run the model or change saved context definitions.

Load the matching tokenizer once for this tab

Select tokenizer.json and tokenizer_config.json from the official model revision 04c3b9ad9361b824064a1f25ea60a8be9599b127. Exact hashes are checked. Files stay local; no automatic download occurs.

Official tokenizer filesTokenizer JSONTokenizer configuration JSON

Tokenizer not loaded. Token fit is not measured.

Rendered instructions and source before the contextFull review context and dependency definitionsRendered text after the context

Include the exact chat formatting and special-token strings required by your runtime. The checker measures only the text you supply. Word count means whitespace-separated words; characters are Unicode code points.

Reserved answer tokensAdditional safety margin tokens

512 and 128 are editable planning reserves, not inherent BitNet limits. The remaining allowance follows from your measured prompt.

Not measured

Over-budget input is refused for the checked handoff. It remains intact here for editing. Saving a context in the main form still uses its separate UTF-8 storage budget; this page does not silently change that policy.