LLM Stopping Criteria

Simulate token-by-token generation with max_tokens, stop sequences, and natural EOS — see how finish_reason is determined without calling any API.

Presets Load a scenario, then Generate

Stopping controls

max_tokens 64

Stream speed 80ms

Stop sequence

Include stop text in output (OpenAI default: exclude)

Streaming output Scenario: ReAct tool handoff

Output appears here as tokens stream…

Tokens generated

Remaining budget

finish_reason

—

Stop matched

—

Generation events log here…

stop — EOS or stop sequence length — max_tokens hit end_turn — model finished naturally