ShareGPTTraceFlavorConfig¶
class ShareGPTTraceFlavorConfig ( BaseTraceFlavorConfig )
ShareGPT conversation trace flavor configuration.
Polymorphic Type:
type: sharegptAll
BaseTraceFlavorConfigtypes:
timed_synthetic_session: TimedSyntheticSessionTraceFlavorConfig
shared_prefix: SharedPrefixTraceFlavorConfig
rag: RAGTraceFlavorConfig
request_log: RequestLogTraceFlavorConfig
untimed_content_multi_turn: UntimedContentMultiTurnTraceFlavorConfig
sharegpt: ShareGPTTraceFlavorConfig
audio: AudioTraceFlavorConfig
seed_tts_text: SeedTTSTextTraceFlavorConfig
Fields:
assistant_rolestr ="gpt"Role name for assistant turns in the ShareGPT data (common values: ‘gpt’, ‘assistant’).
min_tokensint =20Minimum input token count. Turns shorter than this are skipped. Ignored when min_chars/max_chars are set.
max_tokensint =100Maximum input token count. Text is truncated to sampled length. Ignored when min_chars/max_chars are set.
min_charsint =-1If >= 0, enables char-based input length control (mutually exclusive with min_tokens/max_tokens). Turns with fewer than this many chars are skipped.
max_charsint =-1If >= 0, enables char-based input length control (mutually exclusive with min_tokens/max_tokens). Text is truncated to a char length sampled uniformly in [min_chars, max_chars].
min_alpha_ratiofloat =0.5Minimum ratio of alphabetic characters to total non-space characters. Filters out junk entries like number sequences or code snippets. Set to 0.0 to disable.