Struct FTryllSamplingOverrides¶

ClassList > FTryllSamplingOverrides

#include <TryllSamplingOverrides.h>

Public Attributes¶

Type	Name
float	FrequencyPenalty = `0.0`
int32	MaxTokens = `0`
float	MinP = `0.0`
float	PresencePenalty = `0.0`
float	RepeatPenalty = `0.0`
int32	Seed = `0`
float	Temperature = `0.0`
int32	TopK = `0`
float	TopP = `0.0`
bool	bOverrideFrequencyPenalty = `false`
bool	bOverrideMaxTokens = `false`
bool	bOverrideMinP = `false`
bool	bOverridePresencePenalty = `false`
bool	bOverrideRepeatPenalty = `false`
bool	bOverrideSeed = `false`
bool	bOverrideTemperature = `false`
bool	bOverrideTopK = `false`
bool	bOverrideTopP = `false`

Detailed Description¶

Sparse sampling override block. Null on any field means "inherit the selected model's default". No per-field (on_change) attributes: the outer sub-table field carries one callback that re-resolves the whole block and commits derived state.

Public Attributes Documentation¶

variable FrequencyPenalty¶

float FTryllSamplingOverrides::FrequencyPenalty;

variable MaxTokens¶

int32 FTryllSamplingOverrides::MaxTokens;

variable MinP¶

float FTryllSamplingOverrides::MinP;

variable PresencePenalty¶

float FTryllSamplingOverrides::PresencePenalty;

variable RepeatPenalty¶

float FTryllSamplingOverrides::RepeatPenalty;

variable Seed¶

int32 FTryllSamplingOverrides::Seed;

variable Temperature¶

float FTryllSamplingOverrides::Temperature;

variable TopK¶

int32 FTryllSamplingOverrides::TopK;

variable TopP¶

float FTryllSamplingOverrides::TopP;

variable bOverrideFrequencyPenalty¶

bool FTryllSamplingOverrides::bOverrideFrequencyPenalty;

Penalty that grows with how often a token was already used. Unset = the model's default.

variable bOverrideMaxTokens¶

bool FTryllSamplingOverrides::bOverrideMaxTokens;

Longest answer this node may generate, in tokens. Must be smaller than Context Size. Unset = the model's default.

variable bOverrideMinP¶

bool FTryllSamplingOverrides::bOverrideMinP;

Drop tokens less likely than this fraction of the top token (0–1). Unset = the model's default.

variable bOverridePresencePenalty¶

bool FTryllSamplingOverrides::bOverridePresencePenalty;

Penalty for any token already used, pushing toward new topics. Unset = the model's default.

variable bOverrideRepeatPenalty¶

bool FTryllSamplingOverrides::bOverrideRepeatPenalty;

Penalty for repeating recent tokens; 1 = none, higher = less repetition. Unset = the model's default.

variable bOverrideSeed¶

bool FTryllSamplingOverrides::bOverrideSeed;

Random seed; the same seed and input give the same output. Unset = the model's default.

variable bOverrideTemperature¶

bool FTryllSamplingOverrides::bOverrideTemperature;

Randomness of the output: 0 = always the most likely token, 2 = very random. Unset = the model's default.

variable bOverrideTopK¶

bool FTryllSamplingOverrides::bOverrideTopK;

Pick only from this many most likely tokens. Unset = the model's default.

variable bOverrideTopP¶

bool FTryllSamplingOverrides::bOverrideTopP;

Nucleus sampling: pick only from the most likely tokens that together reach this probability (0–1). Unset = the model's default.

The documentation for this class was generated from the following file C:/_tryll/_monorepo3/tryll/clients/unreal/Source/TryllClient/Public/Generated/Nodes/TryllSamplingOverrides.h