- JSON representation
- RewriterModelCallSignals
- SearchSnippet
- AnswerGenerationModelCallSignals
- AnswerPart
- CitedSnippet
- GroundingSignals
- GroundingDecision
- GroundingScoreBucket
- SafetySignals
- SafetyDecision
- BannedPhraseMatch
Data store connection feature output signals. Might be only partially field if processing stop before the final answer. Reasons for this can be, but are not limited to: empty UCS search results, positive RAI check outcome, grounding failure, ...
JSON representation |
---|
{ "rewriterModelCallSignals": { object ( |
Fields | |
---|---|
rewriterModelCallSignals |
Optional. Diagnostic info related to the rewriter model call. |
rewrittenQuery |
Optional. Rewritten string query used for search. |
searchSnippets[] |
Optional. Search snippets included in the answer generation prompt. |
answerGenerationModelCallSignals |
Optional. Diagnostic info related to the answer generation model call. |
answer |
Optional. The final compiled answer. |
answerParts[] |
Optional. Answer parts with relevant citations. Concatenation of texts should add up the |
citedSnippets[] |
Optional. Snippets cited by the answer generation model from the most to least relevant. |
groundingSignals |
Optional. Grounding signals. |
safetySignals |
Optional. Safety check result. |
RewriterModelCallSignals
Diagnostic info related to the rewriter model call.
JSON representation |
---|
{ "renderedPrompt": string, "modelOutput": string, "model": string } |
Fields | |
---|---|
renderedPrompt |
Prompt as sent to the model. |
modelOutput |
Output of the generative model. |
model |
Name of the generative model. For example, "gemini-ultra", "gemini-pro", "gemini-1.5-flash" etc. Defaults to "Other" if the model is unknown. |
SearchSnippet
Search snippet details.
JSON representation |
---|
{ "documentTitle": string, "documentUri": string, "text": string } |
Fields | |
---|---|
documentTitle |
Title of the enclosing document. |
documentUri |
Uri for the document. Present if specified for the document. |
text |
Text included in the prompt. |
AnswerGenerationModelCallSignals
Diagnostic info related to the answer generation model call.
JSON representation |
---|
{ "renderedPrompt": string, "modelOutput": string, "model": string } |
Fields | |
---|---|
renderedPrompt |
Prompt as sent to the model. |
modelOutput |
Output of the generative model. |
model |
Name of the generative model. For example, "gemini-ultra", "gemini-pro", "gemini-1.5-flash" etc. Defaults to "Other" if the model is unknown. |
AnswerPart
Answer part with citation.
JSON representation |
---|
{ "text": string, "supportingIndices": [ integer ] } |
Fields | |
---|---|
text |
Substring of the answer. |
supportingIndices[] |
Citations for this answer part. Indices of |
CitedSnippet
Snippet cited by the answer generation model.
JSON representation |
---|
{
"searchSnippet": {
object ( |
Fields | |
---|---|
searchSnippet |
Details of the snippet. |
snippetIndex |
Index of the snippet in |
GroundingSignals
Grounding signals.
JSON representation |
---|
{ "decision": enum ( |
Fields | |
---|---|
decision |
Represents the decision of the grounding check. |
score |
Grounding score bucket setting. |
GroundingDecision
Represents the decision of the grounding check.
Enums | |
---|---|
GROUNDING_DECISION_UNSPECIFIED |
Decision not specified. |
ACCEPTED_BY_GROUNDING |
Grounding have accepted the answer. |
REJECTED_BY_GROUNDING |
Grounding have rejected the answer. |
GroundingScoreBucket
Grounding score buckets.
Enums | |
---|---|
GROUNDING_SCORE_BUCKET_UNSPECIFIED |
Score not specified. |
VERY_LOW |
We have very low confidence that the answer is grounded. |
LOW |
We have low confidence that the answer is grounded. |
MEDIUM |
We have medium confidence that the answer is grounded. |
HIGH |
We have high confidence that the answer is grounded. |
VERY_HIGH |
We have very high confidence that the answer is grounded. |
SafetySignals
Safety check results.
JSON representation |
---|
{ "decision": enum ( |
Fields | |
---|---|
decision |
Safety decision. |
bannedPhraseMatch |
Specifies banned phrase match subject. |
matchedBannedPhrase |
The matched banned phrase if there was a match. |
SafetyDecision
Safety decision. All kinds of check are incorporated into this final decision, including banned phrases check.
Enums | |
---|---|
SAFETY_DECISION_UNSPECIFIED |
Decision not specified. |
ACCEPTED_BY_SAFETY_CHECK |
No manual or automatic safety check fired. |
REJECTED_BY_SAFETY_CHECK |
One ore more safety checks fired. |
BannedPhraseMatch
Specifies banned phrase match subject.
Enums | |
---|---|
BANNED_PHRASE_MATCH_UNSPECIFIED |
No banned phrase check was executed. |
BANNED_PHRASE_MATCH_NONE |
All banned phrase checks led to no match. |
BANNED_PHRASE_MATCH_QUERY |
A banned phrase matched the query. |
BANNED_PHRASE_MATCH_RESPONSE |
A banned phrase matched the response. |