Mit Funktionsaufrufen können Sie auf Gemini basierende Anwendungen und Agents erstellen, die mit Echtzeitinformationen und Diensten wie Datenbanken, Customer-Relationship-Management-Systemen und Dokument-Repositories interagieren können. Dadurch kann Ihre Anwendung relevantere und kontextbezogenere Antworten liefern.
Mit dem überwachten Fine-Tuning können Sie Ihren eigenen Tuning-Datensatz verwenden, um die allgemeine Genauigkeit der funktionsaufrufbezogenen Antworten Ihrer Anwendung zu verbessern.
Format des Abstimmungs-Datasets
Der fileUri
für Ihr Feinabstimmungs-Dataset kann der URI für eine Datei in einem Cloud Storage-Bucket oder eine öffentlich verfügbare HTTP- oder HTTPS-URL sein.
Ein Beispiel für das generische Format finden Sie im Dataset-Beispiel für Gemini.
In den folgenden Abschnitten finden Sie Beispiele für Datasets für Funktionsaufrufe, die zum Erstellen eines Abstimmungsjobs für Funktionsaufrufe verwendet werden können.
Funktionsaufrufe für die Textgenerierung optimieren
Das folgende Beispiel zeigt ein Dataset zum Abstimmen von Funktionsaufrufen zum Generieren einer einzelnen Textantwort.
{
"system_instruction": {
"role": "system",
"parts": [
{
"text": "You are an assistant that helps users find the best product for them."
}
]
},
"contents": [
{
"role": "user",
"parts": [
{
"text": "Do you have the White Pixel 8 Pro 128GB in stock in the US?"
}
]
},
{
"role": "model",
"parts": [
{
"functionCall": {
"name": "get_product_sku",
"args": {
"product_name": "Pixel 8 Pro 128GB"
}
}
}
]
}
],
"tools": [
{
"functionDeclarations": [
{
"name": "get_product_sku",
"description": "Get the available inventory for a Google products, e.g: Pixel phones, Pixel Watches, Google Home etc",
"parameters": {
"type": "OBJECT",
"properties": {
"product_name": {
"type": "STRING",
"description": "Product name",
"enum": [
"Pixel 8 Pro 128GB",
"Pixel 8 Pro 256GB",
"Pixel 8 Pro 512GB",
"Pixel 8 Pro 1TB"
]
}
}
}
},
{
"name": "get_store_location",
"description": "Get the location of the closest store",
"parameters": {
"type": "OBJECT",
"properties": {
"location": {
"type": "STRING",
"description": "Location"
}
}
}
}
]
}
]
}
Funktionsaufrufe für eine Chatsitzung optimieren
Das folgende Beispiel zeigt ein Dataset für die Feinabstimmung von Funktionsaufrufen zur Unterstützung einer Chatsitzung.
{
"system_instruction": {
"role": "system",
"parts": [
{
"text": "You are an assistant that helps users find the best product for them."
}
]
},
"contents": [
{
"role": "user",
"parts": [
{
"text": "Do you have the Porcelain Pixel 8 Pro 128GB in stock in the US?"
}
]
},
{
"role": "model",
"parts": [
{
"functionCall": {
"name": "get_product_sku",
"args": {
"product_name": "Pixel 8 Pro 128GB"
}
}
}
]
},
{
"parts": [
{
"functionResponse": {
"name": "get_product_sku",
"response": {
"output": "True"
}
}
}
]
},
{
"role": "model",
"parts": [
{
"text": "Yes, we have the Porcelain Pixel 8 Pro 128GB in stock in the US."
}
]
}
],
"tools": [
{
"functionDeclarations": [
{
"name": "get_product_sku",
"description": "Get the available inventory for a Google products, e.g: Pixel phones, Pixel Watches, Google Home etc",
"parameters": {
"type": "OBJECT",
"properties": {
"product_name": {
"type": "STRING",
"description": "Product name",
"enum": [
"Pixel 8 Pro 128GB",
"Pixel 8 Pro 256GB",
"Pixel 8 Pro 512GB",
"Pixel 8 Pro 1TB"
]
}
}
}
},
{
"name": "get_store_location",
"description": "Get the location of the closest store",
"parameters": {
"type": "OBJECT",
"properties": {
"location": {
"type": "STRING",
"description": "Location"
}
}
}
}
]
}
]
}
Parallele Funktionsaufrufe optimieren
Im Folgenden finden Sie ein Beispiel für ein Dataset zur Feinabstimmung von parallelen Funktionsaufrufen.
{
"system_instruction": {
"role": "system",
"parts": [
{
"text": "You are an assistant that helps users find the best product for them."
}
]
},
"contents": [
{
"role": "user",
"parts": [
{
"text": "Where is the closest Google Store to Mountain View, CA that has Pixel 8 Pro 128GB in stock?"
}
]
},
{
"role": "model",
"parts": [
{
"functionCall": {
"name": "get_product_sku",
"args": {
"product_name": "Pixel 8 Pro 128GB"
}
}
},
{
"functionCall": {
"name": "get_store_location",
"args": {
"location": "Mountain View, CA"
}
}
}
]
}
],
"tools": [
{
"functionDeclarations": [
{
"name": "get_product_sku",
"description": "Get the available inventory for a Google products, e.g: Pixel phones, Pixel Watches, Google Home etc",
"parameters": {
"type": "OBJECT",
"properties": {
"product_name": {
"type": "STRING",
"description": "Product name",
"enum": [
"Pixel 8 Pro 128GB",
"Pixel 8 Pro 256GB",
"Pixel 8 Pro 512GB",
"Pixel 8 Pro 1TB"
]
}
}
}
},
{
"name": "get_store_location",
"description": "Get the location of the closest store",
"parameters": {
"type": "OBJECT",
"properties": {
"location": {
"type": "STRING",
"description": "Location"
}
}
}
}
]
}
]
}
Nächste Schritte
- Informationen zum Erstellen eines Abstimmungsjobs und zum Testen des abgestimmten Modells finden Sie unter Gemini-Modelle mithilfe der überwachten Feinabstimmung abstimmen.
- Weitere Informationen zur Optimierung von Gemini-Modellen finden Sie unter Einführung in die Optimierung.
- Weitere Informationen zu Funktionsaufrufen finden Sie unter Einführung in Funktionsaufrufe.