Skip to main content

Intent recognition

Identifie les intentions des utilisateurs en quelques millisecondes et sélectionne les outils appropriés pour répondre aux requêtes.

China (Beijing) région uniquement. Utilisez une clé API provenant de China (Beijing).
Alibaba Cloud Model Studio a publié des domaines spécifiques à chaque espace de travail pour les régions China (Beijing), Singapour et China (Hong Kong). Ces nouveaux domaines dédiés offrent des performances supérieures et une stabilité accrue pour les requêtes d'inférence. Nous vous recommandons de migrer vers ces nouveaux domaines :
  • China (Beijing) : de https://dashscope.aliyuncs.com vers https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com
  • Singapour : de https://dashscope-intl.aliyuncs.com vers https://{WorkspaceId}.ap-southeast-1.maas.aliyuncs.com
  • China (Hong Kong) : de https://cn-hongkong.dashscope.aliyuncs.com vers https://{WorkspaceId}.cn-hongkong.maas.aliyuncs.com
{WorkspaceId} correspond à l'ID de votre espace de travail, disponible sur la page Workspace Details dans la console Alibaba Cloud Model Studio. Le domaine existant reste pleinement fonctionnel.

Modèles pris en charge

Modèle

Fenêtre de contexte

Entrée max.

Sortie max.

Prix d'entrée

Prix de sortie

(tokens)

(pour 1 M tokens)

tongyi-intent-detect-v3

8 192

8 192

1 024

$0,058

$0,144

Utilisation

Prérequis

Obtenir une clé API et exporter la clé API en tant que variable d'environnement. Si vous utilisez le SDK OpenAI ou le SDK DashScope pour effectuer vos appels, installez le SDK.

Retourner à la fois l'intention et les informations d'appel de fonction

Pour retourner simultanément l'intention et les informations d'appel de fonction, configurez le message système comme suit :
You are Qwen, created by Alibaba Cloud. You are a helpful assistant. You may call one or more tools to assist with the user query. The tools you can use are as follows:
{Tool information}
Response in INTENT_MODE.
Incluez Response in INTENT_MODE. dans le message système et spécifiez les outils disponibles. Le format des informations relatives aux outils est le suivant :
[{
    "name": "Name of tool 1",
    "description": "Description of tool 1",
    "parameters": {
        "type": "The type of the parameter, typically object",
        "properties": {
            "parameter_1": {
                "description": "Description of parameter_1",
                "type": "Type of parameter_1",
                "default": "Default value of parameter_1"
            },
            ...
            "parameter_n": {
                "description": "Description of parameter_n",
                "type": "Type of parameter_n",
                "default": "Default value of parameter_n"
            }
        },
        "required": [
        "parameter_1",
        ...
        "parameter_n"
    ]
    },
},
...
{
    "name": "Name of tool n",
    "description": "Description of tool n",
    "parameters": {
        "type": "The type of the parameter, typically object",
        "properties": {
            "parameter_1": {
                "description": "Description of parameter_1",
                "type": "Type of parameter_1",
                "default": "Default value of parameter_1"
            },
            ...
            "parameter_n": {
                "description": "Description of parameter_n",
                "type": "Type of parameter_n",
                "default": "Default value of parameter_n"
            }
        },
        "required": [
        "parameter_1",
        ...
        "parameter_n"
    ]
    },
}]
Exemple avec deux outils (requête d'heure et requête météo) :
[
    {
        "name": "get_current_time",
        "description": "This is useful when you want to know the current time.",
        "parameters": {}
    },
    {
        "name": "get_current_weather",
        "description": "This is useful when you want to query the weather of a specified city.",
        "parameters": {
            "type": "object",
            "properties": {
                "location": {
                    "type": "string",
                    "description": "A city or district, such as Beijing, Hangzhou, or Yuhang District."
                }
            },
            "required": ["location"]
        }
    }
]

Exemple de requête

import os
import json
from openai import OpenAI

# Define tools
tools = [
    {
        "name": "get_current_time",
        "description": "This is useful when you want to know the current time.",
        "parameters": {}
    },
    {
        "name": "get_current_weather",
        "description": "This is useful when you want to query the weather of a specified city.",
        "parameters": {
            "type": "object",
            "properties": {
                "location": {
                    "type": "string",
                    "description": "A city or district, such as Beijing, Hangzhou, or Yuhang District.",
                }
            },
            "required": ["location"]
        }
    }
]

tools_string = json.dumps(tools,ensure_ascii=False)

system_prompt = f"""You are Qwen, created by Alibaba Cloud. You are a helpful assistant. You may call one or more tools to assist with the user query. The tools you can use are as follows:
{tools_string}
Response in INTENT_MODE."""
client = OpenAI(
    api_key=os.getenv("DASHSCOPE_API_KEY"),
    base_url="https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/compatible-mode/v1",
)
messages = [
    {'role': 'system', 'content': system_prompt},
    {'role': 'user', 'content': "Weather in Hangzhou"}
    ]
response = client.chat.completions.create(
    model="tongyi-intent-detect-v3",
    messages=messages
)

print(response.choices[0].message.content)

Exemple de réponse

<tags>
[function call, json response]
</tags><tool_call>
[{"name": "get_current_weather", "arguments": {"location": "Hangzhou"}}]
</tool_call><content>

</content>
Analysez la réponse à l'aide de la fonction parse_text :
import re

def parse_text(text):
    # Define regular expression patterns to match <tags>, <tool_call>, <content>, and their content
    tags_pattern = r'<tags>(.*?)</tags>'
    tool_call_pattern = r'<tool_call>(.*?)</tool_call>'
    content_pattern = r'<content>(.*?)</content>'
    # Use regular expressions to find matching content
    tags_match = re.search(tags_pattern, text, re.DOTALL)
    tool_call_match = re.search(tool_call_pattern, text, re.DOTALL)
    content_match = re.search(content_pattern, text, re.DOTALL)
    # Extract matched content (returns empty string if no match)
    tags = tags_match.group(1).strip() if tags_match else ""
    tool_call = tool_call_match.group(1).strip() if tool_call_match else ""
    content = content_match.group(1).strip() if content_match else ""
    # Store the extracted content in a dictionary
    result = {
      "tags": tags,
      "tool_call": tool_call,
      "content": content
    }
    return result

response = """<tags>
[function call, json response]
</tags><tool_call>
[{"name": "get_current_weather", "arguments": {"location": "Hangzhou"}}]
</tool_call><content>

</content>"""
print(parse_text(response))
Le résultat obtenu est le suivant :
{
    "tags": "[function call, json response]",
    "tool_call": [
        {
            "name": "get_current_weather",
            "arguments": {
                "location": "Hangzhou"
            }
        }
    ],
    "content": ""
}

Retourner uniquement les informations d'intention

Pour ne retourner que les informations d'intention, configurez le message système comme suit :
You are Qwen, created by Alibaba Cloud. You are a helpful assistant. \nYou should choose one tag from the tag list:\n{intent information}\njust reply with the chosen tag.
Le format des informations d'intention se présente ainsi :
{
    "Intent 1": "Description of Intent 1",
    "Intent 2": "Description of Intent 2",
    "Intent 3": "Description of Intent 3",
    ...
}

Exemple de requête

import os
import json
from openai import OpenAI

intent_dict = {
    "play_game": "Play game",
    "email_querycontact": "Email query contact",
    "general_quirky": "quirky",
    "email_addcontact": "Email add contact",
    "takeaway_query": "Takeaway query",
    "recommendation_locations": "Location recommendation",
    "transport_traffic": "Transportation",
    "iot_cleaning": "IoT - vacuum cleaner, cleaner",
    "general_joke": "Joke",
    "lists_query": "Query list/checklist",
    "calendar_remove": "Calendar delete event",
    "transport_taxi": "Taxi, taxi booking",
    "qa_factoid": "Factual Q&A",
    "transport_ticket": "Transportation ticket",
    "play_radio": "Play radio",
    "alarm_set": "Set alarm",
}

intent_string = json.dumps(intent_dict,ensure_ascii=False)

system_prompt = f"""You are Qwen, created by Alibaba Cloud. You are a helpful assistant.
You should choose one tag from the tag list:
{intent_string}
Just reply with the chosen tag."""

client = OpenAI(
    api_key=os.getenv("DASHSCOPE_API_KEY"),
    base_url="https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/compatible-mode/v1",
)
messages = [
    {'role': 'system', 'content': system_prompt},
    {'role': 'user', 'content': "Wake me up at nine on Friday morning"}
    ]
response = client.chat.completions.create(
    model="tongyi-intent-detect-v3",
    messages=messages
)

print(response.choices[0].message.content)

Exemple de réponse

alarm_set

Améliorer le temps de réponse de la reconnaissance d'intention

Pour réduire le temps de réponse, utilisez des lettres majuscules uniques pour les catégories d'intention. Cela génère des réponses composées d'un seul token, optimisant ainsi la latence des appels au modèle.
import os
import json
from openai import OpenAI

intent_dict = {
    "A": "Play game",
    "B": "Email query contact",
    "C": "quirky",
    "D": "Email add contact",
    "E": "Takeaway query",
    "F": "Location recommendation",
    "G": "Transportation",
    "H": "IoT - vacuum cleaner, cleaner",
    "I": "Joke",
    "J": "Query list/checklist",
    "K": "Calendar delete event",
    "L": "Taxi, taxi booking",
    "M": "Factual Q&A",
    "N": "Transportation ticket",
    "O": "Play radio",
    "P": "Set alarm",
}

intent_string = json.dumps(intent_dict, ensure_ascii=False)

system_prompt = f"""You are Qwen, created by Alibaba Cloud. You are a helpful assistant.
You should choose one tag from the tag list:
{intent_string}
Just reply with the chosen tag."""

client = OpenAI(
    api_key=os.getenv("DASHSCOPE_API_KEY"),
    base_url="https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/compatible-mode/v1",
)
messages = [
    {"role": "system", "content": system_prompt},
    {"role": "user", "content": "What is the earliest flight from Beijing to Hangzhou?"},
]
response = client.chat.completions.create(
    model="tongyi-intent-detect-v3", messages=messages
)

print(response.choices[0].message.content)
Résultat : une intention sous forme de token unique.
M

Retourner uniquement les informations d'appel de fonction

Pour ne retourner que les informations d'appel de fonction, configurez le message système comme suit :
You are Qwen, created by Alibaba Cloud. You are a helpful assistant. You may call one or more tools to assist with the user query. The tools you can use are as follows:\n{Tool information}\nResponse in NORMAL_MODE.
Le format des informations relatives aux outils est identique à celui décrit dans la section Retourner à la fois l'intention et les informations d'appel de fonction.

Exemple de requête

import os
import json
from openai import OpenAI

# Define tools
tools = [
    {
        "name": "get_current_time",
        "description": "This is useful when you want to know the current time.",
        "parameters": {}
    },
    {
        "name": "get_current_weather",
        "description": "This is useful when you want to query the weather of a specified city.",
        "parameters": {
            "type": "object",
            "properties": {
                "location": {
                    "type": "string",
                    "description": "A city or district, such as Beijing, Hangzhou, or Yuhang District.",
                }
            },
            "required": ["location"]
        }
    }
]

tools_string = json.dumps(tools,ensure_ascii=False)

system_prompt = f"""You are Qwen, created by Alibaba Cloud. You are a helpful assistant. You may call one or more tools to assist with the user query. The tools you can use are as follows:
{tools_string}
Response in NORMAL_MODE."""
client = OpenAI(
    api_key=os.getenv("DASHSCOPE_API_KEY"),
    base_url="https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/compatible-mode/v1",
)
messages = [
    {'role': 'system', 'content': system_prompt},
    {'role': 'user', 'content': "Weather in Hangzhou"}
    ]
response = client.chat.completions.create(
    model="tongyi-intent-detect-v3",
    messages=messages
)

print(response.choices[0].message.content)

Exemple de réponse

<tool_call>
{"name": "get_current_weather", "arguments": {"location": "Hangzhou"}}
</tool_call>
Une fois la réponse reçue, utilisez la fonction parse_text pour extraire les informations relatives à l'outil et aux paramètres retournés :
import re

def parse_text(text):
    tool_call_pattern = r'<tool_call>(.*?)</tool_call>'
    # Use regular expressions to find matching content
    tool_call_match = re.search(tool_call_pattern, text, re.DOTALL)
    # Extract matched content (returns empty string if no match)
    tool_call = tool_call_match.group(1).strip() if tool_call_match else ""
    return tool_call

response = """<tool_call>
{"name": "get_current_weather", "arguments": {"location": "Hangzhou"}}
</tool_call>"""
print(parse_text(response))
Le résultat obtenu est le suivant :
{"name": "get_current_weather", "arguments": {"location": "Hangzhou"}}

Conversations multi-tours

Si une requête manque d'informations suffisantes, le modèle pose des questions complémentaires. Une fois les paramètres nécessaires collectés, il retourne les informations d'appel de fonction.
  • Retourner à la fois l'intention et les informations d'appel de fonction
  • Retourner uniquement les informations d'appel de fonction
Exemple de requête
import os
import json
from openai import OpenAI

# Define tools
tools = [
    {
        "name": "get_current_time",
        "description": "This is useful when you want to know the current time.",
        "parameters": {},
    },
    {
        "name": "get_current_weather",
        "description": "This is useful when you want to query the weather of a specified city.",
        "parameters": {
            "type": "object",
            "properties": {
                "location": {
                    "type": "string",
                    "description": "A city or district, such as Beijing, Hangzhou, or Yuhang District.",
                }
            },
            "required": ["location"],
        },
    },
]

tools_string = json.dumps(tools, ensure_ascii=False)

system_prompt = f"""You are Qwen, created by Alibaba Cloud. You are a helpful assistant. You may call one or more tools to assist with the user query. The tools you can use are as follows:
{tools_string}
Response in INTENT_MODE."""
client = OpenAI(
    api_key=os.getenv("DASHSCOPE_API_KEY"),
    base_url="https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/compatible-mode/v1",
)
messages = [
    {"role": "system", "content": system_prompt},
    # First round question
    {"role": "user", "content": "I want to check the weather"},
]
response = client.chat.completions.create(
    model="tongyi-intent-detect-v3", messages=messages
)

print("Query: I want to check the weather")
print("First-round output:\n")
print(response.choices[0].message.content)
messages.append(response.choices[0].message)
# Second round question
messages.append({"role": "user", "content": "In Hangzhou"})
response = client.chat.completions.create(
    model="tongyi-intent-detect-v3", messages=messages
)
print("\nQuery: In Hangzhou")
print("Second-round output:\n")
print(response.choices[0].message.content)
Exemple de réponse
Query: I want to check the weather
First-round output:

<tags>
[weather inquiry]
</tags><tool_call>
[]
</tool_call><content>
OK. Which city's weather would you like to check?
</content>

Query: Hangzhou
Second-round output:

<tags>
[function call, json response]
</tags><tool_call>
[{"name": "get_current_weather", "arguments": {"location": "Hangzhou"}}]
</tool_call><content>

</content>

FAQ

Quel est le nombre maximal d'outils pouvant être transmis ?

Ne fournissez pas plus de 10 outils. Dépasser cette limite risque de réduire la précision des appels d'outils.