Dans cet article, nous explorerons comment créer des agents d'IA pour la recherche sur le Web, l'analyse financière, le raisonnement et la génération augmentée par récupération à l'aide de phidata et du LLM local Ollama. Le code utilise le modèle lama3.2. Si vous souhaitez utiliser un modèle différent, vous devrez télécharger le modèle que vous souhaitez utiliser et remplacer la variable model_id dans le code.
Une plateforme open source pour créer, expédier et surveiller des systèmes agentiques.
https://www.phidata.com/
Ollama est une plateforme et un ensemble d'outils conçus pour simplifier le déploiement et l'utilisation de grands modèles linguistiques locaux (LLM).
https://ollama.ai/
Dans cet article, nous utiliserons le modèle lama3.2.
ollama pull llama3.2
Un package Python extrêmement rapide et un gestionnaire de projet, écrit en Rust.
https://github.com/astral-sh/uv
Si vous ne souhaitez pas utiliser uv, vous pouvez utiliser pip au lieu de uv. Ensuite, vous devrez utiliser pip install au lieu de uv add.
https://docs.astral.sh/uv/getting-started/installation/
Si vous décidez d'utiliser pip, vous devrez créer un dossier de projet.
uv init phidata-ollama
uv add phidata ollama duckduckgo-search yfinance pypdf lancedb tantivy sqlalchemy
Dans cet article, nous allons essayer de créer 5 agents IA avec phidata et Ollama.
Remarque : Avant de commencer, assurez-vous que votre serveur ollama est en cours d'exécution en exécutant ollama serve.
Le premier agent que nous allons créer est un agent de recherche Web qui utilisera le moteur de recherche DuckDuckGo.
from phi.agent import Agent
from phi.model.ollama import Ollama
from phi.tools.duckduckgo import DuckDuckGo
model_id = "llama3.2"
model = Ollama(id=model_id)
web_agent = Agent(
name="Web Agent",
model=model,
tools=[DuckDuckGo()],
instructions=["Always include sources"],
show_tool_calls=True,
markdown=True,
)
web_agent.print_response("Tell me about OpenAI Sora?", stream=True)
Sortie :
┏━ Message ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━┓ ┃ ┃ ┃ Tell me about OpenAI Sora? ┃ ┃ ┃ ┗━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━┛ ┏━ Response (12.0s) ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━┓ ┃ ┃ ┃ ┃ ┃ • Running: duckduckgo_news(query=OpenAI Sora) ┃ ┃ ┃ ┃ OpenAI's Sora is a video-generating model that has been trained on ┃ ┃ copyrighted content, which has raised concerns about its legality. ┃ ┃ According to TechCrunch, it appears that OpenAI trained Sora on game ┃ ┃ content, which could be a problem. Additionally, MSN reported that the ┃ ┃ model doesn't feel like the game-changer it was supposed to be. ┃ ┃ ┃ ┃ In other news, Yahoo reported that when asked to generate gymnastics ┃ ┃ videos, Sora produces horrorshow videos with whirling and morphing ┃ ┃ limbs. A lawyer told ExtremeTech that it's "overwhelmingly likely" that ┃ ┃ copyrighted materials are included in Sora's training dataset. ┃ ┃ ┃ ┃ Geeky Gadgets reviewed OpenAI's Sora, stating that while it is included ┃ ┃ in the 0/month Pro Plan, its standalone value for video generation ┃ ┃ is less clear compared to other options. ┃ ┃ ┃ ┗━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━┛
Le deuxième agent que nous allons créer est un agent financier qui utilisera les outils yfinance.
from phi.agent import Agent
from phi.model.ollama import Ollama
from phi.tools.yfinance import YFinanceTools
model_id = "llama3.2"
model = Ollama(id=model_id)
finance_agent = Agent(
name="Finance Agent",
model=model,
tools=[YFinanceTools(stock_price=True, analyst_recommendations=True, company_info=True, company_news=True)],
instructions=["Use tables to display data"],
show_tool_calls=True,
markdown=True,
)
finance_agent.print_response("Summarize analyst recommendations for NVDA", stream=True)
Sortie :
┏━ Message ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━┓ ┃ ┃ ┃ Summarize analyst recommendations for NVDA ┃ ┃ ┃ ┗━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━┛ ┏━ Response (3.9s) ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━┓ ┃ ┃ ┃ ┃ ┃ • Running: get_analyst_recommendations(symbol=NVDA) ┃ ┃ ┃ ┃ Based on the analyst recommendations, here is a summary: ┃ ┃ ┃ ┃ • The overall sentiment is bullish, with 12 strong buy and buy ┃ ┃ recommendations. ┃ ┃ • There are no strong sell or sell recommendations. ┃ ┃ • The average price target for NVDA is around 0-0. ┃ ┃ • Analysts expect NVDA to continue its growth trajectory, driven by ┃ ┃ its strong products and services in the tech industry. ┃ ┃ ┃ ┃ Please note that these recommendations are subject to change and may ┃ ┃ not reflect the current market situation. It's always a good idea to do ┃ ┃ your own research and consult with a financial advisor before making ┃ ┃ any investment decisions. ┃ ┃ ┃ ┗━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━┛
Le troisième agent que nous allons créer est une équipe d'agents qui utilisera le moteur de recherche DuckDuckGo et les outils YFinance.
from phi.agent import Agent
from phi.model.ollama import Ollama
from phi.tools.duckduckgo import DuckDuckGo
from phi.tools.yfinance import YFinanceTools
web_instructions = 'Always include sources'
finance_instructions = 'Use tables to display data'
model_id = "llama3.2"
model = Ollama(id=model_id)
web_agent = Agent(
name="Web Agent",
role="Search the web for information",
model=model,
tools=[DuckDuckGo()],
instructions=[web_instructions],
show_tool_calls=True,
markdown=True,
)
finance_agent = Agent(
name="Finance Agent",
role="Get financial data",
model=model,
tools=[YFinanceTools(stock_price=True, analyst_recommendations=True, company_info=True)],
instructions=[finance_instructions],
show_tool_calls=True,
markdown=True,
)
agent_team = Agent(
model=model,
team=[web_agent, finance_agent],
instructions=[web_instructions, finance_instructions],
show_tool_calls=True,
markdown=True,
)
agent_team.print_response("Summarize analyst recommendations and share the latest news for NVDA", stream=True)
Le quatrième agent que nous allons créer est un agent d'inférence qui va utiliser une tâche.
from phi.agent import Agent from phi.model.ollama import Ollama model_id = "llama3.2" model = Ollama(id=model_id) task = ( "Three missionaries and three cannibals want to cross a river." "There is a boat that can carry up to two people, but if the number of cannibals exceeds the number of missionaries, the missionaries will be eaten." ) reasoning_agent = Agent(model=model, reasoning=True, markdown=True, structured_outputs=True) reasoning_agent.print_response(task, stream=True, show_full_reasoning=True)
Sortie :
┏━ Message ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━┓ ┃ ┃ ┃ Three missionaries and three cannibals want to cross a river.There is a ┃ ┃ boat that can carry up to two people, but if the number of cannibals ┃ ┃ exceeds the number of missionaries, the missionaries will be eaten. ┃ ┃ ┃ ┗━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━┛ [Reasoning steps and output as in the original document]
Le cinquième agent que nous allons créer est un agent RAG qui utilisera une base de connaissances PDF et une base de données vectorielle LanceDB.
from phi.agent import Agent
from phi.model.openai import OpenAIChat
from phi.embedder.openai import OpenAIEmbedder
from phi.embedder.ollama import OllamaEmbedder
from phi.model.ollama import Ollama
from phi.knowledge.pdf import PDFUrlKnowledgeBase
from phi.vectordb.lancedb import LanceDb, SearchType
model_id = "llama3.2"
model = Ollama(id=model_id)
embeddings = OllamaEmbedder().get_embedding("The quick brown fox jumps over the lazy dog.")
knowledge_base = PDFUrlKnowledgeBase(
urls=["https://phi-public.s3.amazonaws.com/recipes/ThaiRecipes.pdf"],
vector_db=LanceDb(
table_name="recipes",
uri="tmp/lancedb",
search_type=SearchType.vector,
embedder=OllamaEmbedder(),
),
)
knowledge_base.load()
agent = Agent(
model=model,
knowledge=knowledge_base,
show_tool_calls=True,
markdown=True,
)
agent.print_response("Please tell me how to make green curry.", stream=True)
Sortie :
uv run rag_agent.py
WARNING model "openhermes" not found, try pulling it first
WARNING model "openhermes" not found, try pulling it first
INFO Creating collection
INFO Loading knowledge base
INFO Reading:
https://phi-public.s3.amazonaws.com/recipes/ThaiRecipes.pdf
WARNING model "openhermes" not found, try pulling it first
WARNING model "openhermes" not found, try pulling it first
WARNING model "openhermes" not found, try pulling it first
WARNING model "openhermes" not found, try pulling it first
WARNING model "openhermes" not found, try pulling it first
WARNING model "openhermes" not found, try pulling it first
WARNING model "openhermes" not found, try pulling it first
WARNING model "openhermes" not found, try pulling it first
WARNING model "openhermes" not found, try pulling it first
WARNING model "openhermes" not found, try pulling it first
WARNING model "openhermes" not found, try pulling it first
WARNING model "openhermes" not found, try pulling it first
WARNING model "openhermes" not found, try pulling it first
WARNING model "openhermes" not found, try pulling it first
INFO Added 14 documents to knowledge base
WARNING model "openhermes" not found, try pulling it first
ERROR Error searching for documents: list index out of range
┏━ Message ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━┓
┃ ┃
┃ Please tell me how to make green curry. ┃
┃ ┃
┗━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━┛
┏━ Response (5.4s) ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━┓
┃ ┃
┃ ┃
┃ • Running: search_knowledge_base(query=green curry recipe) ┃
┃ ┃
┃ ┏━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━┓ ┃
┃ ┃ Green Curry Recipe ┃ ┃
┃ ┗━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━┛ ┃
┃ ┃
┃ ** Servings: 4-6 people** ┃
┃ ┃
┃ Ingredients: ┃
┃ ┃
┃ • 2 tablespoons vegetable oil ┃
┃ • 2 cloves garlic, minced ┃
┃ • 1 tablespoon grated fresh ginger ┃
┃ • 2 tablespoons Thai red curry paste ┃
┃ • 2 cups coconut milk ┃
┃ • 1 cup mixed vegetables (such as bell peppers, bamboo shoots, and ┃
┃ Thai eggplant) ┃
┃ • 1 pound boneless, skinless chicken breasts or thighs, cut into ┃
┃ bite-sized pieces ┃
┃ • 2 tablespoons fish sauce ┃
┃ • 1 tablespoon palm sugar ┃
┃ • 1/4 teaspoon ground white pepper ┃
┃ • Salt to taste ┃
┃ • Fresh basil leaves for garnish ┃
┃ ┃
┃ Instructions: ┃
┃ ┃
┃ 1 Prepare the curry paste: In a blender or food processor, combine the ┃
┃ curry paste, garlic, ginger, fish sauce, palm sugar, and white ┃
┃ pepper. Blend until smooth. ┃
┃ 2 Heat oil in a pan: Heat the oil in a large skillet or Dutch oven ┃
┃ over medium-high heat. ┃
┃ 3 Add the curry paste: Pour the blended curry paste into the hot oil ┃
┃ and stir constantly for 1-2 minutes, until fragrant. ┃
┃ 4 Add coconut milk: Pour in the coconut milk and bring the mixture to ┃
┃ a simmer. ┃
┃ 5 Add vegetables and chicken: Add the mixed vegetables and chicken ┃
┃ pieces to the pan. Stir gently to combine. ┃
┃ 6 Reduce heat and cook: Reduce the heat to medium-low and let the ┃
┃ curry simmer, uncovered, for 20-25 minutes or until the chicken is ┃
┃ cooked through and the sauce has thickened. ┃
┃ 7 Season with salt and taste: Season the curry with salt to taste. ┃
┃ Serve hot garnished with fresh basil leaves. ┃
┃ ┃
┃ Tips and Variations: ┃
┃ ┃
┃ • Adjust the level of spiciness by using more or less Thai red curry ┃
┃ paste. ┃
┃ • Add other protein sources like shrimp, tofu, or tempeh for a ┃
┃ vegetarian or vegan option. ┃
┃ • Experiment with different vegetables, such as zucchini or carrots, ┃
┃ to add variety. ┃
┃ ┃
┃ Tools Used: Python ┃
┃ ┃
┗━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━┛
Dans cet article, nous avons exploré comment créer des agents d'IA pour la recherche sur le Web, l'analyse financière, le raisonnement et la génération augmentée par récupération à l'aide de phidata et du LLM local Ollama.
Ce qui précède est le contenu détaillé de. pour plus d'informations, suivez d'autres articles connexes sur le site Web de PHP en chinois!
![[Web front-end] Démarrage rapide de Node.js](https://img.php.cn/upload/course/000/000/067/662b5d34ba7c0227.png)
