MaxCompute:Develop AI Functions on GU resources

import os
import maxframe.dataframe as md
from maxframe import new_session
from maxframe.config import options
from maxframe.learn.contrib.llm.models.managed import ManagedTextGenLLM
from odps import ODPS

# Initialize session
o = ODPS(
    os.getenv('ALIBABA_CLOUD_ACCESS_KEY_ID'),
    os.getenv('ALIBABA_CLOUD_ACCESS_KEY_SECRET'),
    project='<your project>',
    endpoint='https://service.cn-<your region>.maxcompute.aliyun.com/api',
)
session = new_session(o)
options.session.gu_quota_name = "<your GU quota name>"  # Required to route tasks to GPU workers.

# Prepare input data
df = md.DataFrame({"query": ["What is the boiling point of water?"]})

# Initialize the managed LLM client
llm = ManagedTextGenLLM(name="Qwen3-4B-Instruct-2507-FP8")  # Name must be an exact match.

# Define the prompt template and run inference
messages = [
    {"role": "system", "content": "You are a helpful assistant."},
    {"role": "user", "content": "Please answer the following question: {query}"},
]
result_df = llm.generate(df, prompt_template=messages)
result_df.execute()

import os
import maxframe.dataframe as md
import numpy as np
from maxframe import new_session
from maxframe.config import options
from maxframe.udf import with_running_options  # Helper for attaching resource options to UDF-based tasks.
from odps import ODPS
import logging

# Route tasks to the DPE and MCSQL engines; disable SPE for GPU workloads.
options.dag.settings = {
    "engine_order": ["DPE", "MCSQL"],
    "unavailable_engines": ["SPE"],
}

logging.basicConfig(level=logging.INFO)

# Initialize the MaxFrame session with your project credentials.
o = ODPS(
    # Make sure the ALIBABA_CLOUD_ACCESS_KEY_ID environment variable is set to your AccessKey ID,
    # and the ALIBABA_CLOUD_ACCESS_KEY_SECRET environment variable is set to your AccessKey secret.
    # Do not use the AccessKey ID and AccessKey secret strings directly.
    os.getenv('ALIBABA_CLOUD_ACCESS_KEY_ID'),
    os.getenv('ALIBABA_CLOUD_ACCESS_KEY_SECRET'),
    project='<your project>',
    endpoint='https://service.cn-<your region>.maxcompute.aliyun.com/api',
)

session = new_session(o)

options.session.gu_quota_name = "xxxxx"  # Replace with your GU quota name.

print("LogView address:", session.get_logview_address())

Etapa 1: Prepare os dados de entrada

import pandas as pd
from IPython.display import HTML

# Set display options for debugging.
pd.set_option("display.max_colwidth", None)
pd.set_option("display.max_columns", None)
HTML("<style>div.output_area pre {white-space: pre-wrap;}</style>")

# Create a query list.
query_list = [
    "What is the average distance between the Earth and the Sun?",
    "In what year did the American Revolutionary War begin?",
    "What is the boiling point of water?",
    "How can I quickly relieve a headache?",
    "Who is the main character in the Harry Potter series?",
]

# Convert to a MaxFrame DataFrame.
df = md.DataFrame({"query": query_list})
df.execute()

Etapa 2: Inicialize a instância do LLM

ManagedTextGenLLM é o cliente para modelos de geração de texto gerenciados hospedados no MaxCompute. Informe o nome exato do modelo para identificar qual deles será usado.

from maxframe.learn.contrib.llm.models.managed import ManagedTextGenLLM

llm = ManagedTextGenLLM(
    name="Qwen3-4B-Instruct-2507-FP8"  # The model name must be an exact match.
)

Etapa 3: Defina o modelo de prompt

messages = [
    {"role": "system", "content": "You are a helpful assistant."},
    {"role": "user", "content": "Please answer the following question: {query}"},
]

Sintaxe do modelo:

• Use {column_name} como espaço reservado. O sistema substitui esse valor automaticamente pelo conteúdo da coluna correspondente do DataFrame durante a execução.

• Forneça uma lista messages para dar suporte a conversas com múltiplos turnos.

• Use a função system para definir o comportamento do modelo.

Etapa 4: Execute a tarefa de geração

result_df = llm.generate(
    df,                          # Input DataFrame
    prompt_template=messages,
    running_options={
        "max_tokens": 4096,      # Maximum output length
        "verbose": True          # Enable verbose log output
    },
    params={"temperature": 0.7},
)

# Execute and get the result.
result_df.execute()

Otimização de desempenho

Dicas de depuração

Visualize logs de execução

print(session.get_logview_address())  # Click the link to view real-time MaxFrame job logs.

Execute um teste em pequena escala antes da execução completa

df_sample = df.head(2)  # Use two rows for testing.
result_sample = llm.generate(df_sample, prompt_template=messages, running_options={"gu": 2})
result_sample.execute()

Verifique o uso de recursos

Consulte o status detalhado da execução do job no Logview do MaxFrame.