使用 Qwen2.5-32B-Instruct，在隨附的 8xH100 AI 運行環境中對 16,000 則多語言語音助理話語進行分類。 本筆記本示範如何：
- 從 MASSIVE 1.1 建立一個平衡的多語言射線資料集。
- 在每個可用 GPU 上執行一個持久化的 vLLM 模型副本。
- 用 Ray 儀表板和 MLflow 系統的指標來監控工作負載。
- 將完整的預測結果以 Parquet 格式儲存至 Unity Catalog 磁碟區中。
Note
此範例需要 Databricks AI 環境版本 5 或以上。
連接到無伺服器的 GPU 運算
- 從筆記型電腦的運算選擇器中，選擇 無伺服器 GPU。
- 在 環境 面板中，選擇 8xH100 加速器和 AI v5 環境。
- 點選 套用，然後確認環境。
Qwen 模型是公開的，不需要 Hugging Face 驗證。 這台筆記本從亞馬遜的公開檔案庫下載了 MASSIVE 1.1。
匯入程式庫
AI v5 包含本筆記本中使用的 Ray、vLLM、Hugging Face Datasets、Transformers、PyTorch 及 MLflow 套件，因此無需安裝套件。
import json
import re
import time
from pathlib import Path
import mlflow
import pandas as pd
from datasets import DownloadConfig, DownloadManager, concatenate_datasets, load_dataset
from datasets.utils.logging import disable_progress_bar
from pyspark.sql import functions as F
from vllm import LLM, SamplingParams
from vllm.sampling_params import StructuredOutputsParams
設定工作負載
設定模型、地點、樣本大小及推論參數。
MODEL_NAME = "Qwen/Qwen2.5-32B-Instruct"
DATASET_NAME = "AmazonScience/massive"
MASSIVE_ARCHIVE_URL = "https://amazon-massive-nlu-dataset.s3.amazonaws.com/amazon-massive-dataset-1.1.tar.gz"
LOCALES = ["en-US", "es-ES", "de-DE", "ar-SA", "hi-IN", "ja-JP", "sw-KE", "zh-CN"]
ROWS_PER_LOCALE = 2_000
BATCH_SIZE = 64
MAX_MODEL_LEN = 512
MAX_OUTPUT_TOKENS = 8
SEED = 42
配置 Unity 目錄儲存
使用小工具指定現有的 Unity Catalog 目錄、結構描述和磁碟區。 此筆記本會將 MASSIVE 快取和 Parquet 預測結果儲存在此磁碟區中。 你需要以下特權：
-
USE CATALOG在目錄上，USE SCHEMA在綱要上。 - 音量上的
READ VOLUME和WRITE VOLUME。
每次 MLflow 執行都會將預測寫入其設定的 Parquet 輸出根目錄下的子目錄。
widget_defaults = {
"uc_catalog": "main",
"uc_schema": "default",
"uc_volume": "ray_data",
}
for widget_name, default_value in widget_defaults.items():
dbutils.widgets.text(widget_name, default_value)
CATALOG = dbutils.widgets.get("uc_catalog")
SCHEMA = dbutils.widgets.get("uc_schema")
VOLUME = dbutils.widgets.get("uc_volume")
volume_path = f"/Volumes/{CATALOG}/{SCHEMA}/{VOLUME}"
parquet_output_root = f"{volume_path}/sgc-raydata-vllm-batch-inference"
massive_cache_path = f"{volume_path}/hf-cache/amazon-massive-1.1"
print(f"Parquet output root: {parquet_output_root}")
print(f"Dataset cache: {massive_cache_path}")
啟動 Ray
ray_init() 在已連接的計算資源上啟動 Ray，並列印此筆記本的儀表板網址。 只要筆記本保持連線，Ray 連線就會維持作用中。 演員池使用 Ray 報告的 GPU 數量，因此每個可用 GPU 執行一個 vLLM 模型副本。
import ray
from serverless_gpu import ray_init
ray_context = ray_init()
ACTOR_COUNT = int(ray.cluster_resources().get("GPU", 0))
if ACTOR_COUNT < 1:
raise RuntimeError("Ray did not detect a GPU. Attach GPU compute and run the notebook again.")
print(f"Ray detected {ACTOR_COUNT} GPUs; using {ACTOR_COUNT} predictor actors.")
載入與取樣 MASSIVE
下載MASSIVE 1.1到設定好的快取，然後每次執行時從每個地點選取相同的2,000個訓練範例。 第一個地點也會提供用於建立分類提示的情境名稱與意圖名稱。
disable_progress_bar()
download_config = DownloadConfig(cache_dir=f"{massive_cache_path}/downloads")
download_manager = DownloadManager(download_config=download_config)
massive_archive_dir = Path(download_manager.download_and_extract(MASSIVE_ARCHIVE_URL))
massive_data_dir = massive_archive_dir / "1.1" / "data"
locale_datasets = []
scenario_names = None
scenario_intents = None
for locale in LOCALES:
locale_dataset = load_dataset(
"json",
data_files=str(massive_data_dir / f"{locale}.jsonl"),
split="train",
cache_dir=f"{massive_cache_path}/datasets",
)
locale_dataset = locale_dataset.filter(lambda row: row["partition"] == "train")
locale_scenarios = sorted(locale_dataset.unique("scenario"))
if scenario_names is not None and locale_scenarios != scenario_names:
raise ValueError(f"Scenario labels differ for locale {locale}.")
if scenario_names is None:
scenario_names = locale_scenarios
label_frame = locale_dataset.select_columns(["scenario", "intent"]).to_pandas()
scenario_intents = {
scenario: sorted(group["intent"].unique())
for scenario, group in label_frame.groupby("scenario")
}
sample = locale_dataset.shuffle(seed=SEED).select(range(ROWS_PER_LOCALE))
locale_datasets.append(sample.select_columns(["id", "locale", "utt", "scenario"]))
建立射線資料集
合併區域樣本，保留推論與評估所需的欄位，並重新分割資料，讓 Ray 能讓所有預測角色保持忙碌。
massive_sample = concatenate_datasets(locale_datasets)
records = [
{
"input_id": f"{row['locale']}:{row['id']}",
"locale": row["locale"],
"utterance": row["utt"],
"expected_scenario": row["scenario"],
}
for row in massive_sample
]
input_dataset = ray.data.from_items(records).repartition(ACTOR_COUNT * 8)
print(f"Prepared {len(records):,} records across {len(LOCALES)} locales and {len(scenario_names)} scenarios.")
定義 vLLM 預測器
MASSIVE 將發言分成 18 個情境，例如 alarm、 weather和 music。 此筆記本會根據資料集建立允許使用的標籤以及情境對應意圖的指引，而不是將它們硬編碼。
情境到意圖的映射幫助Qwen區分意義相似的標籤。 vLLM 會回傳允許的標籤之一，最後的正規化步驟會將其他回應標記為無效。
scenario_set = set(scenario_names)
scenario_guidance = "\n".join(
f"- {scenario}: {', '.join(scenario_intents[scenario])}"
for scenario in scenario_names
)
system_prompt = (
"Classify the user utterance into exactly one MASSIVE scenario. "
"Use these scenario-to-intent mappings to distinguish similar labels:\n"
f"{scenario_guidance}\n"
"Return only the scenario label."
)
def format_prompt(tokenizer, utterance: str) -> str:
messages = [
{"role": "system", "content": system_prompt},
{"role": "user", "content": utterance},
]
return tokenizer.apply_chat_template(messages, tokenize=False, add_generation_prompt=True)
def normalize_label(response: str) -> str | None:
normalized = re.sub(r"[^a-z]+", " ", response.lower()).strip()
return normalized if normalized in scenario_set else None
class VLLMPredictor:
def __init__(self):
gpu_ids = ray.get_runtime_context().get_accelerator_ids().get("GPU", [])
if len(gpu_ids) != 1:
raise RuntimeError(f"Expected one GPU per actor, but received {gpu_ids}.")
self.gpu_assignment = str(gpu_ids[0])
self.llm = LLM(
model=MODEL_NAME,
tensor_parallel_size=1,
dtype="bfloat16",
max_model_len=MAX_MODEL_LEN,
max_num_seqs=BATCH_SIZE,
gpu_memory_utilization=0.90,
enable_prefix_caching=True,
)
self.tokenizer = self.llm.get_tokenizer()
self.sampling_params = SamplingParams(
temperature=0.0,
max_tokens=MAX_OUTPUT_TOKENS,
structured_outputs=StructuredOutputsParams(choice=scenario_names),
)
def __call__(self, batch: pd.DataFrame) -> pd.DataFrame:
prompts = [format_prompt(self.tokenizer, utterance) for utterance in batch["utterance"]]
outputs = self.llm.generate(prompts, self.sampling_params, use_tqdm=False)
raw_responses = [output.outputs[0].text.strip() for output in outputs]
predicted_scenarios = [normalize_label(response) for response in raw_responses]
result = batch.copy()
result["raw_response"] = raw_responses
# Preserve invalid responses as nulls with a stable string type across batches.
result["predicted_scenario"] = pd.array(predicted_scenarios, dtype="string")
result["valid_prediction"] = result["predicted_scenario"].notna()
result["correct"] = (result["predicted_scenario"] == result["expected_scenario"]).fillna(False)
result["model_name"] = MODEL_NAME
result["ray_gpu_assignment"] = self.gpu_assignment
return result
執行並監控批次推論
VLLMPredictor 會在每個執行個體啟動時載入一次 Qwen，之後對其收到的每個批次都重複使用該模型。 Ray Data 會為每個偵測到的 GPU 啟動一個執行者，並將每個批次分派給下一個可用的執行者。
在推論執行期間，開啟第 10 個儲存格中由 ray_init() 輸出的 Ray 儀表板 URL。 利用儀表板檢查八個預測角色、GPU 保留、任務進度、日誌和落隊者。
predictions = input_dataset.map_batches(
VLLMPredictor,
batch_format="pandas",
batch_size=BATCH_SIZE,
compute=ray.data.ActorPoolStrategy(size=ACTOR_COUNT),
num_gpus=1,
)
具體化並追蹤結果
Ray Data 以延遲方式建立此管線，因此 write_parquet() 可在一步中執行推論並儲存結果。 Spark 接著讀取 Parquet 檔案進行評估，無需重新執行模型。 周邊的 MLflow 執行會擷取工作負載參數、品質指標、時序、吞吐量及系統指標，完成後 Databricks 會在儲存格下方新增可 (1 MLflow run) 點擊連結。
mlflow.set_system_metrics_sampling_interval(2)
with mlflow.start_run(run_name="raydata-massive-qwen25-32b", log_system_metrics=True) as active_run:
parquet_output_path = f"{parquet_output_root}/{active_run.info.run_id}"
print(f"Parquet output: {parquet_output_path}")
mlflow.log_params(
{
"model": MODEL_NAME,
"dataset": DATASET_NAME,
"dataset_version": "1.1",
"locales": json.dumps(LOCALES),
"record_count": len(records),
"actor_count": ACTOR_COUNT,
"batch_size": BATCH_SIZE,
"max_model_len": MAX_MODEL_LEN,
"max_output_tokens": MAX_OUTPUT_TOKENS,
"temperature": 0.0,
"output_constraint": "scenario_choices",
"system_metrics_interval_seconds": 2,
"gpu_memory_utilization": 0.90,
}
)
mlflow.set_tags(
{
"dataset_source": MASSIVE_ARCHIVE_URL,
"parquet_output_path": parquet_output_path,
}
)
start_time = time.perf_counter()
predictions.write_parquet(parquet_output_path)
cold_start_inclusive_duration_seconds = time.perf_counter() - start_time
results_df = spark.read.parquet(parquet_output_path)
aggregate = results_df.agg(
F.count("*").alias("record_count"),
F.avg(F.col("correct").cast("double")).alias("overall_accuracy"),
F.avg(F.col("valid_prediction").cast("double")).alias("valid_prediction_rate"),
F.countDistinct("ray_gpu_assignment").alias("unique_gpu_assignments"),
).first()
scenario_accuracy_df = results_df.groupBy("expected_scenario").agg(
F.count("*").alias("record_count"),
F.avg(F.col("correct").cast("double")).alias("accuracy"),
F.avg(F.col("valid_prediction").cast("double")).alias("valid_prediction_rate"),
).orderBy("expected_scenario")
macro_scenario_accuracy = scenario_accuracy_df.agg(F.avg("accuracy")).first()[0]
cold_start_inclusive_records_per_second = (
aggregate["record_count"] / cold_start_inclusive_duration_seconds
)
mlflow.log_metrics(
{
"overall_accuracy": aggregate["overall_accuracy"],
"macro_scenario_accuracy": macro_scenario_accuracy,
"valid_prediction_rate": aggregate["valid_prediction_rate"],
"cold_start_inclusive_duration_seconds": cold_start_inclusive_duration_seconds,
"cold_start_inclusive_records_per_second": cold_start_inclusive_records_per_second,
}
)
mlflow_run_id = active_run.info.run_id
print(f"MLflow run ID: {mlflow_run_id}")
print("Open the '(1 MLflow run)' link attached to this cell for parameters and metrics.")
驗證該結果
以下檢查可確認輸出中每個輸入各對應一列，且每個預測器 actor 都至少處理了一個批次。
計時會在 Ray 建立演員並載入模型之前開始，因此報告的持續時間與吞吐量包含冷啟動時間。
if aggregate["record_count"] != len(records):
raise RuntimeError("The persisted result count does not match the input count.")
if aggregate["unique_gpu_assignments"] != ACTOR_COUNT:
raise RuntimeError(f"Expected results from {ACTOR_COUNT} Ray GPU assignments.")
print(f"Records: {aggregate['record_count']:,}")
print(f"Overall accuracy: {aggregate['overall_accuracy']:.2%}")
print(f"Macro scenario accuracy: {macro_scenario_accuracy:.2%}")
print(f"Valid prediction rate: {aggregate['valid_prediction_rate']:.2%}")
print(f"Inference duration including actor and model cold start: {cold_start_inclusive_duration_seconds:.1f} seconds")
print(f"Throughput including actor and model cold start: {cold_start_inclusive_records_per_second:.1f} records/second")
print(f"Unique GPU assignments: {aggregate['unique_gpu_assignments']}")
分析預測品質
依地點、預測樣本及 GPU 角色間記錄分布顯示準確度。
locale_accuracy_df = (
results_df.groupBy("locale")
.agg(
F.count("*").alias("record_count"),
F.avg(F.col("correct").cast("double")).alias("accuracy"),
F.avg(F.col("valid_prediction").cast("double")).alias("valid_prediction_rate"),
)
.orderBy("locale")
)
print("Accuracy by locale:")
locale_accuracy_df.show(truncate=False)
prediction_columns = [
"locale", "utterance", "expected_scenario", "predicted_scenario",
"correct", "ray_gpu_assignment",
]
sample_predictions_df = (
results_df.select(prediction_columns)
.orderBy(F.rand(SEED))
.limit(16)
)
actor_distribution_df = (
results_df.groupBy("ray_gpu_assignment")
.agg(F.count("*").alias("record_count"))
.orderBy("ray_gpu_assignment")
)
print("Sample predictions:")
sample_predictions_df.show(truncate=80)
print("Records by Ray GPU assignment:")
actor_distribution_df.show(truncate=False)