# # Copyright (c) 2025 Huawei Technologies Co., Ltd. All Rights Reserved. # Copyright 2023 The vLLM team. # # Licensed under the Apache License, Version 2.0 (the "License"); # you may not use this file except in compliance with the License. # You may obtain a copy of the License at # # http://www.apache.org/licenses/LICENSE-2.0 # # Unless required by applicable law or agreed to in writing, software # distributed under the License is distributed on an "AS IS" BASIS, # WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. # See the License for the specific language governing permissions and # limitations under the License. # This file is a part of the vllm-ascend project. # Adapted from vllm-project/vllm/blob/main/tests/conftest.py # import contextlib import copy import functools import gc import json import logging import multiprocessing import os import shlex import subprocess import sys import threading import time import traceback from pathlib import Path from typing import Any, TypeVar import huggingface_hub import numpy as np import openai import psutil import pytest import requests import torch from modelscope import snapshot_download # type: ignore[import-untyped] from PIL import Image from requests.exceptions import RequestException from torch import nn from transformers import AutoConfig, AutoModelForCausalLM, AutoTokenizer, BatchEncoding, BatchFeature from transformers.models.auto.auto_factory import _BaseAutoModelClass from vllm import LLM, SamplingParams from vllm.config.model import ConvertOption, RunnerOption, _get_and_verify_dtype from vllm.inputs import TextPrompt from vllm.outputs import RequestOutput from vllm.platforms import current_platform from vllm.transformers_utils.utils import maybe_model_redirect from vllm.utils.network_utils import get_open_port from tests.e2e.model_utils import TokensTextLogprobs, TokensTextLogprobsPromptLogprobs from tests.e2e.nightly.multi_node.scripts.multi_node_config import DisaggregatedPrefillCfg, NodeInfo from vllm_ascend.ascend_config import clear_ascend_config # TODO: remove this part after the patch merged into vllm, if # we not explicitly patch here, some of them might be effectiveless # in pytest scenario from vllm_ascend.utils import adapt_patch # noqa E402 adapt_patch(True) adapt_patch(False) from vllm.distributed.parallel_state import ( # noqa E402 destroy_distributed_environment, destroy_model_parallel, ) _T = TypeVar("_T", nn.Module, torch.Tensor, BatchEncoding, BatchFeature, dict) _M = TypeVar("_M") _PromptMultiModalInput = list[_M] | list[list[_M]] PromptImageInput = _PromptMultiModalInput[Image.Image] PromptAudioInput = _PromptMultiModalInput[tuple[np.ndarray, int]] PromptVideoInput = _PromptMultiModalInput[np.ndarray] logger = logging.getLogger(__name__) _TEST_DIR = os.path.dirname(__file__) _LONG_PROMPTS = [os.path.join(_TEST_DIR, "prompts", "long_prompt.txt")] DISAGG_EPD_PROXY_SCRIPT = ( Path(__file__).parent.parent.parent / "examples" / "disaggregated_encoder" / "disagg_epd_proxy.py" ) def _check_npu_memory_worker(target_free_percentage: float, max_wait_seconds: float): # We can try to clean up memory in this subprocess, though it mostly affects this process. # But if there are any lingering contexts in this process (unlikely for a fresh spawn), it helps. gc.collect() torch.npu.empty_cache() _, total_npu_memory = torch.npu.mem_get_info() start_time = time.time() while True: free_bytes, _ = torch.npu.mem_get_info() if free_bytes / total_npu_memory >= target_free_percentage: print("check_npu_memory_worker: npu free memory decreased target value.") return # Success elapsed = time.time() - start_time if elapsed > max_wait_seconds: # Print to stderr so it's visible in test logs even if captured print( f"Timeout: NPU memory free size did not reach " f"{target_free_percentage} of total npu memory within {max_wait_seconds} seconds.", file=sys.stderr, ) sys.exit(1) # Failure print( f"Waiting for NPU memory to be free: " f"{free_bytes / 1024**3:.2f} GB available, " f"Elapsed time: {elapsed:.2f} s." ) # Try to clean up gc.collect() torch.npu.empty_cache() time.sleep(1) def wait_until_npu_memory_free(target_free_percentage: float = 0.5, max_wait_seconds: float = 50): """Decorator to wait until the NPU memory free size is above target_free_percentage. Args: target_free_percentage (float): Target free memory percentage of total. max_wait_seconds (float): Maximum wait time in seconds. """ def decorator(func): @functools.wraps(func) def wrapper(*args, **kwargs): # Clean up non-NPU resources in the main process cleanup_dist_env_and_memory() # Use a spawned subprocess to check NPU memory to avoid initializing NPU in the main process ctx = multiprocessing.get_context("spawn") p = ctx.Process(target=_check_npu_memory_worker, args=(target_free_percentage, max_wait_seconds)) p.start() p.join() if p.exitcode != 0: raise TimeoutError( f"Timeout: NPU memory free size did not reach " f"{target_free_percentage} of total npu memory within {max_wait_seconds} seconds." ) return func(*args, **kwargs) return wrapper return decorator def cleanup_dist_env_and_memory(shutdown_ray: bool = False): destroy_model_parallel() destroy_distributed_environment() with contextlib.suppress(AssertionError): torch.distributed.destroy_process_group() if shutdown_ray: import ray # Lazy import Ray ray.shutdown() gc.collect() # Only clean NPU cache if NPU is already initialized/available in this process. # This prevents accidental initialization of NPU context in the main process, # which would break subsequent forks. if hasattr(torch, "npu") and torch.npu.is_initialized(): torch.npu.empty_cache() torch.npu.reset_peak_memory_stats() class MooncakeLauncher: def __init__( self, mooncake_port, mooncake_metrics_port, eviction_high_watermark_ratio=0.8, eviction_ratio=0.05, ): self.mooncake_port = mooncake_port self.mooncake_metrics_port = mooncake_metrics_port self.eviction_high_watermark_ratio = eviction_high_watermark_ratio self.eviction_ratio = eviction_ratio def __enter__(self): cmd = [ "mooncake_master", "--eviction_high_watermark_ratio", str(self.eviction_high_watermark_ratio), "--eviction_ratio", str(self.eviction_ratio), "--port", str(self.mooncake_port), "--metrics_port", str(self.mooncake_metrics_port), ] logger.info("Launching mooncake: %s", " ".join(cmd)) curr_ld_path = os.environ.get("LD_LIBRARY_PATH", "") mooncake_ld_path = "/usr/local/Ascend/ascend-toolkit/latest/python/site-packages/mooncake:" os.environ["LD_LIBRARY_PATH"] = mooncake_ld_path + curr_ld_path env = os.environ.copy() self.process = subprocess.Popen(cmd, env=env) return self def __exit__(self, exc_type, exc, tb): if not self.process: return logger.info("Stopping mooncake server...") self.process.terminate() try: self.process.wait(timeout=5) except subprocess.TimeoutExpired: self.process.kill() class RemoteOpenAIServer: DUMMY_API_KEY = "token-abc123" # vLLM's OpenAI server does not need API key def _start_server(self, model: str, server_cmd: list[str], env_dict: dict[str, str] | None) -> None: """Subclasses override this method to customize server process launch""" env = os.environ.copy() # the current process might initialize npu, # to be safe, we should use spawn method env["VLLM_WORKER_MULTIPROC_METHOD"] = "spawn" if env_dict is not None: env.update(env_dict) logger.info(f"Starting server with command: {' '.join(server_cmd)}") self.proc: subprocess.Popen = subprocess.Popen( server_cmd, env=env, stdout=sys.stdout, stderr=sys.stderr, ) def __init__( self, model: str, vllm_serve_args: list[str] | str, *, server_host: str = "0.0.0.0", server_port: int = 8080, env_dict: dict[str, str] | None = None, seed: int | None = None, auto_port: bool = True, nodes_info: list[NodeInfo] | None = None, disaggregated_prefill: DisaggregatedPrefillCfg | None = None, proxy_port: int | None = None, max_wait_seconds: float | None = None, override_hf_configs: dict[str, Any] | None = None, ) -> None: if isinstance(vllm_serve_args, str): vllm_serve_args = shlex.split(vllm_serve_args) else: vllm_serve_args = ["vllm", "serve", model, *vllm_serve_args] if auto_port: if "-p" in vllm_serve_args or "--port" in vllm_serve_args: raise ValueError("You have manually specified the port when `auto_port=True`.") # No need for a port if using unix sockets if "--uds" not in vllm_serve_args: # Don't mutate the input args vllm_serve_args = vllm_serve_args + ["--port", str(get_open_port())] if seed is not None: if "--seed" in vllm_serve_args: raise ValueError(f"You have manually specified the seed when `seed={seed}`.") vllm_serve_args = vllm_serve_args + ["--seed", str(seed)] if override_hf_configs is not None: vllm_serve_args = vllm_serve_args + ["--hf-overrides", json.dumps(override_hf_configs)] self.host = str(server_host) self.port = int(server_port) # for multi-nodes test self.nodes_info = nodes_info self.disaggregated_prefill = disaggregated_prefill self.cur_index = os.getenv("LWS_WORKER_INDEX", 0) self.proxy_port = proxy_port self._start_server(model, vllm_serve_args, env_dict) max_wait_seconds = max_wait_seconds or 2800 if self.disaggregated_prefill: assert proxy_port is not None, "for disaggregated_prefill, proxy port must be provided" self._wait_for_server_pd(timeout=max_wait_seconds) else: self._wait_for_multiple_servers([(self.host, self.url_for("health"))], timeout=max_wait_seconds) def __enter__(self): return self def __exit__(self, exc_type, exc_value, traceback): self._terminate_server() def _poll(self) -> int | None: """Subclasses override this method to customize process polling""" return self.proc.poll() def hang_until_terminated(self, url) -> None: """ Wait until the server process terminates. This is for headless mode, where the api server process only exists in the leader node. """ logger.info("Hanging until server process terminates...") client = requests try: while True: try: resp = client.get(url, timeout=5) if resp.status_code != 200: break time.sleep(5) except Exception: break finally: self._terminate_server() def _wait_for_server_pd(self, timeout: float): # Wait for all api_server nodes ready assert self.nodes_info is not None, "cluster info must be provided" proxy_port = self.proxy_port def url_health(ip: str, port: int) -> str: return f"http://{ip}:{port}/health" targets = [ (node_info.ip, url_health(node_info.ip, self.port)) for node_info in self.nodes_info if not node_info.headless ] # Wait for proxy ready master_node = self.nodes_info[0] url_proxy = f"http://{master_node.ip}:{proxy_port}/healthcheck" # Wait for master node proxy first self._wait_for_multiple_servers([(master_node.ip, url_proxy)], timeout=timeout) # Then wait for all api_server nodes self._wait_for_multiple_servers(targets=targets, timeout=timeout) def _wait_for_multiple_servers( self, targets, timeout: float, log_interval: float = 30.0, always_check_nodes: bool = False ): """ targets: List[(node_ip, url)] log_interval """ start = time.time() client = requests ready = {node_ip: False for node_ip, _ in targets} last_log_time = 0.0 while True: now = time.time() all_ready = True should_log = (now - last_log_time) >= log_interval for node_ip, url in targets: if ready[node_ip] and not always_check_nodes: continue try: resp = client.get(url) if resp.status_code == 200: ready[node_ip] = True logger.info(f"[READY] Node {node_ip}: {url} is ready.") except RequestException: all_ready = False if should_log: logger.debug(f"[WAIT] {url}: connection failed") # check unexpected exit result = self._poll() if result is not None and result != 0: raise RuntimeError(f"Server at {node_ip} exited unexpectedly.") from None if should_log: last_log_time = now if all_ready: break if now - start > timeout: not_ready_nodes = [n for n, ok in ready.items() if not ok] self._terminate_server() raise RuntimeError( f"Timeout: these nodes did not become ready: {not_ready_nodes} in time: {timeout}s" ) from None time.sleep(5) @property def url_root(self) -> str: return f"http://{self.host}:{self.port}" def _terminate_server(self) -> None: """Subclasses override this method to customize server process termination""" self.proc.terminate() try: self.proc.wait(8) except subprocess.TimeoutExpired: # force kill if needed self.proc.kill() def url_for(self, *parts: str) -> str: return self.url_root + "/" + "/".join(parts) def get_client(self, **kwargs): if "timeout" not in kwargs: kwargs["timeout"] = 600 return openai.OpenAI( base_url=self.url_for("v1"), api_key=self.DUMMY_API_KEY, max_retries=0, **kwargs, ) def get_async_client(self, **kwargs): if "timeout" not in kwargs: kwargs["timeout"] = 600 return openai.AsyncOpenAI(base_url=self.url_for("v1"), api_key=self.DUMMY_API_KEY, max_retries=0, **kwargs) class RemoteEPDServer(RemoteOpenAIServer): def _start_server(self, model: str, server_cmd: list[str], env_dict: dict[str, str] | None) -> None: """Subclasses override this method to customize server process launch""" raise NotImplementedError("RemoteEPDServer should use _start_server_with_prefix instead") def __init__( self, vllm_serve_args: list[str] | list[list[str]], server_host: str = "0.0.0.0", env_dict: dict[str, str] | None = None, max_wait_seconds: float | None = 2800, ) -> None: self._proc_list = [] self.env_dict: dict[str, str] = {} if env_dict is not None: self.env_dict.update(env_dict) self.env_dict["VLLM_ALLOW_LONG_MAX_MODEL_LEN"] = "1" self.env_dict["PYTORCH_NPU_ALLOC_CONF"] = "expandable_segments:True" self.env_dict["VLLM_WORKER_MULTIPROC_METHOD"] = "spawn" self.vllm_serve_args_list = [] self.health_url_list = [] self.host = server_host if isinstance(vllm_serve_args, list): if not all(isinstance(item, list) for item in vllm_serve_args): args_copy = copy.deepcopy(vllm_serve_args) self.vllm_serve_args_list.append([str(arg) for arg in args_copy]) else: self.vllm_serve_args_list = [ [str(arg) for arg in sublist] for sublist in copy.deepcopy(vllm_serve_args) ] else: raise RuntimeError("vllm_serves_args must be a list") serve_arg_cmd = ["vllm", "serve"] for i, vllm_serve_arg in enumerate(self.vllm_serve_args_list): self.env_dict["ASCEND_RT_VISIBLE_DEVICES"] = str(i) if isinstance(vllm_serve_arg, list): if "--port" not in vllm_serve_arg: raise ValueError("You have manually specified the port ") else: port_arg = "--port" try: index = vllm_serve_arg.index(port_arg) except ValueError: raise ValueError(f"--port not found in args: {vllm_serve_arg}") port_str = vllm_serve_arg[index + 1] self.port = int(port_str) else: vllm_serve_arg_str = str(vllm_serve_arg) if "--port" not in vllm_serve_arg_str: raise ValueError("You have manually specified the port ") else: raise ValueError(f"Unexpected type for vllm_serve_arg: {type(vllm_serve_arg)}") self.health_url_list.append(super().url_for("health")) vllm_serve_arg = [*serve_arg_cmd, *vllm_serve_arg] proc = self._start_server_with_prefix(vllm_serve_arg, self.env_dict, f"[VLLM_{i}] ") self._proc_list.append(proc) timeout_value = float(max_wait_seconds) if max_wait_seconds is not None else 2800.0 super()._wait_for_multiple_servers( [(self.host, url) for url in self.health_url_list], timeout=timeout_value, always_check_nodes=True ) def _poll(self) -> int | None: return None def _delete_shm(self) -> None: for i, arg in enumerate(self.vllm_serve_args_list): if "--ec-transfer-config" in arg: index = arg.index("--ec-transfer-config") config_str = arg[index + 1] config_dict = json.loads(config_str) ec_connector_extra_config = config_dict.get("ec_connector_extra_config", {}) shm_path = ec_connector_extra_config.get("shared_storage_path") if shm_path: args = ["rm", "-r", "-f", str(shm_path)] print(f"delete shm_path is: {shm_path}") self._start_server_with_prefix(args, None, "[DELETE] ") def _read_output(self, pipe, prefix): try: with pipe: for line in iter(pipe.readline, ""): if line: print(f"{prefix}: {line}", end="") except Exception as e: print(f"error: {e}") traceback.print_exc() def _start_server_with_prefix(self, server_cmd: list[str], env_dict: dict[str, str] | None, log_prefix: str): env = os.environ.copy() if env_dict is not None: env.update(env_dict) proc = subprocess.Popen( server_cmd, env=env, stdout=subprocess.PIPE, stderr=subprocess.PIPE, universal_newlines=True, bufsize=1 ) stdout_thread = threading.Thread(target=self._read_output, args=(proc.stdout, log_prefix), daemon=True) stderr_thread = threading.Thread(target=self._read_output, args=(proc.stderr, log_prefix), daemon=True) stdout_thread.start() stderr_thread.start() return proc def _terminate_server(self) -> None: """kill process and its children""" print("vllm instance is stopping") for proc in self._proc_list: parent = psutil.Process(proc.pid) children = parent.children(recursive=True) for child in children: with contextlib.suppress(psutil.NoSuchProcess): child.terminate() gone, still_alive = psutil.wait_procs(children, timeout=10) for child in still_alive: with contextlib.suppress(psutil.NoSuchProcess): child.kill() try: parent.terminate() parent.wait(timeout=10) except (psutil.NoSuchProcess, psutil.TimeoutExpired): with contextlib.suppress(psutil.NoSuchProcess): parent.kill() def __enter__(self): """Context manager entry point.""" return self def __exit__(self, exc_type, exc_val, exc_tb): """Context manager exit point - clean up all processes.""" self._terminate_server() class DisaggEpdProxy(RemoteEPDServer): def __init__( self, proxy_args: list[str] | str | None = None, env_dict: dict[str, str] | None = None, server_host: str = "0.0.0.0", max_wait_seconds: float | None = 2800, ) -> None: if proxy_args is None: proxy_args_list: list[str] = [] elif isinstance(proxy_args, str): proxy_args_list = shlex.split(proxy_args) else: proxy_args_list = proxy_args self.proxy_args = proxy_args_list self.env_dict: dict[str, str] = {} if env_dict is not None: self.env_dict.update(env_dict) self._proc_list = list() self.host = server_host print(f"proxy param is: {self.proxy_args}") proxy_cmd = ["python", str(DISAGG_EPD_PROXY_SCRIPT), *self.proxy_args] proc = self._start_server_with_prefix(proxy_cmd, self.env_dict, "[PROXY] ") self._proc_list.append(proc) if "--port" not in self.proxy_args: raise ValueError("You have manually specified the port ") else: try: index = self.proxy_args.index("--port") except ValueError: raise ValueError("--port not found in proxy args") port_str = self.proxy_args[index + 1] self.port = int(port_str) timeout_value = float(max_wait_seconds) if max_wait_seconds is not None else 2800.0 super()._wait_for_multiple_servers([(self.host, super().url_for("health"))], timeout=timeout_value) def __enter__(self): """Context manager entry point.""" return self def __exit__(self, exc_type, exc_val, exc_tb): """Context manager exit point - clean up all processes.""" super()._terminate_server() _DP_RUNNER_START_TIMEOUT_SECONDS = 900.0 _DP_RUNNER_REQUEST_TIMEOUT_SECONDS = 900.0 _DP_RUNNER_SHUTDOWN_TIMEOUT_SECONDS = 30.0 def _split_data_parallel_indices(num_items: int, dp_size: int) -> list[list[int]]: if num_items < 0: raise ValueError("num_items must be non-negative") if dp_size <= 0: raise ValueError("dp_size must be positive") floor = num_items // dp_size remainder = num_items % dp_size def start(rank: int) -> int: return rank * floor + min(rank, remainder) return [list(range(start(rank), start(rank + 1))) for rank in range(dp_size)] def _slice_optional_inputs(inputs: PromptImageInput | PromptAudioInput | PromptVideoInput | None, indices: list[int]): if inputs is None: return None return [inputs[index] for index in indices] def _slice_list_inputs(items: list[Any], indices: list[int]) -> list[Any]: return [items[index] for index in indices] def _merge_data_parallel_results(total_items: int, shard_results: list[tuple[list[int], list[Any]]]) -> list[Any]: merged: list[Any] = [None] * total_items for indices, results in shard_results: if not indices: continue if len(indices) != len(results): raise RuntimeError("Mismatched result count returned by data parallel worker") for index, result in zip(indices, results): merged[index] = result if any(result is None for result in merged): raise RuntimeError("Some data parallel results were not returned") return merged def _normalize_score_inputs(text_1: str | list[str], text_2: str | list[str]) -> tuple[list[str], list[str]]: if isinstance(text_1, str) and isinstance(text_2, str): return [text_1], [text_2] if isinstance(text_1, str): return [text_1] * len(text_2), list(text_2) if isinstance(text_2, str): return list(text_1), [text_2] * len(text_1) if len(text_1) != len(text_2): raise ValueError("`text_1` and `text_2` must have the same length") return list(text_1), list(text_2) def _run_vllm_runner_dp_worker(conn, llm_kwargs: dict[str, Any], dp_rank: int, dp_size: int, master_port: int) -> None: llm = None try: os.environ["VLLM_DP_RANK"] = str(dp_rank) os.environ["VLLM_DP_RANK_LOCAL"] = str(dp_rank) os.environ["VLLM_DP_SIZE"] = str(dp_size) os.environ["VLLM_DP_MASTER_IP"] = "127.0.0.1" os.environ["VLLM_DP_MASTER_PORT"] = str(master_port) os.environ["VLLM_WORKER_MULTIPROC_METHOD"] = "spawn" llm = LLM(**llm_kwargs) conn.send({"status": "ready", "rank": dp_rank}) while True: request = conn.recv() command = request["command"] if command == "shutdown": break result: Any if command == "generate": req_outputs = llm.generate( request["inputs"], sampling_params=request["sampling_params"], **request["kwargs"] ) result = VllmRunner._finalize_generate_outputs(req_outputs) elif command == "generate_w_logprobs": req_outputs = llm.generate( request["inputs"], sampling_params=request["sampling_params"], **request["kwargs"] ) result = VllmRunner._final_steps_generate_w_logprobs(req_outputs) elif command == "classify": req_outputs = llm.classify(request["prompts"]) result = [req_output.outputs.probs for req_output in req_outputs] elif command == "embed": req_outputs = llm.embed(request["inputs"], *request["args"], **request["kwargs"]) result = [req_output.outputs.embedding for req_output in req_outputs] elif command == "encode": req_outputs = llm.encode(request["prompts"]) result = [req_output.outputs.data for req_output in req_outputs] elif command == "reward": req_outputs = llm.reward(request["prompts"]) result = [req_output.outputs.data for req_output in req_outputs] elif command == "score": req_outputs = llm.score(request["text_1"], request["text_2"], *request["args"], **request["kwargs"]) result = [req_output.outputs.score for req_output in req_outputs] else: raise ValueError(f"Unsupported data parallel command: {command}") conn.send({"status": "ok", "rank": dp_rank, "indices": request["indices"], "result": result}) except Exception: with contextlib.suppress(Exception): conn.send({"status": "error", "rank": dp_rank, "traceback": traceback.format_exc()}) raise finally: if llm is not None: del llm clear_ascend_config() cleanup_dist_env_and_memory() with contextlib.suppress(Exception): conn.close() class VllmRunner: def __init__( self, model_name: str, runner: RunnerOption = "auto", convert: ConvertOption = "auto", tokenizer_name: str | None = None, tokenizer_mode: str = "auto", max_model_len: int | None = 1024, dtype: str = "auto", disable_log_stats: bool = True, tensor_parallel_size: int = 1, block_size: int = 16, enable_chunked_prefill: bool = True, swap_space: int = 4, enforce_eager: bool | None = False, quantization: str | None = None, **kwargs, ) -> None: data_parallel_size = int(kwargs.get("data_parallel_size", 1)) if data_parallel_size > 1: raise ValueError("VllmRunner does not support `data_parallel_size > 1`; use `DPVllmRunner` instead.") self.model = LLM( model=model_name, runner=runner, convert=convert, tokenizer=tokenizer_name, tokenizer_mode=tokenizer_mode, trust_remote_code=True, dtype=dtype, swap_space=swap_space, enforce_eager=enforce_eager, disable_log_stats=disable_log_stats, tensor_parallel_size=tensor_parallel_size, max_model_len=max_model_len, block_size=block_size, enable_chunked_prefill=enable_chunked_prefill, quantization=quantization, **kwargs, ) @staticmethod def _finalize_generate_outputs(req_outputs: list[RequestOutput]) -> list[tuple[list[list[int]], list[str]]]: outputs: list[tuple[list[list[int]], list[str]]] = [] for req_output in req_outputs: prompt_str = req_output.prompt prompt_ids = req_output.prompt_token_ids req_sample_output_ids: list[list[int]] = [] req_sample_output_strs: list[str] = [] for sample in req_output.outputs: output_str = sample.text output_ids = list(sample.token_ids) req_sample_output_ids.append(prompt_ids + output_ids) req_sample_output_strs.append((prompt_str or "") + output_str) outputs.append((req_sample_output_ids, req_sample_output_strs)) return outputs def get_inputs( self, prompts: list[str] | list[torch.Tensor] | list[int], images: PromptImageInput | None = None, videos: PromptVideoInput | None = None, audios: PromptAudioInput | None = None, ) -> list[TextPrompt]: if any(x is not None and len(x) != len(prompts) for x in [images, videos, audios]): raise ValueError("All non-None multimodal inputs must have the same length as prompts") inputs = [] for i, prompt in enumerate(prompts): multi_modal_data = {} if images is not None and (image := images[i]) is not None: multi_modal_data["image"] = image if videos is not None and (video := videos[i]) is not None: multi_modal_data["video"] = video # type: ignore if audios is not None and (audio := audios[i]) is not None: multi_modal_data["audio"] = audio # type: ignore text_prompt_kwargs: dict[str, Any] = {"multi_modal_data": multi_modal_data or None} if isinstance(prompt, str): text_prompt_kwargs["prompt"] = prompt elif isinstance(prompt, list): text_prompt_kwargs["prompt_token_ids"] = prompt else: text_prompt_kwargs["prompt_embeds"] = prompt inputs.append(TextPrompt(**text_prompt_kwargs)) return inputs def generate( self, prompts: list[str] | list[torch.Tensor] | list[list[int]], sampling_params: SamplingParams, images: PromptImageInput | None = None, videos: PromptVideoInput | None = None, audios: PromptAudioInput | None = None, **kwargs: Any, ) -> list[tuple[list[list[int]], list[str]]]: inputs = self.get_inputs(prompts, images=images, videos=videos, audios=audios) req_outputs = self.model.generate(inputs, sampling_params=sampling_params, **kwargs) return self._finalize_generate_outputs(req_outputs) @staticmethod def _final_steps_generate_w_logprobs( req_outputs: list[RequestOutput], ) -> list[TokensTextLogprobsPromptLogprobs]: outputs: list[TokensTextLogprobsPromptLogprobs] = [] for req_output in req_outputs: assert len(req_output.outputs) > 0 for sample in req_output.outputs: output_str = sample.text output_ids = list(sample.token_ids) output_logprobs = sample.logprobs outputs.append((output_ids, output_str, output_logprobs, req_output.prompt_logprobs)) return outputs def generate_w_logprobs( self, prompts: list[str], sampling_params: SamplingParams, images: PromptImageInput | None = None, audios: PromptAudioInput | None = None, videos: PromptVideoInput | None = None, **kwargs: Any, ) -> list[TokensTextLogprobs] | list[TokensTextLogprobsPromptLogprobs]: inputs = self.get_inputs(prompts, images=images, videos=videos, audios=audios) req_outputs = self.model.generate(inputs, sampling_params=sampling_params, **kwargs) toks_str_logsprobs_prompt_logprobs = self._final_steps_generate_w_logprobs(req_outputs) # Omit prompt logprobs if not required by sampling params return ( [x[0:-1] for x in toks_str_logsprobs_prompt_logprobs] if sampling_params.prompt_logprobs is None else toks_str_logsprobs_prompt_logprobs ) def generate_greedy( self, prompts: list[str] | list[torch.Tensor] | list[list[int]], max_tokens: int, images: PromptImageInput | None = None, videos: PromptVideoInput | None = None, audios: PromptAudioInput | None = None, **kwargs: Any, ) -> list[tuple[list[int], str]]: greedy_params = SamplingParams(temperature=0.0, max_tokens=max_tokens) outputs = self.generate(prompts, greedy_params, images=images, videos=videos, audios=audios, **kwargs) return [(output_ids[0], output_str[0]) for output_ids, output_str in outputs] def generate_greedy_logprobs( self, prompts: list[str], max_tokens: int, num_logprobs: int | None, num_prompt_logprobs: int | None = None, images: PromptImageInput | None = None, audios: PromptAudioInput | None = None, videos: PromptVideoInput | None = None, stop_token_ids: list[int] | None = None, stop: list[str] | None = None, **kwargs: Any, ) -> list[TokensTextLogprobs] | list[TokensTextLogprobsPromptLogprobs]: greedy_logprobs_params = SamplingParams( temperature=0.0, max_tokens=max_tokens, logprobs=num_logprobs, prompt_logprobs=num_prompt_logprobs, stop_token_ids=stop_token_ids, stop=stop, ) return self.generate_w_logprobs( prompts, greedy_logprobs_params, images=images, audios=audios, videos=videos, **kwargs ) def classify(self, prompts: list[str]) -> list[list[float]]: req_outputs = self.model.classify(prompts) return [req_output.outputs.probs for req_output in req_outputs] def embed( self, prompts: list[str], images: PromptImageInput | None = None, videos: PromptVideoInput | None = None, audios: PromptAudioInput | None = None, *args, **kwargs, ) -> list[list[float]]: inputs = self.get_inputs(prompts, images=images, videos=videos, audios=audios) req_outputs = self.model.embed(inputs, *args, **kwargs) return [req_output.outputs.embedding for req_output in req_outputs] def encode(self, prompts: list[str]) -> list[list[float]]: req_outputs = self.model.encode(prompts) return [req_output.outputs.data for req_output in req_outputs] def reward(self, prompts: list[str]) -> list[list[float]]: req_outputs = self.model.reward(prompts) return [req_output.outputs.data for req_output in req_outputs] def score( self, text_1: str | list[str], text_2: str | list[str], *args, **kwargs, ) -> list[float]: req_outputs = self.model.score(text_1, text_2, *args, **kwargs) return [req_output.outputs.score for req_output in req_outputs] def __enter__(self): return self def __exit__(self, exc_type, exc_value, traceback): del self.model clear_ascend_config() cleanup_dist_env_and_memory() class DPVllmRunner(VllmRunner): def __init__( self, model_name: str, runner: RunnerOption = "auto", convert: ConvertOption = "auto", tokenizer_name: str | None = None, tokenizer_mode: str = "auto", max_model_len: int | None = 1024, dtype: str = "auto", disable_log_stats: bool = True, tensor_parallel_size: int = 1, block_size: int = 16, enable_chunked_prefill: bool = True, swap_space: int = 4, enforce_eager: bool | None = False, quantization: str | None = None, data_parallel_size: int = 2, **kwargs, ) -> None: if data_parallel_size < 2: raise ValueError("DPVllmRunner requires `data_parallel_size >= 2`") self._dp_size = data_parallel_size self._dp_parent_conns: list[Any] = [] self._dp_processes: list[Any] = [] self._dp_start_timeout = float(kwargs.pop("dp_start_timeout", _DP_RUNNER_START_TIMEOUT_SECONDS)) self._dp_request_timeout = float(kwargs.pop("dp_request_timeout", _DP_RUNNER_REQUEST_TIMEOUT_SECONDS)) llm_kwargs = dict( model=model_name, runner=runner, convert=convert, tokenizer=tokenizer_name, tokenizer_mode=tokenizer_mode, trust_remote_code=True, dtype=dtype, swap_space=swap_space, enforce_eager=enforce_eager, disable_log_stats=disable_log_stats, tensor_parallel_size=tensor_parallel_size, max_model_len=max_model_len, block_size=block_size, enable_chunked_prefill=enable_chunked_prefill, quantization=quantization, **kwargs, ) cleanup_dist_env_and_memory() self._start_data_parallel_workers(llm_kwargs) @property def model(self) -> LLM: raise RuntimeError("Direct access to `runner.model` is not supported by `DPVllmRunner`.") def _start_data_parallel_workers(self, llm_kwargs: dict[str, Any]) -> None: ctx = multiprocessing.get_context("spawn") master_port = get_open_port() try: for dp_rank in range(self._dp_size): parent_conn, child_conn = ctx.Pipe() proc = ctx.Process( target=_run_vllm_runner_dp_worker, args=(child_conn, llm_kwargs, dp_rank, self._dp_size, master_port), ) proc.start() child_conn.close() self._dp_parent_conns.append(parent_conn) self._dp_processes.append(proc) for rank, conn in enumerate(self._dp_parent_conns): if not conn.poll(self._dp_start_timeout): raise TimeoutError(f"Timed out waiting for data parallel worker {rank} to start") message = conn.recv() if message["status"] != "ready": raise RuntimeError( f"Failed to start data parallel worker {rank}:\n{message.get('traceback', 'unknown error')}" ) except Exception: self._stop_data_parallel_workers() raise def _stop_data_parallel_workers(self) -> None: for conn in self._dp_parent_conns: with contextlib.suppress(Exception): conn.send({"command": "shutdown"}) for proc in self._dp_processes: proc.join(timeout=_DP_RUNNER_SHUTDOWN_TIMEOUT_SECONDS) if proc.is_alive(): proc.kill() proc.join(timeout=5) for conn in self._dp_parent_conns: with contextlib.suppress(Exception): conn.close() self._dp_parent_conns.clear() self._dp_processes.clear() def _dispatch_prompt_command( self, command: str, prompts: list[str] | list[torch.Tensor] | list[list[int]], *, images: PromptImageInput | None = None, videos: PromptVideoInput | None = None, audios: PromptAudioInput | None = None, **payload: Any, ) -> list[Any]: if not prompts: return [] shard_results: list[tuple[list[int], list[Any]]] = [] shard_indices = _split_data_parallel_indices(len(prompts), self._dp_size) for rank, conn in enumerate(self._dp_parent_conns): indices = shard_indices[rank] worker_indices = indices or [0] worker_prompts = _slice_list_inputs(prompts, worker_indices) conn.send( { "command": command, "indices": indices, "inputs": self.get_inputs( worker_prompts, images=_slice_optional_inputs(images, worker_indices), videos=_slice_optional_inputs(videos, worker_indices), audios=_slice_optional_inputs(audios, worker_indices), ), "prompts": worker_prompts, **payload, } ) try: for rank, conn in enumerate(self._dp_parent_conns): if not conn.poll(self._dp_request_timeout): raise TimeoutError(f"Timed out waiting for data parallel worker {rank} to finish `{command}`") message = conn.recv() if message["status"] != "ok": raise RuntimeError( f"Data parallel worker {rank} failed during `{command}`:\n" f"{message.get('traceback', 'unknown error')}" ) shard_results.append((message["indices"], message["result"])) except Exception: self._stop_data_parallel_workers() raise return _merge_data_parallel_results(len(prompts), shard_results) def _dispatch_text_command(self, command: str, prompts: list[str]) -> list[Any]: if not prompts: return [] shard_results: list[tuple[list[int], list[Any]]] = [] shard_indices = _split_data_parallel_indices(len(prompts), self._dp_size) for rank, conn in enumerate(self._dp_parent_conns): indices = shard_indices[rank] worker_indices = indices or [0] conn.send( { "command": command, "indices": indices, "prompts": _slice_list_inputs(prompts, worker_indices), } ) try: for rank, conn in enumerate(self._dp_parent_conns): if not conn.poll(self._dp_request_timeout): raise TimeoutError(f"Timed out waiting for data parallel worker {rank} to finish `{command}`") message = conn.recv() if message["status"] != "ok": raise RuntimeError( f"Data parallel worker {rank} failed during `{command}`:\n" f"{message.get('traceback', 'unknown error')}" ) shard_results.append((message["indices"], message["result"])) except Exception: self._stop_data_parallel_workers() raise return _merge_data_parallel_results(len(prompts), shard_results) def generate( self, prompts: list[str] | list[torch.Tensor] | list[list[int]], sampling_params: SamplingParams, images: PromptImageInput | None = None, videos: PromptVideoInput | None = None, audios: PromptAudioInput | None = None, **kwargs: Any, ) -> list[tuple[list[list[int]], list[str]]]: return self._dispatch_prompt_command( "generate", prompts, images=images, videos=videos, audios=audios, sampling_params=sampling_params, kwargs=kwargs, ) def generate_w_logprobs( self, prompts: list[str], sampling_params: SamplingParams, images: PromptImageInput | None = None, audios: PromptAudioInput | None = None, videos: PromptVideoInput | None = None, **kwargs: Any, ) -> list[TokensTextLogprobs] | list[TokensTextLogprobsPromptLogprobs]: toks_str_logsprobs_prompt_logprobs = self._dispatch_prompt_command( "generate_w_logprobs", prompts, images=images, videos=videos, audios=audios, sampling_params=sampling_params, kwargs=kwargs, ) return ( [x[0:-1] for x in toks_str_logsprobs_prompt_logprobs] if sampling_params.prompt_logprobs is None else toks_str_logsprobs_prompt_logprobs ) def classify(self, prompts: list[str]) -> list[list[float]]: return self._dispatch_text_command("classify", prompts) def embed( self, prompts: list[str], images: PromptImageInput | None = None, videos: PromptVideoInput | None = None, audios: PromptAudioInput | None = None, *args, **kwargs, ) -> list[list[float]]: return self._dispatch_prompt_command( "embed", prompts, images=images, videos=videos, audios=audios, args=args, kwargs=kwargs, ) def encode(self, prompts: list[str]) -> list[list[float]]: return self._dispatch_text_command("encode", prompts) def reward(self, prompts: list[str]) -> list[list[float]]: return self._dispatch_text_command("reward", prompts) def score( self, text_1: str | list[str], text_2: str | list[str], *args, **kwargs, ) -> list[float]: normalized_text_1, normalized_text_2 = _normalize_score_inputs(text_1, text_2) if not normalized_text_1: return [] shard_results: list[tuple[list[int], list[Any]]] = [] shard_indices = _split_data_parallel_indices(len(normalized_text_1), self._dp_size) for rank, conn in enumerate(self._dp_parent_conns): indices = shard_indices[rank] worker_indices = indices or [0] conn.send( { "command": "score", "indices": indices, "text_1": _slice_list_inputs(normalized_text_1, worker_indices), "text_2": _slice_list_inputs(normalized_text_2, worker_indices), "args": args, "kwargs": kwargs, } ) try: for rank, conn in enumerate(self._dp_parent_conns): if not conn.poll(self._dp_request_timeout): raise TimeoutError(f"Timed out waiting for data parallel worker {rank} to finish `score`") message = conn.recv() if message["status"] != "ok": raise RuntimeError( f"Data parallel worker {rank} failed during `score`:\n" f"{message.get('traceback', 'unknown error')}" ) shard_results.append((message["indices"], message["result"])) except Exception: self._stop_data_parallel_workers() raise return _merge_data_parallel_results(len(normalized_text_1), shard_results) def __exit__(self, exc_type, exc_value, traceback): self._stop_data_parallel_workers() clear_ascend_config() cleanup_dist_env_and_memory() DataParallelVllmRunner = DPVllmRunner class HfRunner: def get_default_device(self): return "cpu" if current_platform.is_cpu() else current_platform.device_type def wrap_device(self, x: _T, device: str | None = None) -> _T: if x is None or isinstance(x, (bool,)): return x if device is None: device = self.device if isinstance(x, dict): return {k: self.wrap_device(v, device) for k, v in x.items()} if hasattr(x, "device") and x.device.type == device: return x return x.to(device) def __init__( self, model_name: str, dtype: str = "auto", *, model_kwargs: dict[str, Any] | None = None, trust_remote_code: bool = True, is_sentence_transformer: bool = False, is_cross_encoder: bool = False, skip_tokenizer_init: bool = False, auto_cls: type[_BaseAutoModelClass] = AutoModelForCausalLM, ) -> None: model_name = maybe_model_redirect(model_name) self.model_name = model_name self.config = AutoConfig.from_pretrained( model_name, trust_remote_code=trust_remote_code, ) self.device = self.get_default_device() self.dtype = torch_dtype = _get_and_verify_dtype( self.model_name, self.config, dtype=dtype, is_pooling_model=is_sentence_transformer or is_cross_encoder, ) model_kwargs = model_kwargs if model_kwargs is not None else {} model_kwargs.setdefault("torch_dtype", torch_dtype) if is_sentence_transformer: # Lazy init required for AMD CI from sentence_transformers import SentenceTransformer self.model = SentenceTransformer( model_name, device=self.device, model_kwargs=model_kwargs, trust_remote_code=trust_remote_code, ) elif is_cross_encoder: # Lazy init required for AMD CI from sentence_transformers import CrossEncoder self.model = CrossEncoder( model_name, device=self.device, automodel_args=model_kwargs, trust_remote_code=trust_remote_code, ) else: model = auto_cls.from_pretrained( model_name, trust_remote_code=trust_remote_code, **model_kwargs, ) # in case some unquantized custom models are not in same dtype if getattr(model, "quantization_method", None) is None and any( p.dtype != self.dtype for p in model.parameters() ): model = model.to(dtype=self.dtype) if ( getattr(model, "quantization_method", None) != "bitsandbytes" and len({p.device for p in model.parameters()}) < 2 ): model = model.to(device=self.device) self.model = model if not skip_tokenizer_init: self.tokenizer = AutoTokenizer.from_pretrained( model_name, torch_dtype=torch_dtype, trust_remote_code=trust_remote_code, ) # don't put this import at the top level # it will call torch.cuda.device_count() from transformers import AutoProcessor # noqa: F401 self.processor = AutoProcessor.from_pretrained( model_name, torch_dtype=torch_dtype, trust_remote_code=trust_remote_code, ) if skip_tokenizer_init: self.tokenizer = self.processor.tokenizer def get_inputs( self, prompts: list[str], images: PromptImageInput | None = None, videos: PromptVideoInput | None = None, audios: PromptAudioInput | None = None, ) -> list[BatchFeature | BatchEncoding]: if images is not None: assert len(prompts) == len(images) if videos is not None: assert len(prompts) == len(videos) if audios is not None: assert len(prompts) == len(audios) all_inputs: list[BatchFeature | BatchEncoding] = [] for i, prompt in enumerate(prompts): processor_kwargs: dict[str, Any] = { "text": prompt, "return_tensors": "pt", } if images is not None and (image := images[i]) is not None: processor_kwargs["images"] = image if videos is not None and (video := videos[i]) is not None: processor_kwargs["videos"] = video if audios is not None and (audio_inputs := audios[i]) is not None: # HACK - not all processors take sampling_rate; we should # clean this up in the future. if len(audio_inputs) == 2: audio, sr = audio_inputs processor_kwargs["audio"] = audio processor_kwargs["sampling_rate"] = sr else: processor_kwargs["audio"] = audio_inputs inputs = self.processor(**processor_kwargs) if isinstance(inputs, BatchFeature): inputs = inputs.to(dtype=self.dtype) all_inputs.append(inputs) return all_inputs def classify(self, prompts: list[str]) -> list[str]: # output is final logits all_inputs = self.get_inputs(prompts) outputs = [] problem_type = getattr(self.config, "problem_type", "") for inputs in all_inputs: output = self.model(**self.wrap_device(inputs)) if problem_type == "regression": logits = output.logits[0].tolist() elif problem_type == "multi_label_classification": logits = output.logits.sigmoid()[0].tolist() else: logits = output.logits.softmax(dim=-1)[0].tolist() outputs.append(logits) return outputs def encode(self, prompts: list[str], *args, **kwargs) -> list[list[torch.Tensor]]: return self.model.encode(prompts, *args, **kwargs) def predict(self, prompts: list[list[str]], *args, **kwargs) -> torch.Tensor: return self.model.predict(prompts, *args, convert_to_tensor=True, **kwargs) def __enter__(self): return self def __exit__(self, exc_type, exc_value, traceback): del self.model cleanup_dist_env_and_memory() @pytest.fixture(scope="session") def ilama_lora_files(): return snapshot_download( repo_id="vllm-ascend/ilama-text2sql-spider", local_files_only=huggingface_hub.constants.HF_HUB_OFFLINE, ) @pytest.fixture(scope="session") def llama32_lora_files(): from huggingface_hub import snapshot_download as hf_snapshot_download return hf_snapshot_download(repo_id="jeeejeee/llama32-3b-text2sql-spider", local_files_only=True) def qwen_prompt(questions: list[str]) -> list[str]: placeholder = "<|image_pad|>" return [ ( "<|im_start|>system\nYou are a helpful assistant.<|im_end|>\n" f"<|im_start|>user\n<|vision_start|>{placeholder}<|vision_end|>" f"{q}<|im_end|>\n<|im_start|>assistant\n" ) for q in questions ] def hunyuan_prompt(questions: list[str]) -> list[str]: placeholder = "<｜hy_place▁holder▁no▁100｜><｜hy_place▁holder▁no▁102｜><｜hy_place▁holder▁no▁101｜>" # noqa: E501 return [f"<｜hy_begin▁of▁sentence｜>{placeholder}{question}<｜hy_User｜>" for question in questions] PROMPT_CONFIGS = { "qwen-vl": { "model": "Qwen/Qwen3-VL-8B-Instruct", "prompt_fn": qwen_prompt, "mm_processor_kwargs": { "min_pixels": 28 * 28, "max_pixels": 1280 * 28 * 28, "fps": 1, }, }, "hunyuan-vl": { "model": "Tencent-Hunyuan/HunyuanOCR", "prompt_fn": hunyuan_prompt, "mm_processor_kwargs": {}, }, } @pytest.fixture(params=PROMPT_CONFIGS.keys()) def vl_config(request): config = PROMPT_CONFIGS[request.param] if "skip" in config: pytest.skip(config["skip"]) return config