这里的“覆盖”指什么
教程覆盖当前固定版本中扫描到的原生语法、标准库接口及有接收者线索的方法，每节末尾都有全部相关源码位置。标准库知识先用小实验讲清，再到完整源码查看它在业务里的输入、交接和退出。
同名的 .get()、.join()、.close() 不能单凭名字认定为并发接口。没有可靠接收者线索的记录仍放在第 21 节的核对表中，保留原始代码，不混入已确认的 API 数量。
固定版本 2f2318653f6f794dddd321eff7c9d4b7b174643f，共扫描 381 个 Python 文件。下面的“对应”表示有明确讲解入口，不表示执行过 Slime 训练或证明了所有运行时类型。
| 记录 | 数量 | 怎样阅读 |
|---|---|---|
| 直接标准库 API | 518 个位置 / 83 个名称 | 逐项对应下面的讲解 |
| 异步语法 | 747 | 定义、等待、资源作用域、异步迭代 |
| 导入 | 147 | 对应所属机制的入口 |
| 有接收者线索的方法 | 206 | 按构造或注解对应具体接口；仍保留推断标记 |
| 其余同名方法候选 | 670 | 到第 21 节核对类型，不算作已确认的原生 API |
| Markdown 中的示例 | 29 | 本页下方保留全文及对应讲解 |
每种用法到哪里学习
| 讲解 | 对应的 API、方法与语法（含理解所需的补充接口） | 源码位置数 |
|---|---|---|
| 定义协程函数与创建协程对象 | async_def | 281 |
| 运行顶层协程与事件循环入口 | asyncio asyncio.run | 112 |
| 沿当前调用链等待与恢复 | asyncio.sleep await | 442 |
| 阻塞线程与挂起任务的区别 | time.sleep | 57 |
| 创建任务与保留引用 | asyncio.create_task asyncio.Task | 24 |
| 收集整组结果与异常边界 | asyncio.gather | 27 |
| 等待条件与完成集合 | asyncio.wait asyncio.FIRST_COMPLETED | 8 |
| 按完成顺序消费结果 | asyncio.as_completed | 1 |
| 请求取消与等待清理 | asyncio.CancelledError | 11 |
| 超时与取消保护的边界 | asyncio.TimeoutError asyncio.timeout asyncio.wait_for asyncio.shield | 18 |
| 许可区域、限流与归还 | asyncio.Semaphore asyncio.Semaphore.acquire asyncio.Semaphore.release | 11 |
| 异步队列的生产消费与背压 | asyncio.Queue asyncio.Queue.put asyncio.Queue.get asyncio.Queue.empty | 4 |
| 同步队列与立刻尝试取结果 | queue.Queue queue.Empty queue.Queue.put queue.Queue.get queue.Queue.get_nowait queue.Queue.qsize queue.Queue.empty | 15 |
| 把同步函数交给工作线程 | asyncio.to_thread asyncio.AbstractEventLoop.run_in_executor | 15 |
| 线程启动、互斥、通知与退出 | threading.Thread threading.Lock threading.Event threading.get_ident threading.Thread.start threading.Thread.join threading.Thread.is_alive threading.Lock.acquire threading.Lock.release threading.Event.wait threading.Event.set threading.Event.clear threading.Event.is_set | 120 |
| 异步进入、退出与资源回收 | contextlib.asynccontextmanager async_with | 48 |
| 异步迭代与异步生成器 | async_for async_generator_def async_yield | 15 |
| 任务上下文与 token 恢复 | contextvars.ContextVar contextvars.ContextVar.get contextvars.ContextVar.set contextvars.ContextVar.reset | 18 |
| 进程池分配与收集工作 | multiprocessing.Pool.starmap multiprocessing.context.BaseContext.Pool | 2 |
| 回到 Slime 的创建、交接与清理 | 0 | |
| Event 的等待、通知、状态与清零 | asyncio.Event asyncio.Event.wait asyncio.Event.set asyncio.Event.clear asyncio.Event.is_set | 35 |
| asyncio.Lock 保护任务共享状态 | asyncio.Lock asyncio.Lock.acquire asyncio.Lock.release asyncio.Lock.locked | 1 |
| Condition 谓词循环、锁交接与通知 | threading.Condition threading.Condition.wait threading.Condition.wait_for threading.Condition.notify_all | 24 |
| 同一线程重入 RLock | threading.RLock threading.RLock.acquire threading.RLock.release | 6 |
| Timer 启动、取消与退出 | threading.Timer threading.Timer.start threading.Timer.cancel threading.Timer.join | 3 |
| 线程池 submit 与 map 的结果组织 | concurrent.futures.ThreadPoolExecutor concurrent.futures.ThreadPoolExecutor.submit concurrent.futures.ThreadPoolExecutor.map concurrent.futures.Executor.submit concurrent.futures.Executor.map | 39 |
| 同步 Future 的结果、异常与完成状态 | concurrent.futures.CancelledError concurrent.futures.Future.result concurrent.futures.Future.exception concurrent.futures.Future.done | 11 |
| Future 取消边界与完成回调 | concurrent.futures.Future.cancel concurrent.futures.Future.cancelled concurrent.futures.Future.running concurrent.futures.Future.add_done_callback | 0 |
| 同步 wait 与 FIRST_COMPLETED | concurrent.futures.wait concurrent.futures.FIRST_COMPLETED | 4 |
| 按完成次序消费线程池结果 | concurrent.futures.as_completed | 4 |
| 线程池关闭与在途工作清理 | concurrent.futures.ThreadPoolExecutor.shutdown concurrent.futures.Executor.shutdown | 2 |
| 测试替身直接构造 Future 并设置结果 | concurrent.futures.Future concurrent.futures.Future.set_result concurrent.futures.Future.set_exception | 5 |
| 创建进程、启动上下文与所有权 | multiprocessing.Process multiprocessing.get_context multiprocessing.set_start_method multiprocessing.active_children multiprocessing.Process.start multiprocessing.Process.is_alive multiprocessing.context.BaseContext.Process multiprocessing.context.BaseContext.Queue | 20 |
| 跨进程生产消费与结束协议 | multiprocessing.Queue queue.Empty multiprocessing.Queue.put multiprocessing.Queue.get multiprocessing.Queue.empty multiprocessing.Queue.qsize | 3 |
| 等待、终止、回收与队列关闭 | multiprocessing.Process.join multiprocessing.Process.terminate multiprocessing.Process.kill multiprocessing.Process.close multiprocessing.Queue.close multiprocessing.Queue.join_thread | 0 |
| 进程池分派与退出 | multiprocessing.Pool multiprocessing.context.BaseContext.Pool multiprocessing.Pool.starmap multiprocessing.Pool.close multiprocessing.Pool.terminate multiprocessing.Pool.join multiprocessing.Pool.__enter__ multiprocessing.Pool.__exit__ | 0 |
| 子进程通信、超时与回收 | subprocess.Popen subprocess.TimeoutExpired subprocess subprocess.Popen.communicate subprocess.Popen.poll subprocess.Popen.wait subprocess.Popen.kill subprocess.Popen.terminate | 52 |
| 捕获、合并与丢弃输出 | subprocess.PIPE subprocess.STDOUT subprocess.DEVNULL | 16 |
| 同步命令检查与失败状态 | subprocess.run subprocess.check_call subprocess.check_output subprocess.CalledProcessError | 22 |
| 跨进程文件锁与目录所有权 | fcntl.flock fcntl.LOCK_EX fcntl.LOCK_NB fcntl.LOCK_SH | 9 |
| 管道可读等待与就绪协议 | select.select | 2 |
| 进程退出与解锁 | fcntl.LOCK_UN | 1 |
| 父子身份与定向信号 | os.getpid os.getppid os.kill | 35 |
| 保存、安装与恢复信号处理器 | signal.signal signal.getsignal signal.SIGINT signal.SIGTERM signal.SIGHUP signal.SIGKILL | 14 |
| 进程组信号与清理范围 | os.killpg signal.SIG_IGN | 2 |
| 同步工作闹钟与恢复边界 | signal.alarm signal.SIGALRM | 3 |
| 同步 shell 调用与退出状态 | os.system | 4 |
| Task、线程、进程和进程组的不同停止责任 | 0 | |
| Future：待完成结果与所属 loop | asyncio.get_running_loop asyncio.Future asyncio.InvalidStateError asyncio.AbstractEventLoop.create_future asyncio.Future.result | 13 |
| Future 完成、取消与分批回调 | asyncio.AbstractEventLoop.call_soon asyncio.AbstractEventLoop.call_later asyncio.Future.done asyncio.Future.set_result asyncio.Future.set_exception asyncio.Future.cancel asyncio.Future.cancelled asyncio.Future.add_done_callback | 4 |
| 创建、登记、运行与跨线程提交 | asyncio.AbstractEventLoop asyncio.new_event_loop asyncio.set_event_loop asyncio.run_coroutine_threadsafe asyncio.wrap_future asyncio.AbstractEventLoop.run_forever asyncio.AbstractEventLoop.run_until_complete concurrent.futures.Future.result | 23 |
| 事件循环停止、executor 排空与关闭 | asyncio.AbstractEventLoop.call_soon_threadsafe asyncio.AbstractEventLoop.is_running asyncio.AbstractEventLoop.stop asyncio.AbstractEventLoop.close asyncio.AbstractEventLoop.shutdown_default_executor | 2 |
| Task 身份、归属与完成读取 | asyncio.current_task asyncio.all_tasks asyncio.Task asyncio.CancelledError asyncio.Task.done asyncio.Task.result asyncio.Task.exception asyncio.Task.cancel asyncio.Task.cancelled asyncio.Task.get_loop asyncio.Task.add_done_callback | 17 |
| 手动 loop 的异步生成器兜底清理 | asyncio.AbstractEventLoop.shutdown_asyncgens | 0 |
| 创建本地监听与流式连接 | asyncio.start_server asyncio.open_connection asyncio.Server asyncio.StreamReader asyncio.StreamWriter | 1 |
| 流式读取与协议分帧 | asyncio.LimitOverrunError asyncio.StreamReader.read asyncio.StreamReader.readline asyncio.StreamReader.readuntil asyncio.StreamReader.readexactly | 0 |
| 提前 EOF 的部分数据与期待长度 | asyncio.IncompleteReadError | 1 |
| 写入流量控制与连接关闭 | asyncio.StreamWriter.write asyncio.StreamWriter.drain asyncio.StreamWriter.close asyncio.StreamWriter.wait_closed | 0 |
| 监听器与连接任务的独立收尾 | asyncio.Server.is_serving asyncio.Server.serve_forever asyncio.Server.close asyncio.Server.wait_closed | 0 |
| 调用后检查返回对象能否 await | inspect.isawaitable | 3 |
| 装饰时判断协程函数并选择 wrapper | inspect.iscoroutinefunction | 1 |
| Awaitable 和 AsyncIterator 注解契约 | collections.abc.Awaitable collections.abc.AsyncIterator | 7 |
| 上下文与迭代的异步协议入口 | builtins.StopAsyncIteration contextlib.AbstractAsyncContextManager.__aenter__ contextlib.AbstractAsyncContextManager.__aexit__ collections.abc.AsyncIterable.__aiter__ collections.abc.AsyncIterator.__anext__ async_protocol_def | 0 |
| 提前退出异步生成器时显式关闭 | contextlib.aclosing collections.abc.AsyncGenerator.aclose | 0 |
| 同名方法与原生接收者证据边界 | dict.get str.join queue.Queue.get queue.Queue.join asyncio.Queue.get asyncio.Queue.join threading.Thread.join | 670 |
源码位置数只计算扫描到的记录；部分接口为理解完整机制而补充，不能据此认定 Slime 使用了表内每个接口。
文档中的全部 Python 示例
以下代码出自 Slime 文档，保留原文；其中可能有省略号、业务对象和第三方依赖，不能当作独立可运行实验。每节下载的 example.py 才是验证过的最小实验。
DOC-001 · .claude/skills/add-reward-function/SKILL.md:36
固定版本的原文 · 对应讲解： 定义协程函数与创建协程对象
async def custom_rm(args, sample):
return float_reward_or_reward_dict
DOC-002 · .claude/skills/add-reward-function/SKILL.md:41
固定版本的原文 · 对应讲解： 定义协程函数与创建协程对象
async def custom_rm(args, samples):
return list_of_rewards
DOC-003 · docs/en/advanced/sglang-config.md:173
固定版本的原文 · 对应讲解： 定义协程函数与创建协程对象 沿当前调用链等待与恢复
from slime.rollout.sglang_rollout import get_model_url
from slime.utils.http_utils import post
async def my_generate(args, sample, sampling_params):
# Route to the actor model (default)
actor_url = get_model_url(args, "actor", "/generate")
output = await post(actor_url, {"text": sample.prompt, "sampling_params": sampling_params})
# Route to the reference model
ref_url = get_model_url(args, "ref", "/generate")
ref_output = await post(ref_url, {"text": sample.prompt, "sampling_params": sampling_params})
# Route to the reward model (e.g., OpenAI-compatible API)
reward_url = get_model_url(args, "reward", "/v1/chat/completions")
reward_output = await post(reward_url, {...})
...
DOC-004 · docs/en/advanced/sglang-config.md:402
固定版本的原文 · 对应讲解： 定义协程函数与创建协程对象 沿当前调用链等待与恢复
from slime.rollout.sglang_rollout import get_model_url
from slime.utils.http_utils import post
async def generate_with_models(args, sample, sampling_params):
"""Generate using actor, score with reward model, compare with reference."""
# Generate from actor
actor_url = get_model_url(args, "actor", "/generate")
actor_output = await post(actor_url, {
"text": sample.prompt,
"sampling_params": sampling_params,
"return_logprob": True,
})
# Get reference logprobs for KL penalty
ref_url = get_model_url(args, "ref", "/generate")
ref_output = await post(ref_url, {
"text": sample.prompt + actor_output["text"],
"sampling_params": {"max_new_tokens": 0, "temperature": 0},
"return_logprob": True,
})
# Score with reward model
reward_url = get_model_url(args, "reward", "/v1/chat/completions")
reward_output = await post(reward_url, {
"model": "reward",
"messages": [{"role": "user", "content": sample.prompt + actor_output["text"]}],
})
# ... process outputs and return Sample
DOC-005 · docs/en/developer_guide/trace.md:60
固定版本的原文 · 对应讲解： 定义协程函数与创建协程对象
from slime.observability.trace_utils import trace_function
@trace_function("generate_and_rm", target="sample")
async def generate_and_rm(args, sample, sampling_params, evaluation=False):
...
@trace_function(
"generate_and_rm_group",
target="group",
attrs_getter=lambda args, group, sampling_params, evaluation=False: {"group_size": len(group)},
)
async def generate_and_rm_group(args, group, sampling_params, evaluation=False):
...
DOC-006 · docs/en/developer_guide/trace.md:90
固定版本的原文 · 对应讲解： 定义协程函数与创建协程对象
@trace_function(
"custom_rollout_batch",
target="samples",
attrs_getter=lambda samples, **_: {"batch_size": len(samples)},
)
async def custom_rollout_batch(samples, **kwargs):
...
DOC-007 · docs/en/developer_guide/trace.md:107
固定版本的原文 · 对应讲解： 沿当前调用链等待与恢复
from slime.observability.trace_utils import build_sglang_meta_trace_attrs, trace_span
with trace_span(sample, "sglang_generate") as span:
output = await post(url, payload)
span.update(build_sglang_meta_trace_attrs(output["meta_info"]))
DOC-008 · docs/en/get_started/agent.md:41
固定版本的原文 · 对应讲解： 沿当前调用链等待与恢复
from slime.agent.adapters import AnthropicAdapter
adapter = AnthropicAdapter(
tokenizer=tokenizer,
sglang_url=sglang_url,
tool_parser=tool_parser,
reasoning_parser=reasoning_parser,
)
adapter.open_session(session_id, sampling_defaults=sampling_params)
# Agent client sends requests to adapter.app.
segments = await adapter.finish_session(session_id)
DOC-009 · docs/en/get_started/customization.md:77
固定版本的原文 · 对应讲解： 定义协程函数与创建协程对象 异步迭代与异步生成器 沿当前调用链等待与恢复
from slime.data.transport import publish_rollout_async
from slime.rollout.base_types import finalize_rollout_groups
async def generate_stored_batch(args, rollout_id, data_source, completed_groups):
refs = []
async for group in completed_groups:
ref = await publish_rollout_async(
group, args, rollout_id, group=True, controller=data_source.controller
)
refs.append(ref)
return finalize_rollout_groups(args, rollout_id, refs, controller=data_source.controller)
DOC-010 · docs/en/get_started/customization.md:115
固定版本的原文 · 对应讲解： 定义协程函数与创建协程对象
async def custom_generate(args, sample: Sample, sampling_params: dict) -> Sample | list[Sample]
DOC-011 · docs/en/get_started/customization.md:130
固定版本的原文 · 对应讲解： 定义协程函数与创建协程对象 沿当前调用链等待与恢复
import copy
from slime.utils.types import Sample
async def custom_generate(args, sample: Sample, sampling_params: dict) -> list[Sample]:
segments = await run_agent_and_split_segments(args, sample, sampling_params)
rollout_id = sample.rollout_id if sample.rollout_id is not None else sample.index
samples: list[Sample] = []
for segment in segments:
s = copy.copy(sample)
s.tokens = segment.tokens
s.response = segment.response
s.response_length = segment.response_length
s.loss_mask = segment.loss_mask
s.reward = segment.reward
s.status = Sample.Status.COMPLETED
s.rollout_id = rollout_id
samples.append(s)
return samples
DOC-012 · docs/en/get_started/customization.md:167
固定版本的原文 · 对应讲解： 定义协程函数与创建协程对象
async def custom_rm(args, sample: Sample) -> float
DOC-013 · docs/en/get_started/customization.md:172
固定版本的原文 · 对应讲解： 定义协程函数与创建协程对象
async def batched_custom_rm(args, samples: list[Sample]) -> list[float]
DOC-014 · docs/en/get_started/quick_start.md:500
固定版本的原文 · 对应讲解： 定义协程函数与创建协程对象 沿当前调用链等待与恢复
async def generate(args, sample: Sample, sampling_params) -> Sample:
# ... initialization ...
prompt, full_response, loss_masks = sample.prompt, "", []
for _ in range(max_turns):
# 1. Model generates action
model_output = await call_sglang(prompt + full_response, ...)
# ... tokenization and appending ...
loss_masks += [1] * len(model_tokens) # loss_mask = 1
full_response += model_output
# 2. Parse and execute action
action, content = parse_action(model_output)
if action == "search":
# 3 & 4. Get and append observation results
tool_output = await google_search(content)
# ... tokenization and appending ...
loss_masks += [0] * len(tool_tokens) # loss_mask = 0
full_response += tool_output
elif action == "answer":
break # end loop
# 7. Fill and return Sample object
sample.response = full_response
sample.tokens = ...
sample.loss_mask = loss_masks
return sample
DOC-015 · docs/en/get_started/usage.md:356
固定版本的原文 · 对应讲解： 定义协程函数与创建协程对象 沿当前调用链等待与恢复
async def generate(args, sample: Sample, sampling_params) -> Sample:
global TOKENIZER
if TOKENIZER is None:
TOKENIZER = AutoTokenizer.from_pretrained(args.hf_checkpoint, trust_remote_code=True)
# send request to router
output = await post(
f"http://{args.sglang_router_ip}:{args.sglang_router_port}/generate",
{
"text": sample.prompt,
"sampling_params": sampling_params,
}
)
prompt_tokens_ids = TOKENIZER(sample.prompt, add_special_tokens=False)["input_ids"]
response_token_ids = TOKENIZER(output["text"], add_special_tokens=False)["input_ids"]
# set sample
sample.tokens = prompt_tokens_ids + response_token_ids
sample.response_length = len(response_token_ids)
finish_reason = output["meta_info"]["finish_reason"]["type"]
if finish_reason == "length":
sample.status = Sample.Status.TRUNCATED
elif finish_reason == "abort":
sample.status = Sample.Status.ABORTED
else:
sample.status = Sample.Status.COMPLETED
sample.response = output["text"]
return sample
DOC-016 · docs/zh/advanced/sglang-config.md:173
固定版本的原文 · 对应讲解： 定义协程函数与创建协程对象 沿当前调用链等待与恢复
from slime.rollout.sglang_rollout import get_model_url
from slime.utils.http_utils import post
async def my_generate(args, sample, sampling_params):
# 路由到 actor 模型（默认）
actor_url = get_model_url(args, "actor", "/generate")
output = await post(actor_url, {"text": sample.prompt, "sampling_params": sampling_params})
# 路由到 reference 模型
ref_url = get_model_url(args, "ref", "/generate")
ref_output = await post(ref_url, {"text": sample.prompt, "sampling_params": sampling_params})
# 路由到 reward 模型（如 OpenAI 兼容 API）
reward_url = get_model_url(args, "reward", "/v1/chat/completions")
reward_output = await post(reward_url, {...})
...
DOC-017 · docs/zh/advanced/sglang-config.md:401
固定版本的原文 · 对应讲解： 定义协程函数与创建协程对象 沿当前调用链等待与恢复
from slime.rollout.sglang_rollout import get_model_url
from slime.utils.http_utils import post
async def generate_with_models(args, sample, sampling_params):
"""使用 actor 生成，用 reward 模型打分，与 reference 比较。"""
# 从 actor 生成
actor_url = get_model_url(args, "actor", "/generate")
actor_output = await post(actor_url, {
"text": sample.prompt,
"sampling_params": sampling_params,
"return_logprob": True,
})
# 获取 reference logprobs 用于 KL penalty
ref_url = get_model_url(args, "ref", "/generate")
ref_output = await post(ref_url, {
"text": sample.prompt + actor_output["text"],
"sampling_params": {"max_new_tokens": 0, "temperature": 0},
"return_logprob": True,
})
# 用 reward 模型打分
reward_url = get_model_url(args, "reward", "/v1/chat/completions")
reward_output = await post(reward_url, {
"model": "reward",
"messages": [{"role": "user", "content": sample.prompt + actor_output["text"]}],
})
# ... 处理输出并返回 Sample
DOC-018 · docs/zh/developer_guide/trace.md:60
固定版本的原文 · 对应讲解： 定义协程函数与创建协程对象
from slime.observability.trace_utils import trace_function
@trace_function("generate_and_rm", target="sample")
async def generate_and_rm(args, sample, sampling_params, evaluation=False):
...
@trace_function(
"generate_and_rm_group",
target="group",
attrs_getter=lambda args, group, sampling_params, evaluation=False: {"group_size": len(group)},
)
async def generate_and_rm_group(args, group, sampling_params, evaluation=False):
...
DOC-019 · docs/zh/developer_guide/trace.md:90
固定版本的原文 · 对应讲解： 定义协程函数与创建协程对象
@trace_function(
"custom_rollout_batch",
target="samples",
attrs_getter=lambda samples, **_: {"batch_size": len(samples)},
)
async def custom_rollout_batch(samples, **kwargs):
...
DOC-020 · docs/zh/developer_guide/trace.md:107
固定版本的原文 · 对应讲解： 沿当前调用链等待与恢复
from slime.observability.trace_utils import build_sglang_meta_trace_attrs, trace_span
with trace_span(sample, "sglang_generate") as span:
output = await post(url, payload)
span.update(build_sglang_meta_trace_attrs(output["meta_info"]))
DOC-021 · docs/zh/get_started/agent.md:41
固定版本的原文 · 对应讲解： 沿当前调用链等待与恢复
from slime.agent.adapters import AnthropicAdapter
adapter = AnthropicAdapter(
tokenizer=tokenizer,
sglang_url=sglang_url,
tool_parser=tool_parser,
reasoning_parser=reasoning_parser,
)
adapter.open_session(session_id, sampling_defaults=sampling_params)
# Agent client 向 adapter.app 发送请求。
segments = await adapter.finish_session(session_id)
DOC-022 · docs/zh/get_started/customization.md:72
固定版本的原文 · 对应讲解： 定义协程函数与创建协程对象 异步迭代与异步生成器 沿当前调用链等待与恢复
from slime.data.transport import publish_rollout_async
from slime.rollout.base_types import finalize_rollout_groups
async def generate_stored_batch(args, rollout_id, data_source, completed_groups):
refs = []
async for group in completed_groups:
ref = await publish_rollout_async(
group, args, rollout_id, group=True, controller=data_source.controller
)
refs.append(ref)
return finalize_rollout_groups(args, rollout_id, refs, controller=data_source.controller)
DOC-023 · docs/zh/get_started/customization.md:100
固定版本的原文 · 对应讲解： 定义协程函数与创建协程对象
async def custom_generate(args, sample: Sample, sampling_params: dict) -> Sample | list[Sample]
DOC-024 · docs/zh/get_started/customization.md:115
固定版本的原文 · 对应讲解： 定义协程函数与创建协程对象 沿当前调用链等待与恢复
import copy
from slime.utils.types import Sample
async def custom_generate(args, sample: Sample, sampling_params: dict) -> list[Sample]:
segments = await run_agent_and_split_segments(args, sample, sampling_params)
rollout_id = sample.rollout_id if sample.rollout_id is not None else sample.index
samples: list[Sample] = []
for segment in segments:
s = copy.copy(sample)
s.tokens = segment.tokens
s.response = segment.response
s.response_length = segment.response_length
s.loss_mask = segment.loss_mask
s.reward = segment.reward
s.status = Sample.Status.COMPLETED
s.rollout_id = rollout_id
samples.append(s)
return samples
DOC-025 · docs/zh/get_started/customization.md:152
固定版本的原文 · 对应讲解： 定义协程函数与创建协程对象
async def custom_rm(args, sample: Sample) -> float
DOC-026 · docs/zh/get_started/customization.md:157
固定版本的原文 · 对应讲解： 定义协程函数与创建协程对象
async def batched_custom_rm(args, samples: list[Sample]) -> list[float]
DOC-027 · docs/zh/get_started/quick_start.md:506
固定版本的原文 · 对应讲解： 定义协程函数与创建协程对象 沿当前调用链等待与恢复
async def generate(args, sample: Sample, sampling_params) -> Sample:
# ... 初始化 ...
prompt, full_response, loss_masks = sample.prompt, "", []
for _ in range(max_turns):
# 1. 模型生成动作
model_output = await call_sglang(prompt + full_response, ...)
# ... tokenization and appending ...
loss_masks += [1] * len(model_tokens) # loss_mask = 1
full_response += model_output
# 2. 解析并执行动作
action, content = parse_action(model_output)
if action == "search":
# 3 & 4. 获取并追加观察结果
tool_output = await google_search(content)
# ... tokenization and appending ...
loss_masks += [0] * len(tool_tokens) # loss_mask = 0
full_response += tool_output
elif action == "answer":
break # 结束循环
# 7. 填充并返回 Sample 对象
sample.response = full_response
sample.tokens = ...
sample.loss_mask = loss_masks
return sample
DOC-028 · docs/zh/get_started/usage.md:358
固定版本的原文 · 对应讲解： 定义协程函数与创建协程对象 沿当前调用链等待与恢复
async def generate(args, sample: Sample, sampling_params) -> Sample:
global TOKENIZER
if TOKENIZER is None:
TOKENIZER = AutoTokenizer.from_pretrained(args.hf_checkpoint, trust_remote_code=True)
# send request to router
output = await post(
f"http://{args.sglang_router_ip}:{args.sglang_router_port}/generate",
{
"text": sample.prompt,
"sampling_params": sampling_params,
}
)
prompt_tokens_ids = TOKENIZER(sample.prompt, add_special_tokens=False)["input_ids"]
response_token_ids = TOKENIZER(output["text"], add_special_tokens=False)["input_ids"]
# set sample
sample.tokens = prompt_tokens_ids + response_token_ids
sample.response_length = len(response_token_ids)
finish_reason = output["meta_info"]["finish_reason"]["type"]
if finish_reason == "length":
sample.status = Sample.Status.TRUNCATED
elif finish_reason == "abort":
sample.status = Sample.Status.ABORTED
else:
sample.status = Sample.Status.COMPLETED
sample.response = output["text"]
return sample
DOC-029 · examples/coding_agent_rl/README.md:194
固定版本的原文 · 对应讲解： 异步进入、退出与资源回收 沿当前调用链等待与恢复
await sb.exec(cmd, user=..., check=..., timeout=...)
await sb.write_file(sandbox_path, content_or_host_path, user=...)
await sb.read_file(sandbox_path, user=...)
async with E2BSandbox(...) as sb: ...
下载与重新核对
这份对照不声称覆盖其他 Slime 版本、动态生成的代码或 Python 标准库里本仓库未使用的所有接口。更新源码后需要重新扫描并补齐新增映射，不能沿用旧的“无缺项”结论。