[FDConfig] Reduce FD_CUSTOM_AR_MAX_SIZE_MB default from 64 to 8

[gongweibao]

Motivation

The default FD_CUSTOM_AR_MAX_SIZE_MB of 64MB is unnecessarily large for most single-GPU and small-model deployments. Reducing it to 8MB lowers shared memory allocation overhead. Multi-GPU or large-model scenarios that need bigger buffers can set the env var explicitly.

Modifications

• fastdeploy/envs.py: Change default value from 64 to 8, update comment accordingly.
• tests/e2e/4cards_cases/test_determinism_long.py: Explicitly set FD_CUSTOM_AR_MAX_SIZE_MB=64 (was using 57 as fallback; now aligned with other test files).

Usage or Command

# Use default 8MB buffer
python -m fastdeploy.entrypoints.openai.api_server ...

# Override for multi-GPU with large tensors
export FD_CUSTOM_AR_MAX_SIZE_MB=64

Checklist

• Add at least a tag in the PR title.
• Format your code, run pre-commit before commit.
• Add unit tests. No new unit tests needed — this is a config default change; existing e2e tests cover the behavior.
• Provide accuracy results.

🤖 Generated with Claude Code

[paddle-bot]

Thanks for your contribution!

[CLAassistant]

Thank you for your submission! We really appreciate it. Like many open source projects, we ask that you sign our Contributor License Agreement before we can accept your contribution.

---

gongweibao seems not to be a GitHub user. You need a GitHub account to be able to sign the CLA. If you have already a GitHub account, please add the email address used for this commit to your account.
_{You have signed the CLA already but the status is still pending? Let us recheck it.}

[Copilot]

Pull request overview

该 PR 调整自定义 all-reduce 共享 buffer 的默认配置：将 FD_CUSTOM_AR_MAX_SIZE_MB 默认值从 64MB 下调到 8MB，以降低多数小规模部署的共享内存占用，并在确定性相关 e2e 用例中显式使用更大的 buffer 以保持稳定性。

Changes:

• 将 FD_CUSTOM_AR_MAX_SIZE_MB 默认值从 64 改为 8，并更新对应注释
• 将 tests/e2e/4cards_cases/test_determinism_long.py 中该环境变量的默认回退值改为 64

Reviewed changes

Copilot reviewed 2 out of 2 changed files in this pull request and generated 3 comments.

File	Description
fastdeploy/envs.py	下调 custom all-reduce buffer 默认值并更新注释
tests/e2e/4cards_cases/test_determinism_long.py	调整 determinism long e2e 用例中 buffer 的默认回退值

[Copilot]

这里的注释与实际行为不一致：在 FD_DETERMINISTIC_MODE=1 时，如果输入 tensor 超过 max_size，会直接抛 RuntimeError（communication._ensure_deterministic_ready），不会“fallback 到 NCCL”。建议把注释改成“超大 tensor 会报错/需要调大该值以满足 deterministic all-reduce 的 max_size 限制”，避免误导。

Suggested change

	# Custom all-reduce max buffer size in MB (default 8MB).
	# Increase this to avoid NCCL fallback for large tensors in deterministic mode.
	# Custom deterministic all-reduce max buffer size in MB (default 8MB).
	# When FD_DETERMINISTIC_MODE=1, tensors larger than this limit will raise an error
	# instead of falling back to NCCL. Increase this value to avoid max_size errors.

[Copilot]

FD_CUSTOM_AR_MAX_SIZE_MB 这里直接 int(os.getenv(...))，未校验取值范围；如果被设置为 0/负数，会导致 CustomAllreduce 分配共享 buffer 时 size_in_bytes 非法并在更底层报错，排查困难。建议新增一个校验（类似 _validate_split_kv_size），确保该值为正整数（例如 >=1），并在不合法时给出明确异常信息。

[Copilot]

该测试这里允许外部环境变量覆盖 FD_CUSTOM_AR_MAX_SIZE_MB；如果 CI/本地恰好设置为更小值（比如现在默认 8），可能导致 deterministic all-reduce 在大 tensor 场景直接报错而引入不稳定。考虑与同目录其他 determinism e2e 测试保持一致，直接固定为 "64"（如 _test_determinism_offline.py），让用例结果不受外部环境影响。

Suggested change

	"FD_CUSTOM_AR_MAX_SIZE_MB": os.environ.get("FD_CUSTOM_AR_MAX_SIZE_MB", "64"),
	"FD_CUSTOM_AR_MAX_SIZE_MB": "64",

[codecov-commenter]

Codecov Report

✅ All modified and coverable lines are covered by tests.
⚠️ Please upload report for BASE (develop@6f5aa88). Learn more about missing BASE report.

Additional details and impacted files

@@            Coverage Diff             @@
##             develop    #6997   +/-   ##
==========================================
  Coverage           ?   73.85%           
==========================================
  Files              ?      399           
  Lines              ?    56045           
  Branches           ?     8849           
==========================================
  Hits               ?    41392           
  Misses             ?    11727           
  Partials           ?     2926           

Flag	Coverage Δ
GPU	73.85% <ø> (?)

Flags with carried forward coverage won't be shown. Click here to find out more.

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.