EngineX/xc-llm-ascend
3
0
Fork 0
Code Issues Pull Requests Projects Releases Wiki Activity
2,880 Commits 3 Branches 0 Tags
9fd01a52c09fe653faa9498afe920c81775a0f9e
Commit Graph

3 Commits

Author SHA1 Message Date
hucong
8a671a109c [CI][Cherry-pick] Relax TTFT benefits threshold from 0.4 to 0.5 to account for DP load imbalance (#8684) 
Cherry-pick https://github.com/vllm-project/vllm-ascend/pull/8683

### What this PR does / why we need it?

This PR relaxes the TTFT threshold from `0.4` to `0.5` to improve
robustness under Data Parallel (DP) load imbalance.

#### Background

The current assertion enforces: prefix75 < prefix0 * 0.4

#### ❌ Nightly Failure Cases (Observed)

| prefix0 | threshold (0.4x) | prefix75 | delta |
|--------|------------------|----------|--------|
| 4696.24 | 1878.50 | 1883.99 | +5.49 |
| 4696.20 | 1878.48 | 1896.01 | +17.53 |
| 4636.73 | 1854.69 | 1902.48 | +47.79 |
| 4655.17 | 1862.07 | 1913.54 | +51.47 |
| 4685.35 | 1874.14 | 1919.36 | +45.22 |
| 4660.33 | 1864.13 | 1915.41 | +51.28 |
| 4648.30 | 1859.32 | 1950.50 | +91.18 |
| 4655.30 | 1862.12 | 1962.32 | +100.20 |

---

#### ✅ Nightly Passing Cases (Observed)

| prefix0 | threshold (0.4x) | prefix75 | margin |
|--------|------------------|----------|---------|
| 4685.64 | 1874.26 | 1864.46 | -9.80 |
| 5520.28 | 2208.11 | 1928.97 | -279.14 |
| 4639.23 | 1855.69 | 1846.86 | -8.83 |
| 4651.64 | 1860.66 | 1854.30 | -6.36 |
| 4640.39 | 1856.15 | 1840.32 | -15.83 |
| 4677.20 | 1870.88 | 1848.35 | -22.53 |
---

#### Key Observations

- Failures exceed the threshold by only **~5 ms to ~100 ms (~0.3%–5%)**
- Passing cases often have **very tight margins (~5–10 ms)**
- There is clear **overlap between pass and fail boundaries**
- Many failures are **borderline violations**, not real regressions

---

#### Root Cause

The instability is caused by **Data Parallel (DP) load imbalance**,
which introduces systematic variance:

- Uneven request distribution across workers  
- Queueing delays  
- Increased TTFT variance (especially for `prefix75`)

---

#### Conclusion

- The current threshold (`0.4x`) is **too strict**
- Observed natural fluctuation:
  - Absolute: up to ~100 ms  
  - Relative: up to ~5% over threshold  
- Pass/fail boundary is currently **too sensitive to runtime jitter**

---

#### Change

We relax the threshold: **0.4 → 0.5**


This adjustment:

- Accounts for expected runtime variance  
- Reduces false negatives  
- Maintains a meaningful performance constraint  

Even with `0.5`, the requirement remains strict (`prefix75 < 50% of
prefix0`) and does not mask real regressions.


---

### Does this PR introduce _any_ user-facing change?

No.

This change only affects internal test assertions and does not impact
user-facing behavior or model performance.


---

### How was this patch tested?

- Verified against existing TTFT test cases:
  - Previously failing cases (due to small variance) now pass
  - No regressions observed in other scenarios
- Confirmed that failures were due to DP load imbalance rather than
actual performance degradation
- Ensured the updated threshold still enforces a meaningful constraint
on TTFT

Signed-off-by: underfituu <hzhucong@163.com>
 2026-04-27 16:38:07 +08:00
hucong
4a628f1042 [UT][v0.18.0] Fix APC nightly UT and TTFT ratio (cherry-pick #7468) (#8053) 
<!--  Thanks for sending a pull request!

BEFORE SUBMITTING, PLEASE READ
https://docs.vllm.ai/en/latest/contributing/overview.html

-->
### What this PR does / why we need it?
<!--
- Please clarify what changes you are proposing. The purpose of this
section is to outline the changes and how this PR fixes the issue.
If possible, please consider writing useful notes for better and faster
reviews in your PR.

- Please clarify why the changes are needed. For instance, the use case
and bug description.

- Fixes #
-->
Cherry-pick from https://github.com/vllm-project/vllm-ascend/pull/7468

- Fix TTFT ratio threshold from 0.8 to 0.4 for prefix cache benchmarks
- Fix max_out_len values for warm_up and benchmark configs
- Applied to both DeepSeek-R1-0528-W8A8 and Qwen3-32B-Int8 configs

### Does this PR introduce _any_ user-facing change?
<!--
Note that it means *any* user-facing change including all aspects such
as API, interface or other behavior changes.
Documentation-only updates are not considered user-facing changes.
-->

### How was this patch tested?
<!--
CI passed with new added/existing test.
If it was tested in a way different from regular unit tests, please
clarify how you tested step by step, ideally copy and paste-able, so
that other reviewers can test and check, and descendants can verify in
the future.
If tests were not added, please describe why they were not added and/or
why it was difficult to add.
-->

Signed-off-by: underfituu <hzhucong@163.com>
 2026-04-08 21:08:26 +08:00
SILONG ZENG
859f2c25b9 [Nightly][Refactor]Migrate nightly single-node model tests from .py to .yaml (#6503) 
### What this PR does / why we need it?
This PR refactors the nightly single-node model test by migrating test
configurations from Python scripts to a more maintainable `YAML-based`
format.

| Original PR | Python (`.py`) | YAML (`.yaml`) |
| :--- | :--- | :--- |
| [#3568](https://github.com/vllm-project/vllm-ascend/pull/3568) |
`test_deepseek_r1_0528_w8a8_eplb.py` | `DeepSeek-R1-0528-W8A8.yaml` |
| [#3631](https://github.com/vllm-project/vllm-ascend/pull/3631) |
`test_deepseek_r1_0528_w8a8.py` | `DeepSeek-R1-0528-W8A8.yaml` |
| [#5874](https://github.com/vllm-project/vllm-ascend/pull/5874) |
`test_deepseek_r1_w8a8_hbm.py` | `DeepSeek-R1-W8A8-HBM.yaml` |
| [#3908](https://github.com/vllm-project/vllm-ascend/pull/3908) |
`test_deepseek_v3_2_w8a8.py` | `DeepSeek-V3.2-W8A8.yaml` |
| [#5682](https://github.com/vllm-project/vllm-ascend/pull/5682) |
`test_kimi_k2_thinking.py` | `Kimi-K2-Thinking.yaml` |
| [#4111](https://github.com/vllm-project/vllm-ascend/pull/4111) |
`test_mtpx_deepseek_r1_0528_w8a8.py` | `MTPX-DeepSeek-R1-0528-W8A8.yaml`
|
| [#3733](https://github.com/vllm-project/vllm-ascend/pull/3733) |
`test_prefix_cache_deepseek_r1_0528_w8a8.py` |
`Prefix-Cache-DeepSeek-R1-0528-W8A8.yaml` |
| [#6543](https://github.com/vllm-project/vllm-ascend/pull/6543) |
`test_qwen3_235b_w8a8.py` | `Qwen3-235B-A22B-W8A8.yaml` |
| [#6543](https://github.com/vllm-project/vllm-ascend/pull/6543) |
`test_qwen3_235b_a22b_w8a8_eplb.py` | `Qwen3-235B-A22B-W8A8.yaml` |
| [#3973](https://github.com/vllm-project/vllm-ascend/pull/3973) |
`test_qwen3_30b_w8a8.py` | `Qwen3-30B-A3B-W8A8.yaml` |
| [#3541](https://github.com/vllm-project/vllm-ascend/pull/3541) |
`test_qwen3_32b_int8.py` | `Qwen3-32B-Int8.yaml` |
| [#3757](https://github.com/vllm-project/vllm-ascend/pull/3757) |
`test_qwq_32b.py` | `QwQ-32B.yaml` |
| [#5616](https://github.com/vllm-project/vllm-ascend/pull/5616) |
`test_qwen3_next_w8a8.py` | `Qwen3-Next-80B-A3B-Instruct-W8A8.yaml` |
| [#3541](https://github.com/vllm-project/vllm-ascend/pull/3541) |
`test_qwen2_5_vl_7b.py` | `Qwen2.5-VL-7B-Instruct.yaml` |
| [#5301](https://github.com/vllm-project/vllm-ascend/pull/5301) |
`test_qwen2_5_vl_7b_epd.py` | `Qwen2.5-VL-7B-Instruct-EPD.yaml` |
| [#3707](https://github.com/vllm-project/vllm-ascend/pull/3707) |
`test_qwen2_5_vl_32b.py` | `Qwen2.5-VL-32B-Instruct.yaml` |
| [#3676](https://github.com/vllm-project/vllm-ascend/pull/3676) |
`test_qwen3_32b_int8_a3_feature_stack3.py` |
`Qwen3-32B-Int8-A3-Feature-Stack3.yaml` |
| [#3709](https://github.com/vllm-project/vllm-ascend/pull/3709) |
`test_prefix_cache_qwen3_32b_int8.py` |
`Prefix-Cache-Qwen3-32B-Int8.yaml` |
| [#5395](https://github.com/vllm-project/vllm-ascend/pull/5395) |
`test_qwen3_next.py` | `Qwen3-Next-80B-A3B-Instruct-A2.yaml` |
| [#3474](https://github.com/vllm-project/vllm-ascend/pull/3474) |
`test_qwen3_32b.py` | `Qwen3-32B.yaml` |
| [#3541](https://github.com/vllm-project/vllm-ascend/pull/3541) |
`test_qwen3_32b_int8.py` | `Qwen3-32B-Int8-A2.yaml` |
### Does this PR introduce _any_ user-facing change?

### How was this patch tested?

- vLLM version: v0.15.0
- vLLM main: https://github.com/vllm-project/vllm/commit/v0.15.0

---------

Signed-off-by: MrZ20 <2609716663@qq.com>
 2026-03-03 20:13:43 +08:00