vllm - ✅(Solved) Fix [Bug]: Gemma-3 specific heterogeneous TP failures with PD disagg [1 pull requests, 1 comments, 2 participants]

vllm2026-03-17 18:09:32

ON THIS PAGE

Recommended Tools

×6

Utilities matched from this issue’s tags and category — try them while you read without losing context.

GitHub issue graph ai analysis

Paste a GitHub issue URL. We fetch that issue, discover linked issues from bodies/comments/timeline, collect linked pull requests, and produce a structured English report.

The report is written in English Markdown for sharing and archival.

GitHub issue URL

Helpful · Quick feedback

GitHub stats

vllm-project/vllm#37333•Fetched 2026-04-08 00:53:29

View on GitHub

Comments

Participants

Timeline

Reactions

Author

Participants

Assignees

Timeline (top)

cross-referenced ×3assigned ×1closed ×1commented ×1

Error Message

======================================================================== SUMMARY

[PASS] tp_gemma_sw_fa.yaml 167.7s logs: /home/yzong-rh/vllm/tests/v1/kv_connector/nixl_integration/logs/20260317_163330/tp_gemma_sw_fa.{stdout,stderr} [PASS] tp_gemma_sw_fa_hma.yaml 171.1s logs: /home/yzong-rh/vllm/tests/v1/kv_connector/nixl_integration/logs/20260317_163330/tp_gemma_sw_fa_hma.{stdout,stderr} [PASS] tp_gemma_sw_fa_hma_2p2d.yaml 200.4s logs: /home/yzong-rh/vllm/tests/v1/kv_connector/nixl_integration/logs/20260317_163330/tp_gemma_sw_fa_hma_2p2d.{stdout,stderr} [FAIL] xfail_tp_gemma_sw_fa_1p2d.yaml 239.5s logs: /home/yzong-rh/vllm/tests/v1/kv_connector/nixl_integration/logs/20260317_163330/xfail_tp_gemma_sw_fa_1p2d.{stdout,stderr} error: Command '['bash', '/home/yzong-rh/vllm/tests/v1/kv_connector/nixl_integration/run_accuracy_test.sh']' returned non-zero exit status 1. [FAIL] xfail_tp_gemma_sw_fa_2p1d.yaml 237.2s logs: /home/yzong-rh/vllm/tests/v1/kv_connector/nixl_integration/logs/20260317_163330/xfail_tp_gemma_sw_fa_2p1d.{stdout,stderr} error: Command '['bash', '/home/yzong-rh/vllm/tests/v1/kv_connector/nixl_integration/run_accuracy_test.sh']' returned non-zero exit status 1.

Passed: 3 | Failed: 2 | Total: 5

Fix Action

Fix / Workaround

============================== CPU Info

Architecture: x86_64 CPU op-mode(s): 32-bit, 64-bit Address sizes: 46 bits physical, 57 bits virtual Byte Order: Little Endian CPU(s): 160 On-line CPU(s) list: 0-159 Vendor ID: GenuineIntel Model name: Intel Xeon Processor (SapphireRapids) CPU family: 6 Model: 143 Thread(s) per core: 2 Core(s) per socket: 40 Socket(s): 2 Stepping: 4 BogoMIPS: 4200.00 Flags: fpu vme de pse tsc msr pae mce cx8 apic sep mtrr pge mca cmov pat pse36 clflush mmx fxsr sse sse2 ss ht syscall nx pdpe1gb rdtscp lm constant_tsc rep_good nopl xtopology cpuid tsc_known_freq pni pclmulqdq vmx ssse3 fma cx16 pcid sse4_1 sse4_2 x2apic movbe popcnt tsc_deadline_timer aes xsave avx f16c rdrand hypervisor lahf_lm abm 3dnowprefetch cpuid_fault ssbd ibrs ibpb stibp ibrs_enhanced tpr_shadow flexpriority ept vpid ept_ad fsgsbase tsc_adjust bmi1 avx2 smep bmi2 erms invpcid avx512f avx512dq rdseed adx smap avx512ifma clflushopt clwb avx512cd sha_ni avx512bw avx512vl xsaveopt xsavec xgetbv1 xsaves avx_vnni avx512_bf16 wbnoinvd arat vnmi avx512vbmi umip pku ospke waitpkg avx512_vbmi2 gfni vaes vpclmulqdq avx512_vnni avx512_bitalg avx512_vpopcntdq la57 rdpid bus_lock_detect cldemote movdiri movdir64b fsrm md_clear serialize tsxldtrk amx_bf16 avx512_fp16 amx_tile amx_int8 arch_capabilities Virtualization: VT-x Hypervisor vendor: KVM Virtualization type: full L1d cache: 5 MiB (160 instances) L1i cache: 5 MiB (160 instances) L2 cache: 320 MiB (80 instances) L3 cache: 32 MiB (2 instances) NUMA node(s): 2 NUMA node0 CPU(s): 0-79 NUMA node1 CPU(s): 80-159 Vulnerability Gather data sampling: Not affected Vulnerability Indirect target selection: Mitigation; Aligned branch/return thunks Vulnerability Itlb multihit: Not affected Vulnerability L1tf: Not affected Vulnerability Mds: Not affected Vulnerability Meltdown: Not affected Vulnerability Mmio stale data: Unknown: No mitigations Vulnerability Reg file data sampling: Not affected Vulnerability Retbleed: Not affected Vulnerability Spec rstack overflow: Not affected Vulnerability Spec store bypass: Mitigation; Speculative Store Bypass disabled via prctl Vulnerability Spectre v1: Mitigation; usercopy/swapgs barriers and __user pointer sanitization Vulnerability Spectre v2: Mitigation; Enhanced / Automatic IBRS; IBPB conditional; PBRSB-eIBRS SW sequence; BHI SW loop, KVM SW loop Vulnerability Srbds: Not affected Vulnerability Tsa: Not affected Vulnerability Tsx async abort: Not affected Vulnerability Vmscape: Not affected

PR fix notes

PR #37940: [NIXL][BUG] Fix Triton heterogeneous TP

Repository: vllm-project/vllm
Author: yzong-rh
State: closed | merged: True
Link: https://github.com/vllm-project/vllm/pull/37940

Description (problem / solution / changelog)

co-authored with @ZhanqiuHu

Purpose

Fix Triton Attn Heterogeneous TP Disagg: #37703
Also fixes Gemma with Heterogeneous TP bug, also caused by Triton Backend: #37333
Enable cross-layer TP disagg for Triton, which now has the same KV cache layout as FlashInfer

Test Plan

In tests/v1/kv_connector/nixl_integration/config_sweep_accuracy_test.sh, replace tp_configs with the following:

Fixed GEMMA tests:

"GPU_MEMORY_UTILIZATION=0.8 MODEL_NAMES=google/gemma-3-4b-it PREFILLER_TP_SIZE=1 DECODER_TP_SIZE=2 VLLM_SERVE_EXTRA_ARGS=--max-model-len,8192"
"GPU_MEMORY_UTILIZATION=0.8 MODEL_NAMES=google/gemma-3-4b-it PREFILLER_TP_SIZE=2 DECODER_TP_SIZE=1 VLLM_SERVE_EXTRA_ARGS=--max-model-len,8192"

Fixed Triton backend test

"GPU_MEMORY_UTILIZATION=0.6 PREFILLER_TP_SIZE=1 DECODER_TP_SIZE=2 VLLM_SERVE_EXTRA_ARGS=--attention-backend,TRITON_ATTN"
"GPU_MEMORY_UTILIZATION=0.6 PREFILLER_TP_SIZE=2 DECODER_TP_SIZE=1 VLLM_SERVE_EXTRA_ARGS=--attention-backend,TRITON_ATTN"

Run cd tests && v1/kv_connector/nixl_integration/config_sweep_accuracy_test.sh.

Test Result

All tests now pass. Triton backend was tested with CROSS_LAYERS_BLOCKS=0 and CROSS_LAYERS_BLOCKS=1.

cc @NickLucche

<details> <summary> Essential Elements of an Effective PR Description Checklist </summary>

The purpose of the PR, such as "Fix some issue (link existing issues this PR will resolve)".
The test plan, such as providing test command.
The test results, such as pasting the results comparison before and after, or e2e results
(Optional) The necessary documentation update, such as updating supported_models.md and examples for a new model.
(Optional) Release notes update. If your change is user facing, please update the release notes draft in the Google Doc.

</details>

Changed files

tests/v1/kv_connector/nixl_integration/config_sweep_accuracy_test.sh (modified, +7/-0)
tests/v1/kv_connector/unit/test_nixl_connector.py (modified, +17/-15)
vllm/distributed/kv_transfer/kv_connector/v1/nixl_connector.py (modified, +16/-0)
vllm/v1/attention/backends/triton_attn.py (modified, +13/-4)
vllm/v1/attention/ops/triton_reshape_and_cache_flash.py (modified, +9/-3)

Code Example

==============================
        System Info
==============================
OS                           : CentOS Stream 9 (x86_64)
GCC version                  : (GCC) 11.5.0 20240719 (Red Hat 11.5.0-11)
Clang version                : Could not collect
CMake version                : version 4.2.3
Libc version                 : glibc-2.34

==============================
       PyTorch Info
==============================
PyTorch version              : 2.10.0+cu128
Is debug build               : False
CUDA used to build PyTorch   : 12.8
ROCM used to build PyTorch   : N/A

==============================
      Python Environment
==============================
Python version               : 3.12.12 (main, Oct 10 2025, 12:47:49) [Clang 20.1.4 ] (64-bit runtime)
Python platform              : Linux-5.14.0-648.el9.x86_64-x86_64-with-glibc2.34

==============================
       CUDA / GPU Info
==============================
Is CUDA available            : True
CUDA runtime version         : 13.0.88
CUDA_MODULE_LOADING set to   : 
GPU models and configuration : 
GPU 0: NVIDIA H200
GPU 1: NVIDIA H200
GPU 2: NVIDIA H200
GPU 3: NVIDIA H200
GPU 4: NVIDIA H200
GPU 5: NVIDIA H200
GPU 6: NVIDIA H200
GPU 7: NVIDIA H200

Nvidia driver version        : 580.105.08
cuDNN version                : Could not collect
HIP runtime version          : N/A
MIOpen runtime version       : N/A
Is XNNPACK available         : True

==============================
          CPU Info
==============================
Architecture:                            x86_64
CPU op-mode(s):                          32-bit, 64-bit
Address sizes:                           46 bits physical, 57 bits virtual
Byte Order:                              Little Endian
CPU(s):                                  160
On-line CPU(s) list:                     0-159
Vendor ID:                               GenuineIntel
Model name:                              Intel Xeon Processor (SapphireRapids)
CPU family:                              6
Model:                                   143
Thread(s) per core:                      2
Core(s) per socket:                      40
Socket(s):                               2
Stepping:                                4
BogoMIPS:                                4200.00
Flags:                                   fpu vme de pse tsc msr pae mce cx8 apic sep mtrr pge mca cmov pat pse36 clflush mmx fxsr sse sse2 ss ht syscall nx pdpe1gb rdtscp lm constant_tsc rep_good nopl xtopology cpuid tsc_known_freq pni pclmulqdq vmx ssse3 fma cx16 pcid sse4_1 sse4_2 x2apic movbe popcnt tsc_deadline_timer aes xsave avx f16c rdrand hypervisor lahf_lm abm 3dnowprefetch cpuid_fault ssbd ibrs ibpb stibp ibrs_enhanced tpr_shadow flexpriority ept vpid ept_ad fsgsbase tsc_adjust bmi1 avx2 smep bmi2 erms invpcid avx512f avx512dq rdseed adx smap avx512ifma clflushopt clwb avx512cd sha_ni avx512bw avx512vl xsaveopt xsavec xgetbv1 xsaves avx_vnni avx512_bf16 wbnoinvd arat vnmi avx512vbmi umip pku ospke waitpkg avx512_vbmi2 gfni vaes vpclmulqdq avx512_vnni avx512_bitalg avx512_vpopcntdq la57 rdpid bus_lock_detect cldemote movdiri movdir64b fsrm md_clear serialize tsxldtrk amx_bf16 avx512_fp16 amx_tile amx_int8 arch_capabilities
Virtualization:                          VT-x
Hypervisor vendor:                       KVM
Virtualization type:                     full
L1d cache:                               5 MiB (160 instances)
L1i cache:                               5 MiB (160 instances)
L2 cache:                                320 MiB (80 instances)
L3 cache:                                32 MiB (2 instances)
NUMA node(s):                            2
NUMA node0 CPU(s):                       0-79
NUMA node1 CPU(s):                       80-159
Vulnerability Gather data sampling:      Not affected
Vulnerability Indirect target selection: Mitigation; Aligned branch/return thunks
Vulnerability Itlb multihit:             Not affected
Vulnerability L1tf:                      Not affected
Vulnerability Mds:                       Not affected
Vulnerability Meltdown:                  Not affected
Vulnerability Mmio stale data:           Unknown: No mitigations
Vulnerability Reg file data sampling:    Not affected
Vulnerability Retbleed:                  Not affected
Vulnerability Spec rstack overflow:      Not affected
Vulnerability Spec store bypass:         Mitigation; Speculative Store Bypass disabled via prctl
Vulnerability Spectre v1:                Mitigation; usercopy/swapgs barriers and __user pointer sanitization
Vulnerability Spectre v2:                Mitigation; Enhanced / Automatic IBRS; IBPB conditional; PBRSB-eIBRS SW sequence; BHI SW loop, KVM SW loop
Vulnerability Srbds:                     Not affected
Vulnerability Tsa:                       Not affected
Vulnerability Tsx async abort:           Not affected
Vulnerability Vmscape:                   Not affected

==============================
Versions of relevant libraries
==============================
[pip3] flashinfer-python==0.6.6
[pip3] mypy_extensions==1.1.0
[pip3] numpy==2.2.6
[pip3] nvidia-cublas-cu12==12.8.4.1
[pip3] nvidia-cuda-cupti-cu12==12.8.90
[pip3] nvidia-cuda-nvrtc-cu12==12.8.93
[pip3] nvidia-cuda-runtime-cu12==12.8.90
[pip3] nvidia-cudnn-cu12==9.10.2.21
[pip3] nvidia-cudnn-frontend==1.18.0
[pip3] nvidia-cufft-cu12==11.3.3.83
[pip3] nvidia-cufile-cu12==1.13.1.3
[pip3] nvidia-curand-cu12==10.3.9.90
[pip3] nvidia-cusolver-cu12==11.7.3.90
[pip3] nvidia-cusparse-cu12==12.5.8.93
[pip3] nvidia-cusparselt-cu12==0.7.1
[pip3] nvidia-cutlass-dsl==4.4.1
[pip3] nvidia-cutlass-dsl-libs-base==4.4.1
[pip3] nvidia-ml-py==13.590.48
[pip3] nvidia-nccl-cu12==2.27.5
[pip3] nvidia-nvjitlink-cu12==12.8.93
[pip3] nvidia-nvshmem-cu12==3.4.5
[pip3] nvidia-nvtx-cu12==12.8.90
[pip3] open_clip_torch==2.32.0
[pip3] pytorch-lightning==2.6.1
[pip3] pyzmq==27.1.0
[pip3] segmentation_models_pytorch==0.5.0
[pip3] sentence-transformers==5.3.0
[pip3] terratorch==1.2.4
[pip3] torch==2.10.0
[pip3] torch_c_dlpack_ext==0.1.5
[pip3] torchaudio==2.10.0
[pip3] torchgeo==0.9.0
[pip3] torchmetrics==1.9.0
[pip3] torchvision==0.25.0
[pip3] transformers==4.57.5
[pip3] transformers-stream-generator==0.0.5
[pip3] triton==3.6.0
[pip3] tritonclient==2.66.0
[pip3] vector-quantize-pytorch==1.27.21
[conda] Could not collect

==============================
         vLLM Info
==============================
ROCM Version                 : Could not collect
vLLM Version                 : 0.17.1rc1.dev237+g0a0a1a198 (git sha: 0a0a1a198)
vLLM Build Flags:
  CUDA Archs: Not Set; ROCm: Disabled
GPU Topology:
        GPU0    GPU1    GPU2    GPU3    GPU4    GPU5    GPU6    GPU7    CPU Affinity    NUMA Affinity   GPU NUMA ID
GPU0     X      NV18    NV18    NV18    NV18    NV18    NV18    NV18    0-79    0               N/A
GPU1    NV18     X      NV18    NV18    NV18    NV18    NV18    NV18    0-79    0               N/A
GPU2    NV18    NV18     X      NV18    NV18    NV18    NV18    NV18    0-79    0               N/A
GPU3    NV18    NV18    NV18     X      NV18    NV18    NV18    NV18    0-79    0               N/A
GPU4    NV18    NV18    NV18    NV18     X      NV18    NV18    NV18    80-159  1               N/A
GPU5    NV18    NV18    NV18    NV18    NV18     X      NV18    NV18    80-159  1               N/A
GPU6    NV18    NV18    NV18    NV18    NV18    NV18     X      NV18    80-159  1               N/A
GPU7    NV18    NV18    NV18    NV18    NV18    NV18    NV18     X      80-159  1               N/A

Legend:

  X    = Self
  SYS  = Connection traversing PCIe as well as the SMP interconnect between NUMA nodes (e.g., QPI/UPI)
  NODE = Connection traversing PCIe as well as the interconnect between PCIe Host Bridges within a NUMA node
  PHB  = Connection traversing PCIe as well as a PCIe Host Bridge (typically the CPU)
  PXB  = Connection traversing multiple PCIe bridges (without traversing the PCIe Host Bridge)
  PIX  = Connection traversing at most a single PCIe bridge
  NV#  = Connection traversing a bonded set of # NVLinks

==============================
     Environment Variables
==============================
VLLM_CACHE_ROOT==<...>
LD_LIBRARY_PATH=:/usr/local/cuda-13.0/lib64:/usr/local/cuda-13.0/lib64:/usr/local/cuda-13.0/lib64:/usr/local/cuda-13.0/lib64
CUDA_HOME=/usr/local/cuda-13.0
CUDA_HOME=/usr/local/cuda-13.0
PYTORCH_NVML_BASED_CUDA_CHECK=1
TORCHINDUCTOR_COMPILE_THREADS=1
TORCHINDUCTOR_CACHE_DIR=/tmp/torchinductor_yzong-rh
UCX_TLS=cuda_ipc,cuda_copy,tcp

---

========================================================================
SUMMARY
========================================================================
  [PASS]  tp_gemma_sw_fa.yaml                        167.7s  logs: /home/yzong-rh/vllm/tests/v1/kv_connector/nixl_integration/logs/20260317_163330/tp_gemma_sw_fa.{stdout,stderr}
  [PASS]  tp_gemma_sw_fa_hma.yaml                    171.1s  logs: /home/yzong-rh/vllm/tests/v1/kv_connector/nixl_integration/logs/20260317_163330/tp_gemma_sw_fa_hma.{stdout,stderr}
  [PASS]  tp_gemma_sw_fa_hma_2p2d.yaml               200.4s  logs: /home/yzong-rh/vllm/tests/v1/kv_connector/nixl_integration/logs/20260317_163330/tp_gemma_sw_fa_hma_2p2d.{stdout,stderr}
  [FAIL]  xfail_tp_gemma_sw_fa_1p2d.yaml             239.5s  logs: /home/yzong-rh/vllm/tests/v1/kv_connector/nixl_integration/logs/20260317_163330/xfail_tp_gemma_sw_fa_1p2d.{stdout,stderr}
         error: Command '['bash', '/home/yzong-rh/vllm/tests/v1/kv_connector/nixl_integration/run_accuracy_test.sh']' returned non-zero exit status 1.
  [FAIL]  xfail_tp_gemma_sw_fa_2p1d.yaml             237.2s  logs: /home/yzong-rh/vllm/tests/v1/kv_connector/nixl_integration/logs/20260317_163330/xfail_tp_gemma_sw_fa_2p1d.{stdout,stderr}
         error: Command '['bash', '/home/yzong-rh/vllm/tests/v1/kv_connector/nixl_integration/run_accuracy_test.sh']' returned non-zero exit status 1.
------------------------------------------------------------------------
  Passed: 3  |  Failed: 2  |  Total: 5
========================================================================

---

E       AssertionError: Expected: 0.76 | Measured: 0.000758150113722517
E       assert ((np.float64(0.000758150113722517) - 0.03) < 0.76 and (np.float64(0.000758150113722517) + 0.03) > 0.76)

---

E       AssertionError: Expected: 0.76 | Measured: 0.000758150113722517
E       assert ((np.float64(0.000758150113722517) - 0.03) < 0.76 and (np.float64(0.000758150113722517) + 0.03) > 0.76)

---

tp_configs = (
    "GPU_MEMORY_UTILIZATION=0.8 MODEL_NAMES=google/gemma-3-4b-it PREFILLER_TP_SIZE=1 DECODER_TP_SIZE=2 VLLM_SERVE_EXTRA_ARGS=--max-model-len,8192"
)

---

E       AssertionError: Expected: 0.76 | Measured: 0.000758150113722517
E       assert ((np.float64(0.000758150113722517) - 0.03) < 0.76 and (np.float64(0.000758150113722517) + 0.03) > 0.76)

v1/kv_connector/nixl_integration/test_accuracy.py:73: AssertionError
============================================================== warnings summary ===============================================================
<frozen importlib._bootstrap>:488
  <frozen importlib._bootstrap>:488: DeprecationWarning: builtin type SwigPyPacked has no __module__ attribute

<frozen importlib._bootstrap>:488
  <frozen importlib._bootstrap>:488: DeprecationWarning: builtin type SwigPyObject has no __module__ attribute

-- Docs: https://docs.pytest.org/en/stable/how-to/capture-warnings.html
=========================================================== short test summary info ===========================================================
FAILED v1/kv_connector/nixl_integration/test_accuracy.py::test_accuracy - AssertionError: Expected: 0.76 | Measured: 0.000758150113722517
!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!! stopping after 1 failures !!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!
================================================== 1 failed, 2 warnings in 149.34s (0:02:29) ==================================================

RAW_BUFFERClick to expand / collapse

Your current environment

<details> <summary>The output of <code>python collect_env.py</code></summary>

==============================
        System Info
==============================
OS                           : CentOS Stream 9 (x86_64)
GCC version                  : (GCC) 11.5.0 20240719 (Red Hat 11.5.0-11)
Clang version                : Could not collect
CMake version                : version 4.2.3
Libc version                 : glibc-2.34

==============================
       PyTorch Info
==============================
PyTorch version              : 2.10.0+cu128
Is debug build               : False
CUDA used to build PyTorch   : 12.8
ROCM used to build PyTorch   : N/A

==============================
      Python Environment
==============================
Python version               : 3.12.12 (main, Oct 10 2025, 12:47:49) [Clang 20.1.4 ] (64-bit runtime)
Python platform              : Linux-5.14.0-648.el9.x86_64-x86_64-with-glibc2.34

==============================
       CUDA / GPU Info
==============================
Is CUDA available            : True
CUDA runtime version         : 13.0.88
CUDA_MODULE_LOADING set to   : 
GPU models and configuration : 
GPU 0: NVIDIA H200
GPU 1: NVIDIA H200
GPU 2: NVIDIA H200
GPU 3: NVIDIA H200
GPU 4: NVIDIA H200
GPU 5: NVIDIA H200
GPU 6: NVIDIA H200
GPU 7: NVIDIA H200

Nvidia driver version        : 580.105.08
cuDNN version                : Could not collect
HIP runtime version          : N/A
MIOpen runtime version       : N/A
Is XNNPACK available         : True

==============================
          CPU Info
==============================
Architecture:                            x86_64
CPU op-mode(s):                          32-bit, 64-bit
Address sizes:                           46 bits physical, 57 bits virtual
Byte Order:                              Little Endian
CPU(s):                                  160
On-line CPU(s) list:                     0-159
Vendor ID:                               GenuineIntel
Model name:                              Intel Xeon Processor (SapphireRapids)
CPU family:                              6
Model:                                   143
Thread(s) per core:                      2
Core(s) per socket:                      40
Socket(s):                               2
Stepping:                                4
BogoMIPS:                                4200.00
Flags:                                   fpu vme de pse tsc msr pae mce cx8 apic sep mtrr pge mca cmov pat pse36 clflush mmx fxsr sse sse2 ss ht syscall nx pdpe1gb rdtscp lm constant_tsc rep_good nopl xtopology cpuid tsc_known_freq pni pclmulqdq vmx ssse3 fma cx16 pcid sse4_1 sse4_2 x2apic movbe popcnt tsc_deadline_timer aes xsave avx f16c rdrand hypervisor lahf_lm abm 3dnowprefetch cpuid_fault ssbd ibrs ibpb stibp ibrs_enhanced tpr_shadow flexpriority ept vpid ept_ad fsgsbase tsc_adjust bmi1 avx2 smep bmi2 erms invpcid avx512f avx512dq rdseed adx smap avx512ifma clflushopt clwb avx512cd sha_ni avx512bw avx512vl xsaveopt xsavec xgetbv1 xsaves avx_vnni avx512_bf16 wbnoinvd arat vnmi avx512vbmi umip pku ospke waitpkg avx512_vbmi2 gfni vaes vpclmulqdq avx512_vnni avx512_bitalg avx512_vpopcntdq la57 rdpid bus_lock_detect cldemote movdiri movdir64b fsrm md_clear serialize tsxldtrk amx_bf16 avx512_fp16 amx_tile amx_int8 arch_capabilities
Virtualization:                          VT-x
Hypervisor vendor:                       KVM
Virtualization type:                     full
L1d cache:                               5 MiB (160 instances)
L1i cache:                               5 MiB (160 instances)
L2 cache:                                320 MiB (80 instances)
L3 cache:                                32 MiB (2 instances)
NUMA node(s):                            2
NUMA node0 CPU(s):                       0-79
NUMA node1 CPU(s):                       80-159
Vulnerability Gather data sampling:      Not affected
Vulnerability Indirect target selection: Mitigation; Aligned branch/return thunks
Vulnerability Itlb multihit:             Not affected
Vulnerability L1tf:                      Not affected
Vulnerability Mds:                       Not affected
Vulnerability Meltdown:                  Not affected
Vulnerability Mmio stale data:           Unknown: No mitigations
Vulnerability Reg file data sampling:    Not affected
Vulnerability Retbleed:                  Not affected
Vulnerability Spec rstack overflow:      Not affected
Vulnerability Spec store bypass:         Mitigation; Speculative Store Bypass disabled via prctl
Vulnerability Spectre v1:                Mitigation; usercopy/swapgs barriers and __user pointer sanitization
Vulnerability Spectre v2:                Mitigation; Enhanced / Automatic IBRS; IBPB conditional; PBRSB-eIBRS SW sequence; BHI SW loop, KVM SW loop
Vulnerability Srbds:                     Not affected
Vulnerability Tsa:                       Not affected
Vulnerability Tsx async abort:           Not affected
Vulnerability Vmscape:                   Not affected

==============================
Versions of relevant libraries
==============================
[pip3] flashinfer-python==0.6.6
[pip3] mypy_extensions==1.1.0
[pip3] numpy==2.2.6
[pip3] nvidia-cublas-cu12==12.8.4.1
[pip3] nvidia-cuda-cupti-cu12==12.8.90
[pip3] nvidia-cuda-nvrtc-cu12==12.8.93
[pip3] nvidia-cuda-runtime-cu12==12.8.90
[pip3] nvidia-cudnn-cu12==9.10.2.21
[pip3] nvidia-cudnn-frontend==1.18.0
[pip3] nvidia-cufft-cu12==11.3.3.83
[pip3] nvidia-cufile-cu12==1.13.1.3
[pip3] nvidia-curand-cu12==10.3.9.90
[pip3] nvidia-cusolver-cu12==11.7.3.90
[pip3] nvidia-cusparse-cu12==12.5.8.93
[pip3] nvidia-cusparselt-cu12==0.7.1
[pip3] nvidia-cutlass-dsl==4.4.1
[pip3] nvidia-cutlass-dsl-libs-base==4.4.1
[pip3] nvidia-ml-py==13.590.48
[pip3] nvidia-nccl-cu12==2.27.5
[pip3] nvidia-nvjitlink-cu12==12.8.93
[pip3] nvidia-nvshmem-cu12==3.4.5
[pip3] nvidia-nvtx-cu12==12.8.90
[pip3] open_clip_torch==2.32.0
[pip3] pytorch-lightning==2.6.1
[pip3] pyzmq==27.1.0
[pip3] segmentation_models_pytorch==0.5.0
[pip3] sentence-transformers==5.3.0
[pip3] terratorch==1.2.4
[pip3] torch==2.10.0
[pip3] torch_c_dlpack_ext==0.1.5
[pip3] torchaudio==2.10.0
[pip3] torchgeo==0.9.0
[pip3] torchmetrics==1.9.0
[pip3] torchvision==0.25.0
[pip3] transformers==4.57.5
[pip3] transformers-stream-generator==0.0.5
[pip3] triton==3.6.0
[pip3] tritonclient==2.66.0
[pip3] vector-quantize-pytorch==1.27.21
[conda] Could not collect

==============================
         vLLM Info
==============================
ROCM Version                 : Could not collect
vLLM Version                 : 0.17.1rc1.dev237+g0a0a1a198 (git sha: 0a0a1a198)
vLLM Build Flags:
  CUDA Archs: Not Set; ROCm: Disabled
GPU Topology:
        GPU0    GPU1    GPU2    GPU3    GPU4    GPU5    GPU6    GPU7    CPU Affinity    NUMA Affinity   GPU NUMA ID
GPU0     X      NV18    NV18    NV18    NV18    NV18    NV18    NV18    0-79    0               N/A
GPU1    NV18     X      NV18    NV18    NV18    NV18    NV18    NV18    0-79    0               N/A
GPU2    NV18    NV18     X      NV18    NV18    NV18    NV18    NV18    0-79    0               N/A
GPU3    NV18    NV18    NV18     X      NV18    NV18    NV18    NV18    0-79    0               N/A
GPU4    NV18    NV18    NV18    NV18     X      NV18    NV18    NV18    80-159  1               N/A
GPU5    NV18    NV18    NV18    NV18    NV18     X      NV18    NV18    80-159  1               N/A
GPU6    NV18    NV18    NV18    NV18    NV18    NV18     X      NV18    80-159  1               N/A
GPU7    NV18    NV18    NV18    NV18    NV18    NV18    NV18     X      80-159  1               N/A

Legend:

  X    = Self
  SYS  = Connection traversing PCIe as well as the SMP interconnect between NUMA nodes (e.g., QPI/UPI)
  NODE = Connection traversing PCIe as well as the interconnect between PCIe Host Bridges within a NUMA node
  PHB  = Connection traversing PCIe as well as a PCIe Host Bridge (typically the CPU)
  PXB  = Connection traversing multiple PCIe bridges (without traversing the PCIe Host Bridge)
  PIX  = Connection traversing at most a single PCIe bridge
  NV#  = Connection traversing a bonded set of # NVLinks

==============================
     Environment Variables
==============================
VLLM_CACHE_ROOT==<...>
LD_LIBRARY_PATH=:/usr/local/cuda-13.0/lib64:/usr/local/cuda-13.0/lib64:/usr/local/cuda-13.0/lib64:/usr/local/cuda-13.0/lib64
CUDA_HOME=/usr/local/cuda-13.0
CUDA_HOME=/usr/local/cuda-13.0
PYTORCH_NVML_BASED_CUDA_CHECK=1
TORCHINDUCTOR_COMPILE_THREADS=1
TORCHINDUCTOR_CACHE_DIR=/tmp/torchinductor_yzong-rh
UCX_TLS=cuda_ipc,cuda_copy,tcp

</details>

🐛 Describe the bug

[NixlConnector] Gemma-3 specific heterogeneous TP failures with PD disagg

google/gemma-3-4b-it uses sliding-window + full-attention.

When running with PD disagg in heterogeneous TP (1 prefill + 2 decodes or 2 prefills + 1 decode), gemma-3-4b-it has ~0% accuracy on GSM8K 5-shot, down from ~76%.

Existing sweep in tests/v1/kv_connector/nixl_integration/config_sweep_accuracy_test.sh only tests gemma-3-4b-it with 1 prefill + 1 decode. So this is not caught in CI.

To reproduce:

Pull #37069, which adds to the existing test suite, and run python3 tests/v1/kv_connector/nixl_integration/test_config.py '*gemma*'.

========================================================================
SUMMARY
========================================================================
  [PASS]  tp_gemma_sw_fa.yaml                        167.7s  logs: /home/yzong-rh/vllm/tests/v1/kv_connector/nixl_integration/logs/20260317_163330/tp_gemma_sw_fa.{stdout,stderr}
  [PASS]  tp_gemma_sw_fa_hma.yaml                    171.1s  logs: /home/yzong-rh/vllm/tests/v1/kv_connector/nixl_integration/logs/20260317_163330/tp_gemma_sw_fa_hma.{stdout,stderr}
  [PASS]  tp_gemma_sw_fa_hma_2p2d.yaml               200.4s  logs: /home/yzong-rh/vllm/tests/v1/kv_connector/nixl_integration/logs/20260317_163330/tp_gemma_sw_fa_hma_2p2d.{stdout,stderr}
  [FAIL]  xfail_tp_gemma_sw_fa_1p2d.yaml             239.5s  logs: /home/yzong-rh/vllm/tests/v1/kv_connector/nixl_integration/logs/20260317_163330/xfail_tp_gemma_sw_fa_1p2d.{stdout,stderr}
         error: Command '['bash', '/home/yzong-rh/vllm/tests/v1/kv_connector/nixl_integration/run_accuracy_test.sh']' returned non-zero exit status 1.
  [FAIL]  xfail_tp_gemma_sw_fa_2p1d.yaml             237.2s  logs: /home/yzong-rh/vllm/tests/v1/kv_connector/nixl_integration/logs/20260317_163330/xfail_tp_gemma_sw_fa_2p1d.{stdout,stderr}
         error: Command '['bash', '/home/yzong-rh/vllm/tests/v1/kv_connector/nixl_integration/run_accuracy_test.sh']' returned non-zero exit status 1.
------------------------------------------------------------------------
  Passed: 3  |  Failed: 2  |  Total: 5
========================================================================

Only the heterogeneous configs fail.

xfail_tp_gemma_sw_fa_1p2d:

E       AssertionError: Expected: 0.76 | Measured: 0.000758150113722517
E       assert ((np.float64(0.000758150113722517) - 0.03) < 0.76 and (np.float64(0.000758150113722517) + 0.03) > 0.76)

xfail_tp_gemma_sw_fa_2p1d:

E       AssertionError: Expected: 0.76 | Measured: 0.000758150113722517
E       assert ((np.float64(0.000758150113722517) - 0.03) < 0.76 and (np.float64(0.000758150113722517) + 0.03) > 0.76)

Alternatively,

Replace tp_configs in config_sweep_accuracy_test.sh with:

tp_configs = (
    "GPU_MEMORY_UTILIZATION=0.8 MODEL_NAMES=google/gemma-3-4b-it PREFILLER_TP_SIZE=1 DECODER_TP_SIZE=2 VLLM_SERVE_EXTRA_ARGS=--max-model-len,8192"
)

and run bash v1/kv_connector/nixl_integration/config_sweep_accuracy_test.sh

E       AssertionError: Expected: 0.76 | Measured: 0.000758150113722517
E       assert ((np.float64(0.000758150113722517) - 0.03) < 0.76 and (np.float64(0.000758150113722517) + 0.03) > 0.76)

v1/kv_connector/nixl_integration/test_accuracy.py:73: AssertionError
============================================================== warnings summary ===============================================================
<frozen importlib._bootstrap>:488
  <frozen importlib._bootstrap>:488: DeprecationWarning: builtin type SwigPyPacked has no __module__ attribute

<frozen importlib._bootstrap>:488
  <frozen importlib._bootstrap>:488: DeprecationWarning: builtin type SwigPyObject has no __module__ attribute

-- Docs: https://docs.pytest.org/en/stable/how-to/capture-warnings.html
=========================================================== short test summary info ===========================================================
FAILED v1/kv_connector/nixl_integration/test_accuracy.py::test_accuracy - AssertionError: Expected: 0.76 | Measured: 0.000758150113722517
!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!! stopping after 1 failures !!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!
================================================== 1 failed, 2 warnings in 149.34s (0:02:29) ==================================================

Note that this seems to be a gemma-3-4b-it specific issue since gpt-oss-20b, which is also a FA + SWA hybrid, does not suffer from this -- it has similar accuracy whether in 1p1d or 2p1d/1p2d.

Full logs: results.tar.gz

Before submitting a new issue...

Make sure you already searched for relevant issues, and asked the chatbot living at the bottom right corner of the documentation page, which can answer lots of frequently asked questions.

extent analysis

Fix Plan

To address the issue of gemma-3-4b-it having ~0% accuracy on GSM8K 5-shot when running with PD disagg in heterogeneous TP, we need to modify the configuration and potentially the code. Here are the steps:

Update tp_configs in config_sweep_accuracy_test.sh: Modify the tp_configs variable to include the correct settings for gemma-3-4b-it. For example:

tp_configs = ( "GPU_MEMORY_UTILIZATION=0.8 MODEL_NAMES=google/gemma-3-4b-it PREFILLER_TP_SIZE=1 DECODER_TP_SIZE=2 VLLM_SERVE_EXTRA_ARGS=--max-model-len,8192" )

*   **Adjust `PREFILLER_TP_SIZE` and `DECODER_TP_SIZE`**: Experiment with different values for `PREFILLER_TP_SIZE` and `DECODER_TP_SIZE` to find the optimal combination for `gemma-3-4b-it`.
*   **Modify `test_accuracy.py`**: Update the `test_accuracy` function to handle the new configuration and ensure that the accuracy calculation is correct.
*   **Verify `VLLM_SERVE_EXTRA_ARGS`**: Confirm that the `--max-model-len` argument is set correctly and adjust if necessary.

Example code snippet for updating `test_accuracy.py`:
```python
import numpy as np

def test_accuracy():
    # ... (existing code)

    # Update the accuracy calculation to handle the new configuration
    expected_accuracy = 0.76
    measured_accuracy = np.float64(0.000758150113722517)
    assert ((measured_accuracy - 0.03) < expected_accuracy and (measured_accuracy + 0.03) > expected_accuracy)

    # ... (existing code)

Verification

To verify that the fix worked, run the updated config_sweep_accuracy_test.sh script and check the output for the gemma-3-4b-it model. The accuracy should be closer to the expected value of 0.76.

Extra Tips

Ensure that the gemma-3-4b-it model is correctly configured and that the PREFILLER_TP_SIZE and DECODER_TP_SIZE values are optimal for the model.
If issues persist, try adjusting the GPU_MEMORY_UTILIZATION value or exploring other configuration options.
Consider adding additional logging or debugging statements to test_accuracy.py to help identify any remaining issues.

Vote matrix · Quick signals

Works

Did the solution work? Tap to confirm.

Easy Fix

Was it a quick fix?

Time Saver

Did it save you time?

Blocking

Was it severely blocking?

Common Issue

Are others likely hitting this too?

Flaky / Intermittent

Is it intermittent?

Verified / Reproducible

Can you reproduce it reliably?

#api #ssr #installation #tensor shape #environment variable #ISR setup #authentication setup #request error

Still need to ship something?

×6

Another batch ranked right after the header list — different links, same matching logic.

Data

Security

Network

Code

UI/UX

Text

System

Multimedia

Protocol

API

Engineering

vllm - ✅(Solved) Fix [Bug]: Gemma-3 specific heterogeneous TP failures with PD disagg [1 pull requests, 1 comments, 2 participants]

Recommended Tools

GitHub issue graph ai analysis

Error Message

======================================================================== SUMMARY

Passed: 3 | Failed: 2 | Total: 5

Fix Action

Fix / Workaround

============================== CPU Info

PR fix notes

PR #37940: [NIXL][BUG] Fix Triton heterogeneous TP

Description (problem / solution / changelog)

Purpose

Test Plan

Test Result

Changed files

Code Example

Your current environment

🐛 Describe the bug

[NixlConnector] Gemma-3 specific heterogeneous TP failures with PD disagg

To reproduce:

Before submitting a new issue...

extent analysis

Fix Plan

Verification

Extra Tips

Still need to ship something?

TRENDING

vllm - ✅(Solved) Fix [Bug]: Gemma-3 specific heterogeneous TP failures with PD disagg [1 pull requests, 1 comments, 2 participants]

Recommended Tools

GitHub issue graph ai analysis

Error Message

======================================================================== SUMMARY

Passed: 3 | Failed: 2 | Total: 5

Fix Action

Fix / Workaround

============================== CPU Info

PR fix notes

PR #37940: [NIXL][BUG] Fix Triton heterogeneous TP

Description (problem / solution / changelog)

Purpose

Test Plan

Test Result

Changed files

Code Example

Your current environment

🐛 Describe the bug

[NixlConnector] Gemma-3 specific heterogeneous TP failures with PD disagg

To reproduce:

Before submitting a new issue...

extent analysis

Fix Plan

Verification

Extra Tips

Still need to ship something?

RELATED_DISCOVERY

TRENDING