Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Jump to bottom

[Bug]: VLMs shutting down MQLLMEngine with more than 1 concurrent requests #841

Open
1 task done
akarX23 opened this issue Feb 19, 2025 · 3 comments
Open
1 task done

[Bug]: VLMs shutting down MQLLMEngine with more than 1 concurrent requests #841

akarX23 opened this issue Feb 19, 2025 · 3 comments
Labels
bug Something isn't working

Comments

@akarX23
Copy link

akarX23 commented Feb 19, 2025

Your current environment

The output of `python collect_env.py` 
INFO 02-19 17:34:38 __init__.py:188] Automatically detected platform hpu.
Collecting environment information...
PyTorch version: 2.5.1a0+git6fc067b
Is debug build: False
CUDA used to build PyTorch: None
ROCM used to build PyTorch: N/A

OS: Ubuntu 24.04 LTS (x86_64)
GCC version: (Ubuntu 13.3.0-6ubuntu2~24.04) 13.3.0
Clang version: Could not collect
CMake version: version 3.28.3
Libc version: glibc-2.39

Python version: 3.12.3 (main, Nov  6 2024, 18:32:19) [GCC 13.2.0] (64-bit runtime)
Python platform: Linux-6.8.0-52-generic-x86_64-with-glibc2.39
Is CUDA available: False
CUDA runtime version: No CUDA
CUDA_MODULE_LOADING set to: N/A
GPU models and configuration: No CUDA
Nvidia driver version: No CUDA
cuDNN version: No CUDA
HIP runtime version: N/A
MIOpen runtime version: N/A
Is XNNPACK available: True

CPU:
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(R) Xeon(R) Platinum 8380 CPU @ 2.30GHz
CPU family:                           6
Model:                                106
Thread(s) per core:                   2
Core(s) per socket:                   40
Socket(s):                            2
Stepping:                             6
CPU(s) scaling MHz:                   36%
CPU max MHz:                          3400.0000
CPU min MHz:                          800.0000
BogoMIPS:                             4600.00
Flags:                                fpu vme de pse tsc msr pae mce cx8 apic sep mtrr pge mca cmov pat pse36 clflush dts acpi mmx fxsr sse sse2 ss ht tm pbe syscall nx pdpe1gb rdtscp lm constant_tsc art arch_perfmon pebs bts rep_good nopl xtopology nonstop_tsc cpuid aperfmperf pni pclmulqdq dtes64 monitor ds_cpl vmx smx est tm2 ssse3 sdbg fma cx16 xtpr pdcm pcid dca sse4_1 sse4_2 x2apic movbe popcnt tsc_deadline_timer aes xsave avx f16c rdrand lahf_lm abm 3dnowprefetch cpuid_fault epb cat_l3 intel_ppin ssbd mba ibrs ibpb stibp ibrs_enhanced tpr_shadow flexpriority ept vpid ept_ad fsgsbase tsc_adjust bmi1 avx2 smep bmi2 erms invpcid cqm rdt_a avx512f avx512dq rdseed adx smap avx512ifma clflushopt clwb intel_pt avx512cd sha_ni avx512bw avx512vl xsaveopt xsavec xgetbv1 xsaves cqm_llc cqm_occup_llc cqm_mbm_total cqm_mbm_local split_lock_detect wbnoinvd dtherm ida arat pln pts hwp hwp_act_window hwp_epp hwp_pkg_req vnmi avx512vbmi umip pku ospke avx512_vbmi2 gfni vaes vpclmulqdq avx512_vnni avx512_bitalg tme avx512_vpopcntdq la57 rdpid fsrm md_clear pconfig flush_l1d arch_capabilities
Virtualization:                       VT-x
L1d cache:                            3.8 MiB (80 instances)
L1i cache:                            2.5 MiB (80 instances)
L2 cache:                             100 MiB (80 instances)
L3 cache:                             120 MiB (2 instances)
NUMA node(s):                         2
NUMA node0 CPU(s):                    0-39,80-119
NUMA node1 CPU(s):                    40-79,120-159
Vulnerability Gather data sampling:   Mitigation; Microcode
Vulnerability Itlb multihit:          Not affected
Vulnerability L1tf:                   Not affected
Vulnerability Mds:                    Not affected
Vulnerability Meltdown:               Not affected
Vulnerability Mmio stale data:        Mitigation; Clear CPU buffers; SMT vulnerable
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; RSB filling; PBRSB-eIBRS SW sequence; BHI SW loop, KVM SW loop
Vulnerability Srbds:                  Not affected
Vulnerability Tsx async abort:        Not affected

Versions of relevant libraries:
[pip3] habana-torch-dataloader==1.19.0.561
[pip3] habana-torch-plugin==1.19.0.561
[pip3] numpy==1.26.4
[pip3] pynvml==8.0.4
[pip3] pytorch-lightning==2.4.0
[pip3] pyzmq==26.2.1
[pip3] torch==2.5.1a0+git6fc067b
[pip3] torch_tb_profiler==0.4.0
[pip3] torchaudio==2.5.1a0+1661daf
[pip3] torchdata==0.9.0+d4bb3e6
[pip3] torchmetrics==1.6.0
[pip3] torchtext==0.18.0a0+9bed85d
[pip3] torchvision==0.20.1a0+3ac97aa
[pip3] transformers==4.49.0
[pip3] triton==3.1.0
[conda] Could not collect
ROCM Version: Could not collect
Neuron SDK Version: N/A
vLLM Version: 0.6.3.dev2006+gc3ce6ae77
vLLM Build Flags:
CUDA Archs: Not Set; ROCm: Disabled; Neuron: Disabled
GPU Topology:
Could not collect

LD_LIBRARY_PATH=/usr/local/lib/python3.12/dist-packages/cv2/../../lib64:/opt/habanalabs/libfabric-1.22.0/lib:/opt/amazon/openmpi/lib:/usr/lib/habanalabs:

🐛 Describe the bug

When I try to use VLMs like llava-hf/llava-v1.6-mistral-7b-hf or Llama-3.2-11B-Vision-Instruct, the vLLM server starts without issues. Command used:

VLLM_SKIP_WARMUP=true vllm serve meta-llama/Llama-3.2-11B-Vision-Instruct --enforce-eager --max-model-len 8192 --max_num_seqs 16

Now I have a Python script that sends requests to this server using images in a particular folder. I can specify the total number of requests that need to be completed and the concurrency with which requests are sent.

When I specify concurrency as 1, everything seems to run fine and the vLLM server output looks like this:

INFO 02-19 17:45:38 engine.py:270] Added request chatcmpl-f999587430c74f35b54a8776a8738206.
INFO:     127.0.0.1:47262 - "POST /v1/chat/completions HTTP/1.1" 200 OK
INFO 02-19 17:45:40 metrics.py:467] Avg prompt throughput: 20.9 tokens/s, Avg generation throughput: 4.1 tokens/s, Running: 0 reqs, Swapped: 0 reqs, Pending: 0 reqs, GPU KV cache usage: 0.0%, CPU KV cache usage: 0.0%.
INFO 02-19 17:45:40 logger.py:37] Received request chatcmpl-b1110b43342547b59ccde5040ef50cf2: prompt: '<|begin_of_text|><|start_header_id|>user<|end_header_id|>\n\nDoes this image contain sexually suggestive or provocative content? Just say yes or no.<|image|><|eot_id|><|start_header_id|>assistant<|end_header_id|>\n\n', params: SamplingParams(n=1, presence_penalty=0.0, frequency_penalty=0.0, repetition_penalty=1.0, temperature=1.0, top_p=1.0, top_k=-1, min_p=0.0, seed=None, stop=[], stop_token_ids=[], bad_words=[], include_stop_str_in_output=False, ignore_eos=False, max_tokens=10, min_tokens=0, logprobs=None, prompt_logprobs=None, skip_special_tokens=True, spaces_between_special_tokens=True, truncate_prompt_tokens=None, guided_decoding=None), prompt_token_ids: None, lora_request: None, prompt_adapter_request: None.
INFO 02-19 17:45:40 engine.py:270] Added request chatcmpl-b1110b43342547b59ccde5040ef50cf2.
INFO:     127.0.0.1:47274 - "POST /v1/chat/completions HTTP/1.1" 200 OK
INFO 02-19 17:45:41 logger.py:37] Received request chatcmpl-f12842d229124c8aadcecb85596ed043: prompt: '<|begin_of_text|><|start_header_id|>user<|end_header_id|>\n\nDoes this image contain sexually suggestive or provocative content? Just say yes or no.<|image|><|eot_id|><|start_header_id|>assistant<|end_header_id|>\n\n', params: SamplingParams(n=1, presence_penalty=0.0, frequency_penalty=0.0, repetition_penalty=1.0, temperature=1.0, top_p=1.0, top_k=-1, min_p=0.0, seed=None, stop=[], stop_token_ids=[], bad_words=[], include_stop_str_in_output=False, ignore_eos=False, max_tokens=10, min_tokens=0, logprobs=None, prompt_logprobs=None, skip_special_tokens=True, spaces_between_special_tokens=True, truncate_prompt_tokens=None, guided_decoding=None), prompt_token_ids: None, lora_request: None, prompt_adapter_request: None.
INFO 02-19 17:45:41 engine.py:270] Added request chatcmpl-f12842d229124c8aadcecb85596ed043.
INFO:     127.0.0.1:47276 - "POST /v1/chat/completions HTTP/1.1" 200 OK
INFO 02-19 17:45:42 logger.py:37] Received request chatcmpl-b755a73bbe214cbea59145874cfbeddf: prompt: '<|begin_of_text|><|start_header_id|>user<|end_header_id|>\n\nDoes this image contain sexually suggestive or provocative content? Just say yes or no.<|image|><|eot_id|><|start_header_id|>assistant<|end_header_id|>\n\n', params: SamplingParams(n=1, presence_penalty=0.0, frequency_penalty=0.0, repetition_penalty=1.0, temperature=1.0, top_p=1.0, top_k=-1, min_p=0.0, seed=None, stop=[], stop_token_ids=[], bad_words=[], include_stop_str_in_output=False, ignore_eos=False, max_tokens=10, min_tokens=0, logprobs=None, prompt_logprobs=None, skip_special_tokens=True, spaces_between_special_tokens=True, truncate_prompt_tokens=None, guided_decoding=None), prompt_token_ids: None, lora_request: None, prompt_adapter_request: None.
INFO 02-19 17:45:42 engine.py:270] Added request chatcmpl-b755a73bbe214cbea59145874cfbeddf.
INFO:     127.0.0.1:47288 - "POST /v1/chat/completions HTTP/1.1" 200 OK

However, when I make the concurrency to 2, first few requests go through and then the MQLLMEngine crashes, giving this output:

INFO 02-19 17:47:46 engine.py:270] Added request chatcmpl-b320da69e6614d9d968d70c838a62bbe.
INFO:     127.0.0.1:33578 - "POST /v1/chat/completions HTTP/1.1" 200 OK
INFO:     127.0.0.1:33584 - "POST /v1/chat/completions HTTP/1.1" 200 OK
INFO 02-19 17:47:48 logger.py:37] Received request chatcmpl-c7fabc22bf4f4888ab26c65d51cec8e1: prompt: '<|begin_of_text|><|start_header_id|>user<|end_header_id|>\n\nDoes this image contain sexually suggestive or provocative content? Just say yes or no.<|image|><|eot_id|><|start_header_id|>assistant<|end_header_id|>\n\n', params: SamplingParams(n=1, presence_penalty=0.0, frequency_penalty=0.0, repetition_penalty=1.0, temperature=1.0, top_p=1.0, top_k=-1, min_p=0.0, seed=None, stop=[], stop_token_ids=[], bad_words=[], include_stop_str_in_output=False, ignore_eos=False, max_tokens=10, min_tokens=0, logprobs=None, prompt_logprobs=None, skip_special_tokens=True, spaces_between_special_tokens=True, truncate_prompt_tokens=None, guided_decoding=None), prompt_token_ids: None, lora_request: None, prompt_adapter_request: None.
INFO 02-19 17:47:48 logger.py:37] Received request chatcmpl-9997c92d2aa04412b56c7154d5fdc156: prompt: '<|begin_of_text|><|start_header_id|>user<|end_header_id|>\n\nDoes this image contain sexually suggestive or provocative content? Just say yes or no.<|image|><|eot_id|><|start_header_id|>assistant<|end_header_id|>\n\n', params: SamplingParams(n=1, presence_penalty=0.0, frequency_penalty=0.0, repetition_penalty=1.0, temperature=1.0, top_p=1.0, top_k=-1, min_p=0.0, seed=None, stop=[], stop_token_ids=[], bad_words=[], include_stop_str_in_output=False, ignore_eos=False, max_tokens=10, min_tokens=0, logprobs=None, prompt_logprobs=None, skip_special_tokens=True, spaces_between_special_tokens=True, truncate_prompt_tokens=None, guided_decoding=None), prompt_token_ids: None, lora_request: None, prompt_adapter_request: None.
INFO 02-19 17:47:48 engine.py:270] Added request chatcmpl-c7fabc22bf4f4888ab26c65d51cec8e1.
INFO 02-19 17:47:48 engine.py:270] Added request chatcmpl-9997c92d2aa04412b56c7154d5fdc156.
INFO:     127.0.0.1:33600 - "POST /v1/chat/completions HTTP/1.1" 200 OK
INFO:     127.0.0.1:33610 - "POST /v1/chat/completions HTTP/1.1" 200 OK
INFO 02-19 17:47:49 logger.py:37] Received request chatcmpl-6e65512fe55742598110dc6439fa4cb7: prompt: '<|begin_of_text|><|start_header_id|>user<|end_header_id|>\n\nDoes this image contain sexually suggestive or provocative content? Just say yes or no.<|image|><|eot_id|><|start_header_id|>assistant<|end_header_id|>\n\n', params: SamplingParams(n=1, presence_penalty=0.0, frequency_penalty=0.0, repetition_penalty=1.0, temperature=1.0, top_p=1.0, top_k=-1, min_p=0.0, seed=None, stop=[], stop_token_ids=[], bad_words=[], include_stop_str_in_output=False, ignore_eos=False, max_tokens=10, min_tokens=0, logprobs=None, prompt_logprobs=None, skip_special_tokens=True, spaces_between_special_tokens=True, truncate_prompt_tokens=None, guided_decoding=None), prompt_token_ids: None, lora_request: None, prompt_adapter_request: None.
INFO 02-19 17:47:49 logger.py:37] Received request chatcmpl-31efce9bad0643979e4dff9163dbde73: prompt: '<|begin_of_text|><|start_header_id|>user<|end_header_id|>\n\nDoes this image contain sexually suggestive or provocative content? Just say yes or no.<|image|><|eot_id|><|start_header_id|>assistant<|end_header_id|>\n\n', params: SamplingParams(n=1, presence_penalty=0.0, frequency_penalty=0.0, repetition_penalty=1.0, temperature=1.0, top_p=1.0, top_k=-1, min_p=0.0, seed=None, stop=[], stop_token_ids=[], bad_words=[], include_stop_str_in_output=False, ignore_eos=False, max_tokens=10, min_tokens=0, logprobs=None, prompt_logprobs=None, skip_special_tokens=True, spaces_between_special_tokens=True, truncate_prompt_tokens=None, guided_decoding=None), prompt_token_ids: None, lora_request: None, prompt_adapter_request: None.
INFO 02-19 17:47:49 engine.py:270] Added request chatcmpl-6e65512fe55742598110dc6439fa4cb7.
INFO 02-19 17:47:49 engine.py:270] Added request chatcmpl-31efce9bad0643979e4dff9163dbde73.
CRITICAL 02-19 17:47:49 launcher.py:99] MQLLMEngine is already dead, terminating server process
INFO:     127.0.0.1:32912 - "POST /v1/chat/completions HTTP/1.1" 500 Internal Server Error
CRITICAL 02-19 17:47:49 launcher.py:99] MQLLMEngine is already dead, terminating server process
INFO:     127.0.0.1:32926 - "POST /v1/chat/completions HTTP/1.1" 500 Internal Server Error

The traceback looks like this in the vLLM server:

ERROR 02-19 17:47:49 engine.py:136] AssertionError()
ERROR 02-19 17:47:49 engine.py:136] Traceback (most recent call last):
ERROR 02-19 17:47:49 engine.py:136]   File "/root/akarx/vllm-fork/vllm/engine/multiprocessing/engine.py", line 134, in start
ERROR 02-19 17:47:49 engine.py:136]     self.run_engine_loop()
ERROR 02-19 17:47:49 engine.py:136]   File "/root/akarx/vllm-fork/vllm/engine/multiprocessing/engine.py", line 197, in run_engine_loop
ERROR 02-19 17:47:49 engine.py:136]     request_outputs = self.engine_step()
ERROR 02-19 17:47:49 engine.py:136]                       ^^^^^^^^^^^^^^^^^^
ERROR 02-19 17:47:49 engine.py:136]   File "/root/akarx/vllm-fork/vllm/engine/multiprocessing/engine.py", line 215, in engine_step
ERROR 02-19 17:47:49 engine.py:136]     raise e
ERROR 02-19 17:47:49 engine.py:136]   File "/root/akarx/vllm-fork/vllm/engine/multiprocessing/engine.py", line 206, in engine_step
ERROR 02-19 17:47:49 engine.py:136]     return self.engine.step()
ERROR 02-19 17:47:49 engine.py:136]            ^^^^^^^^^^^^^^^^^^
ERROR 02-19 17:47:49 engine.py:136]   File "/root/akarx/vllm-fork/vllm/engine/llm_engine.py", line 1359, in step
ERROR 02-19 17:47:49 engine.py:136]     outputs = self.model_executor.execute_model(
ERROR 02-19 17:47:49 engine.py:136]               ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
ERROR 02-19 17:47:49 engine.py:136]   File "/root/akarx/vllm-fork/vllm/executor/executor_base.py", line 113, in execute_model
ERROR 02-19 17:47:49 engine.py:136]     output = self.collective_rpc("execute_model",
ERROR 02-19 17:47:49 engine.py:136]              ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
ERROR 02-19 17:47:49 engine.py:136]   File "/root/akarx/vllm-fork/vllm/executor/uniproc_executor.py", line 51, in collective_rpc
ERROR 02-19 17:47:49 engine.py:136]     answer = run_method(self.driver_worker, method, args, kwargs)
ERROR 02-19 17:47:49 engine.py:136]              ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
ERROR 02-19 17:47:49 engine.py:136]   File "/root/akarx/vllm-fork/vllm/utils.py", line 2288, in run_method
ERROR 02-19 17:47:49 engine.py:136]     return func(*args, **kwargs)
ERROR 02-19 17:47:49 engine.py:136]            ^^^^^^^^^^^^^^^^^^^^^
ERROR 02-19 17:47:49 engine.py:136]   File "/root/akarx/vllm-fork/vllm/worker/hpu_worker.py", line 281, in execute_model
ERROR 02-19 17:47:49 engine.py:136]     output = LocalOrDistributedWorkerBase.execute_model(
ERROR 02-19 17:47:49 engine.py:136]              ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
ERROR 02-19 17:47:49 engine.py:136]   File "/root/akarx/vllm-fork/vllm/worker/worker_base.py", line 402, in execute_model
ERROR 02-19 17:47:49 engine.py:136]     output = self.model_runner.execute_model(
ERROR 02-19 17:47:49 engine.py:136]              ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
ERROR 02-19 17:47:49 engine.py:136]   File "/usr/local/lib/python3.12/dist-packages/torch/utils/_contextlib.py", line 116, in decorate_context
ERROR 02-19 17:47:49 engine.py:136]     return func(*args, **kwargs)
ERROR 02-19 17:47:49 engine.py:136]            ^^^^^^^^^^^^^^^^^^^^^
ERROR 02-19 17:47:49 engine.py:136]   File "/root/akarx/vllm-fork/vllm/worker/hpu_enc_dec_model_runner.py", line 658, in execute_model
ERROR 02-19 17:47:49 engine.py:136]     hidden_states = self.model.forward(
ERROR 02-19 17:47:49 engine.py:136]                     ^^^^^^^^^^^^^^^^^^^
ERROR 02-19 17:47:49 engine.py:136]   File "/root/akarx/vllm-fork/vllm/worker/hpu_enc_dec_model_runner.py", line 162, in forward
ERROR 02-19 17:47:49 engine.py:136]     hidden_states = self.model(*args, **kwargs)
ERROR 02-19 17:47:49 engine.py:136]                     ^^^^^^^^^^^^^^^^^^^^^^^^^^^
ERROR 02-19 17:47:49 engine.py:136]   File "/usr/local/lib/python3.12/dist-packages/torch/nn/modules/module.py", line 1739, in _wrapped_call_impl
ERROR 02-19 17:47:49 engine.py:136]     return self._call_impl(*args, **kwargs)
ERROR 02-19 17:47:49 engine.py:136]            ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
ERROR 02-19 17:47:49 engine.py:136]   File "/usr/local/lib/python3.12/dist-packages/torch/nn/modules/module.py", line 1750, in _call_impl
ERROR 02-19 17:47:49 engine.py:136]     return forward_call(*args, **kwargs)
ERROR 02-19 17:47:49 engine.py:136]            ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
ERROR 02-19 17:47:49 engine.py:136]   File "/root/akarx/vllm-fork/vllm/model_executor/models/mllama.py", line 1419, in forward
ERROR 02-19 17:47:49 engine.py:136]     outputs = self.language_model(
ERROR 02-19 17:47:49 engine.py:136]               ^^^^^^^^^^^^^^^^^^^^
ERROR 02-19 17:47:49 engine.py:136]   File "/usr/local/lib/python3.12/dist-packages/torch/nn/modules/module.py", line 1739, in _wrapped_call_impl
ERROR 02-19 17:47:49 engine.py:136]     return self._call_impl(*args, **kwargs)
ERROR 02-19 17:47:49 engine.py:136]            ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
ERROR 02-19 17:47:49 engine.py:136]   File "/usr/local/lib/python3.12/dist-packages/torch/nn/modules/module.py", line 1847, in _call_impl
ERROR 02-19 17:47:49 engine.py:136]     return inner()
ERROR 02-19 17:47:49 engine.py:136]            ^^^^^^^
ERROR 02-19 17:47:49 engine.py:136]   File "/usr/local/lib/python3.12/dist-packages/torch/nn/modules/module.py", line 1793, in inner
ERROR 02-19 17:47:49 engine.py:136]     result = forward_call(*args, **kwargs)
ERROR 02-19 17:47:49 engine.py:136]              ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
ERROR 02-19 17:47:49 engine.py:136]   File "/usr/local/lib/python3.12/dist-packages/habana_frameworks/torch/hpu/graphs.py", line 736, in forward
ERROR 02-19 17:47:49 engine.py:136]     return wrapped_hpugraph_forward(
ERROR 02-19 17:47:49 engine.py:136]            ^^^^^^^^^^^^^^^^^^^^^^^^^
ERROR 02-19 17:47:49 engine.py:136]   File "/usr/local/lib/python3.12/dist-packages/habana_frameworks/torch/hpu/graphs.py", line 586, in wrapped_hpugraph_forward
ERROR 02-19 17:47:49 engine.py:136]     return orig_fwd(*args, **kwargs)
ERROR 02-19 17:47:49 engine.py:136]            ^^^^^^^^^^^^^^^^^^^^^^^^^
ERROR 02-19 17:47:49 engine.py:136]   File "/root/akarx/vllm-fork/vllm/model_executor/models/mllama.py", line 1107, in forward
ERROR 02-19 17:47:49 engine.py:136]     hidden_states = self.model(
ERROR 02-19 17:47:49 engine.py:136]                     ^^^^^^^^^^^
ERROR 02-19 17:47:49 engine.py:136]   File "/usr/local/lib/python3.12/dist-packages/torch/nn/modules/module.py", line 1739, in _wrapped_call_impl
ERROR 02-19 17:47:49 engine.py:136]     return self._call_impl(*args, **kwargs)
ERROR 02-19 17:47:49 engine.py:136]            ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
ERROR 02-19 17:47:49 engine.py:136]   File "/usr/local/lib/python3.12/dist-packages/torch/nn/modules/module.py", line 1847, in _call_impl
ERROR 02-19 17:47:49 engine.py:136]     return inner()
ERROR 02-19 17:47:49 engine.py:136]            ^^^^^^^
ERROR 02-19 17:47:49 engine.py:136]   File "/usr/local/lib/python3.12/dist-packages/torch/nn/modules/module.py", line 1793, in inner
ERROR 02-19 17:47:49 engine.py:136]     result = forward_call(*args, **kwargs)
ERROR 02-19 17:47:49 engine.py:136]              ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
ERROR 02-19 17:47:49 engine.py:136]   File "/root/akarx/vllm-fork/vllm/model_executor/models/mllama.py", line 1043, in forward
ERROR 02-19 17:47:49 engine.py:136]     hidden_states = decoder_layer(
ERROR 02-19 17:47:49 engine.py:136]                     ^^^^^^^^^^^^^^
ERROR 02-19 17:47:49 engine.py:136]   File "/usr/local/lib/python3.12/dist-packages/torch/nn/modules/module.py", line 1739, in _wrapped_call_impl
ERROR 02-19 17:47:49 engine.py:136]     return self._call_impl(*args, **kwargs)
ERROR 02-19 17:47:49 engine.py:136]            ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
ERROR 02-19 17:47:49 engine.py:136]   File "/usr/local/lib/python3.12/dist-packages/torch/nn/modules/module.py", line 1847, in _call_impl
ERROR 02-19 17:47:49 engine.py:136]     return inner()
ERROR 02-19 17:47:49 engine.py:136]            ^^^^^^^
ERROR 02-19 17:47:49 engine.py:136]   File "/usr/local/lib/python3.12/dist-packages/torch/nn/modules/module.py", line 1793, in inner
ERROR 02-19 17:47:49 engine.py:136]     result = forward_call(*args, **kwargs)
ERROR 02-19 17:47:49 engine.py:136]              ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
ERROR 02-19 17:47:49 engine.py:136]   File "/root/akarx/vllm-fork/vllm/model_executor/models/mllama.py", line 950, in forward
ERROR 02-19 17:47:49 engine.py:136]     hidden_states = self.cross_attn(
ERROR 02-19 17:47:49 engine.py:136]                     ^^^^^^^^^^^^^^^^
ERROR 02-19 17:47:49 engine.py:136]   File "/usr/local/lib/python3.12/dist-packages/torch/nn/modules/module.py", line 1739, in _wrapped_call_impl
ERROR 02-19 17:47:49 engine.py:136]     return self._call_impl(*args, **kwargs)
ERROR 02-19 17:47:49 engine.py:136]            ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
ERROR 02-19 17:47:49 engine.py:136]   File "/usr/local/lib/python3.12/dist-packages/torch/nn/modules/module.py", line 1847, in _call_impl
ERROR 02-19 17:47:49 engine.py:136]     return inner()
ERROR 02-19 17:47:49 engine.py:136]            ^^^^^^^
ERROR 02-19 17:47:49 engine.py:136]   File "/usr/local/lib/python3.12/dist-packages/torch/nn/modules/module.py", line 1793, in inner
ERROR 02-19 17:47:49 engine.py:136]     result = forward_call(*args, **kwargs)
ERROR 02-19 17:47:49 engine.py:136]              ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
ERROR 02-19 17:47:49 engine.py:136]   File "/root/akarx/vllm-fork/vllm/model_executor/models/mllama.py", line 820, in forward
ERROR 02-19 17:47:49 engine.py:136]     output = self.attn(
ERROR 02-19 17:47:49 engine.py:136]              ^^^^^^^^^^
ERROR 02-19 17:47:49 engine.py:136]   File "/usr/local/lib/python3.12/dist-packages/torch/nn/modules/module.py", line 1739, in _wrapped_call_impl
ERROR 02-19 17:47:49 engine.py:136]     return self._call_impl(*args, **kwargs)
ERROR 02-19 17:47:49 engine.py:136]            ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
ERROR 02-19 17:47:49 engine.py:136]   File "/usr/local/lib/python3.12/dist-packages/torch/nn/modules/module.py", line 1847, in _call_impl
ERROR 02-19 17:47:49 engine.py:136]     return inner()
ERROR 02-19 17:47:49 engine.py:136]            ^^^^^^^
ERROR 02-19 17:47:49 engine.py:136]   File "/usr/local/lib/python3.12/dist-packages/torch/nn/modules/module.py", line 1793, in inner
ERROR 02-19 17:47:49 engine.py:136]     result = forward_call(*args, **kwargs)
ERROR 02-19 17:47:49 engine.py:136]              ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
ERROR 02-19 17:47:49 engine.py:136]   File "/root/akarx/vllm-fork/vllm/attention/layer.py", line 159, in forward
ERROR 02-19 17:47:49 engine.py:136]     return unified_attention(query, key, value, self.layer_name)
ERROR 02-19 17:47:49 engine.py:136]            ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
ERROR 02-19 17:47:49 engine.py:136]   File "/root/akarx/vllm-fork/vllm/attention/layer.py", line 246, in unified_attention
ERROR 02-19 17:47:49 engine.py:136]     return self.impl.forward(query, key, value, kv_cache, attn_metadata,
ERROR 02-19 17:47:49 engine.py:136]            ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
ERROR 02-19 17:47:49 engine.py:136]   File "/root/akarx/vllm-fork/vllm/attention/backends/hpu_attn.py", line 188, in forward
ERROR 02-19 17:47:49 engine.py:136]     return self.forward_encoder_decoder(
ERROR 02-19 17:47:49 engine.py:136]            ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
ERROR 02-19 17:47:49 engine.py:136]   File "/root/akarx/vllm-fork/vllm/attention/backends/hpu_attn.py", line 323, in forward_encoder_decoder
ERROR 02-19 17:47:49 engine.py:136]     assert batched_kv_tokens % batch_size == 0
ERROR 02-19 17:47:49 engine.py:136]            ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
ERROR 02-19 17:47:49 engine.py:136] AssertionError
CRITICAL 02-19 17:47:49 launcher.py:99] MQLLMEngine is already dead, terminating server process
INFO:     127.0.0.1:32928 - "POST /v1/chat/completions HTTP/1.1" 500 Internal Server Error
CRITICAL 02-19 17:47:49 launcher.py:99] MQLLMEngine is already dead, terminating server process
INFO:     127.0.0.1:32932 - "POST /v1/chat/completions HTTP/1.1" 500 Internal Server Error

This concurrency issue didn't exist last month. I had tested this around 24th Jan 2025, and everything worked that time. This same issue is seen with LLava as well. Please assist, thank you!

I am on the v1.20.0 branch currently.

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.
@akarX23 akarX23 added the bug Something isn't working label Feb 19, 2025
@michalkuligowski
Copy link

Hi @akarX23, thank you for the ticket. Vllm version from your collect_env output seems to not reflect release 1.19 ("vLLM Version: 0.6.3.dev2006+gc3ce6ae77" and should be something like "0.6.4...").
Also recently new version 1.20 was released. Can you please try the latest, by calling, and see if that helps:

$ git clone https://github.com/HabanaAI/vllm-fork.git
$ cd vllm-fork
$ git checkout v0.6.6.post1+Gaudi-1.20.0
$ pip install -r requirements-hpu.txt
$ pip install -e .
$ python setup.py develop

https://github.com/HabanaAI/vllm-fork/releases/tag/v0.6.6.post1%2BGaudi-1.20.0

@akarX23
Copy link
Author

akarX23 commented Feb 26, 2025

Hi @michalkuligowski,

Thank you for directing me to the latest stable branch. I installed vLLM from that branch and reran the steps.

For concurrency 1, everything works well. However, for concurrency 2, I am still getting the same error:

Output of collect_env:

INFO 02-26 23:01:04 __init__.py:188] Automatically detected platform hpu.
Collecting environment information...
PyTorch version: 2.5.1a0+git6fc067b
Is debug build: False
CUDA used to build PyTorch: None
ROCM used to build PyTorch: N/A

OS: Ubuntu 22.04.5 LTS (x86_64)
GCC version: (Ubuntu 11.4.0-1ubuntu1~22.04) 11.4.0
Clang version: Could not collect
CMake version: version 3.22.1
Libc version: glibc-2.35

Python version: 3.10.12 (main, Nov  6 2024, 20:22:13) [GCC 11.4.0] (64-bit runtime)
Python platform: Linux-6.8.0-52-generic-x86_64-with-glibc2.35
Is CUDA available: False
CUDA runtime version: No CUDA
CUDA_MODULE_LOADING set to: N/A
GPU models and configuration: No CUDA
Nvidia driver version: No CUDA
cuDNN version: No CUDA
HIP runtime version: N/A
MIOpen runtime version: N/A
Is XNNPACK available: True

CPU:
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(R) Xeon(R) Platinum 8380 CPU @ 2.30GHz
CPU family:                           6
Model:                                106
Thread(s) per core:                   2
Core(s) per socket:                   40
Socket(s):                            2
Stepping:                             6
CPU max MHz:                          3400.0000
CPU min MHz:                          800.0000
BogoMIPS:                             4600.00
Flags:                                fpu vme de pse tsc msr pae mce cx8 apic sep mtrr pge mca cmov pat pse36 clflush dts acpi mmx fxsr sse sse2 ss ht tm pbe syscall nx pdpe1gb rdtscp lm constant_tsc art arch_perfmon pebs bts rep_good nopl xtopology nonstop_tsc cpuid aperfmperf pni pclmulqdq dtes64 monitor ds_cpl vmx smx est tm2 ssse3 sdbg fma cx16 xtpr pdcm pcid dca sse4_1 sse4_2 x2apic movbe popcnt tsc_deadline_timer aes xsave avx f16c rdrand lahf_lm abm 3dnowprefetch cpuid_fault epb cat_l3 intel_ppin ssbd mba ibrs ibpb stibp ibrs_enhanced tpr_shadow flexpriority ept vpid ept_ad fsgsbase tsc_adjust bmi1 avx2 smep bmi2 erms invpcid cqm rdt_a avx512f avx512dq rdseed adx smap avx512ifma clflushopt clwb intel_pt avx512cd sha_ni avx512bw avx512vl xsaveopt xsavec xgetbv1 xsaves cqm_llc cqm_occup_llc cqm_mbm_total cqm_mbm_local split_lock_detect wbnoinvd dtherm ida arat pln pts hwp hwp_act_window hwp_epp hwp_pkg_req vnmi avx512vbmi umip pku ospke avx512_vbmi2 gfni vaes vpclmulqdq avx512_vnni avx512_bitalg tme avx512_vpopcntdq la57 rdpid fsrm md_clear pconfig flush_l1d arch_capabilities
Virtualization:                       VT-x
L1d cache:                            3.8 MiB (80 instances)
L1i cache:                            2.5 MiB (80 instances)
L2 cache:                             100 MiB (80 instances)
L3 cache:                             120 MiB (2 instances)
NUMA node(s):                         2
NUMA node0 CPU(s):                    0-39,80-119
NUMA node1 CPU(s):                    40-79,120-159
Vulnerability Gather data sampling:   Mitigation; Microcode
Vulnerability Itlb multihit:          Not affected
Vulnerability L1tf:                   Not affected
Vulnerability Mds:                    Not affected
Vulnerability Meltdown:               Not affected
Vulnerability Mmio stale data:        Mitigation; Clear CPU buffers; SMT vulnerable
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; RSB filling; PBRSB-eIBRS SW sequence; BHI SW loop, KVM SW loop
Vulnerability Srbds:                  Not affected
Vulnerability Tsx async abort:        Not affected

Versions of relevant libraries:
[pip3] habana-torch-dataloader==1.19.0.561
[pip3] habana-torch-plugin==1.19.0.561
[pip3] numpy==1.26.4
[pip3] pynvml==8.0.4
[pip3] pytorch-lightning==2.4.0
[pip3] pyzmq==26.2.1
[pip3] torch==2.5.1a0+git6fc067b
[pip3] torch_tb_profiler==0.4.0
[pip3] torchaudio==2.5.1a0+1661daf
[pip3] torchdata==0.9.0+d4bb3e6
[pip3] torchmetrics==1.6.0
[pip3] torchtext==0.18.0a0+9bed85d
[pip3] torchvision==0.20.1a0+3ac97aa
[pip3] transformers==4.49.0
[pip3] triton==3.1.0
[conda] Could not collect
ROCM Version: Could not collect
Neuron SDK Version: N/A
vLLM Version: 0.6.6.post1
vLLM Build Flags:
CUDA Archs: Not Set; ROCm: Disabled; Neuron: Disabled
GPU Topology:
Could not collect

LD_LIBRARY_PATH=/usr/local/lib/python3.10/dist-packages/cv2/../../lib64:/opt/habanalabs/libfabric-1.22.0/lib:/opt/amazon/openmpi/lib:/usr/lib/habanalabs:

@akarX23
Copy link
Author

akarX23 commented Feb 26, 2025

For llava-hf/llava-v1.6-mistral-7b-hf, I start the vllm server using this command:

VLLM_SKIP_WARMUP=true vllm serve llava-hf/llava-v1.6-mistral-7b-hf

The error with 2 concurrent requests is different than what I get with llama:

INFO 02-26 23:33:20 engine.py:270] Added request chatcmpl-f8f0bd35667c40ed90aff588b2a39951.
INFO 02-26 23:33:20 engine.py:270] Added request chatcmpl-5b50ade11291442a9d52627f418c8112.
WARNING 02-26 23:33:20 hpu_model_runner.py:2101] Configuration: (prompt, 2, 2304) was not warmed-up!
CRITICAL 02-26 23:33:21 launcher.py:99] MQLLMEngine is already dead, terminating server process
INFO:     127.0.0.1:53148 - "POST /v1/chat/completions HTTP/1.1" 500 Internal Server Error
CRITICAL 02-26 23:33:21 launcher.py:99] MQLLMEngine is already dead, terminating server process
INFO:     127.0.0.1:53154 - "POST /v1/chat/completions HTTP/1.1" 500 Internal Server Error
ERROR 02-26 23:33:21 engine.py:136] ValueError('Attempted to assign 2160 + 2144 = 4304 multimodal tokens to 0 placeholders')
ERROR 02-26 23:33:21 engine.py:136] Traceback (most recent call last):
ERROR 02-26 23:33:21 engine.py:136]   File "/root/akarx/vllm-fork/vllm/engine/multiprocessing/engine.py", line 134, in start
ERROR 02-26 23:33:21 engine.py:136]     self.run_engine_loop()
ERROR 02-26 23:33:21 engine.py:136]   File "/root/akarx/vllm-fork/vllm/engine/multiprocessing/engine.py", line 197, in run_engine_loop
ERROR 02-26 23:33:21 engine.py:136]     request_outputs = self.engine_step()
ERROR 02-26 23:33:21 engine.py:136]   File "/root/akarx/vllm-fork/vllm/engine/multiprocessing/engine.py", line 215, in engine_step
ERROR 02-26 23:33:21 engine.py:136]     raise e
ERROR 02-26 23:33:21 engine.py:136]   File "/root/akarx/vllm-fork/vllm/engine/multiprocessing/engine.py", line 206, in engine_step
ERROR 02-26 23:33:21 engine.py:136]     return self.engine.step()
ERROR 02-26 23:33:21 engine.py:136]   File "/root/akarx/vllm-fork/vllm/engine/llm_engine.py", line 1359, in step
ERROR 02-26 23:33:21 engine.py:136]     outputs = self.model_executor.execute_model(
ERROR 02-26 23:33:21 engine.py:136]   File "/root/akarx/vllm-fork/vllm/executor/executor_base.py", line 113, in execute_model
ERROR 02-26 23:33:21 engine.py:136]     output = self.collective_rpc("execute_model",
ERROR 02-26 23:33:21 engine.py:136]   File "/root/akarx/vllm-fork/vllm/executor/uniproc_executor.py", line 51, in collective_rpc
ERROR 02-26 23:33:21 engine.py:136]     answer = run_method(self.driver_worker, method, args, kwargs)
ERROR 02-26 23:33:21 engine.py:136]   File "/root/akarx/vllm-fork/vllm/utils.py", line 2288, in run_method
ERROR 02-26 23:33:21 engine.py:136]     return func(*args, **kwargs)
ERROR 02-26 23:33:21 engine.py:136]   File "/root/akarx/vllm-fork/vllm/worker/hpu_worker.py", line 281, in execute_model
ERROR 02-26 23:33:21 engine.py:136]     output = LocalOrDistributedWorkerBase.execute_model(
ERROR 02-26 23:33:21 engine.py:136]   File "/root/akarx/vllm-fork/vllm/worker/worker_base.py", line 402, in execute_model
ERROR 02-26 23:33:21 engine.py:136]     output = self.model_runner.execute_model(
ERROR 02-26 23:33:21 engine.py:136]   File "/usr/local/lib/python3.10/dist-packages/torch/utils/_contextlib.py", line 116, in decorate_context
ERROR 02-26 23:33:21 engine.py:136]     return func(*args, **kwargs)
ERROR 02-26 23:33:21 engine.py:136]   File "/root/akarx/vllm-fork/vllm/worker/hpu_model_runner.py", line 2285, in execute_model
ERROR 02-26 23:33:21 engine.py:136]     hidden_states = self.model.forward(
ERROR 02-26 23:33:21 engine.py:136]   File "/usr/local/lib/python3.10/dist-packages/habana_frameworks/torch/hpu/graphs.py", line 736, in forward
ERROR 02-26 23:33:21 engine.py:136]     return wrapped_hpugraph_forward(
ERROR 02-26 23:33:21 engine.py:136]   File "/usr/local/lib/python3.10/dist-packages/habana_frameworks/torch/hpu/graphs.py", line 609, in wrapped_hpugraph_forward
ERROR 02-26 23:33:21 engine.py:136]     outputs = orig_fwd(*args, **kwargs)
ERROR 02-26 23:33:21 engine.py:136]   File "/root/akarx/vllm-fork/vllm/worker/hpu_model_runner.py", line 409, in forward
ERROR 02-26 23:33:21 engine.py:136]     hidden_states = self.model(*args, **kwargs)
ERROR 02-26 23:33:21 engine.py:136]   File "/usr/local/lib/python3.10/dist-packages/torch/nn/modules/module.py", line 1739, in _wrapped_call_impl
ERROR 02-26 23:33:21 engine.py:136]     return self._call_impl(*args, **kwargs)
ERROR 02-26 23:33:21 engine.py:136]   File "/usr/local/lib/python3.10/dist-packages/torch/nn/modules/module.py", line 1750, in _call_impl
ERROR 02-26 23:33:21 engine.py:136]     return forward_call(*args, **kwargs)
ERROR 02-26 23:33:21 engine.py:136]   File "/root/akarx/vllm-fork/vllm/model_executor/models/llava_next.py", line 557, in forward
ERROR 02-26 23:33:21 engine.py:136]     inputs_embeds = self.get_input_embeddings(input_ids,
ERROR 02-26 23:33:21 engine.py:136]   File "/root/akarx/vllm-fork/vllm/model_executor/models/llava_next.py", line 488, in get_input_embeddings
ERROR 02-26 23:33:21 engine.py:136]     inputs_embeds = embed_multimodal(
ERROR 02-26 23:33:21 engine.py:136]   File "/root/akarx/vllm-fork/vllm/model_executor/models/utils.py", line 406, in embed_multimodal
ERROR 02-26 23:33:21 engine.py:136]     return _merge_multimodal_embeddings(
ERROR 02-26 23:33:21 engine.py:136]   File "/root/akarx/vllm-fork/vllm/model_executor/models/utils.py", line 369, in _merge_multimodal_embeddings
ERROR 02-26 23:33:21 engine.py:136]     raise ValueError(
ERROR 02-26 23:33:21 engine.py:136] ValueError: Attempted to assign 2160 + 2144 = 4304 multimodal tokens to 0 placeholders
CRITICAL 02-26 23:33:21 launcher.py:99] MQLLMEngine is already dead, terminating server process
INFO:     127.0.0.1:53166 - "POST /v1/chat/completions HTTP/1.1" 500 Internal Server Error
CRITICAL 02-26 23:33:21 launcher.py:99] MQLLMEngine is already dead, terminating server process
INFO:     127.0.0.1:53176 - "POST /v1/chat/completions HTTP/1.1" 500 Internal Server Error

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Assignees
No one assigned
Labels
bug Something isn't working
Projects
None yet
Milestone
No milestone
Development

No branches or pull requests
2 participants
@michalkuligowski @akarX23