search

vllm-project / vllm

A high-throughput and memory-efficient inference and serving engine for LLMs
https://docs.vllm.ai
Apache License 2.0
27.64k stars 4.08k forks source link

gfx908 architecture not working for version 0.5.1 #6375

Closed erkintelnyx closed 2 months ago

erkintelnyx commented 2 months ago

Your current environment

PyTorch version: 2.4.0.dev20240612+rocm6.1
Is debug build: False
CUDA used to build PyTorch: N/A
ROCM used to build PyTorch: 6.1.40091-a8dbc0c19

OS: Ubuntu 20.04.6 LTS (x86_64)
GCC version: (Ubuntu 9.4.0-1ubuntu1~20.04.2) 9.4.0
Clang version: 17.0.0 (https://github.com/RadeonOpenCompute/llvm-project roc-6.1.2 24193 669db884972e769450470020c06a6f132a8a065b)
CMake version: version 3.30.0
Libc version: glibc-2.31

Python version: 3.9.19 (main, May  6 2024, 19:43:03)  [GCC 11.2.0] (64-bit runtime)
Python platform: Linux-5.15.0-105-generic-x86_64-with-glibc2.31
Is CUDA available: True
CUDA runtime version: Could not collect
CUDA_MODULE_LOADING set to: LAZY
GPU models and configuration: AMD Instinct MI100 (gfx908:sramecc+:xnack-)
Nvidia driver version: Could not collect
cuDNN version: Could not collect
HIP runtime version: 6.1.40093
MIOpen runtime version: 3.1.0
Is XNNPACK available: True

CPU:
Architecture:                       x86_64
CPU op-mode(s):                     32-bit, 64-bit
Byte Order:                         Little Endian
Address sizes:                      48 bits physical, 48 bits virtual
CPU(s):                             16
On-line CPU(s) list:                0-15
Thread(s) per core:                 1
Core(s) per socket:                 16
Socket(s):                          1
NUMA node(s):                       1
Vendor ID:                          AuthenticAMD
CPU family:                         25
Model:                              1
Model name:                         AMD EPYC 7313 16-Core Processor
Stepping:                           1
CPU MHz:                            2999.998
BogoMIPS:                           5999.99
Virtualization:                     AMD-V
Hypervisor vendor:                  KVM
Virtualization type:                full
L1d cache:                          1 MiB
L1i cache:                          1 MiB
L2 cache:                           8 MiB
L3 cache:                           16 MiB
NUMA node0 CPU(s):                  0-15
Vulnerability Gather data sampling: Not affected
Vulnerability Itlb multihit:        Not affected
Vulnerability L1tf:                 Not affected
Vulnerability Mds:                  Not affected
Vulnerability Meltdown:             Not affected
Vulnerability Mmio stale data:      Not affected
Vulnerability Retbleed:             Not affected
Vulnerability Spec rstack overflow: Mitigation; safe RET, no microcode
Vulnerability Spec store bypass:    Mitigation; Speculative Store Bypass disabled via prctl and seccomp
Vulnerability Spectre v1:           Mitigation; usercopy/swapgs barriers and __user pointer sanitization
Vulnerability Spectre v2:           Mitigation; Retpolines, IBPB conditional, IBRS_FW, STIBP disabled, RSB filling, PBRSB-eIBRS Not affected
Vulnerability Srbds:                Not affected
Vulnerability Tsx async abort:      Not affected
Flags:                              fpu vme de pse tsc msr pae mce cx8 apic sep mtrr pge mca cmov pat pse36 clflush mmx fxsr sse sse2 ht syscall nx mmxext fxsr_opt pdpe1gb rdtscp lm rep_good nopl cpuid extd_apicid tsc_known_freq pni pclmulqdq ssse3 fma cx16 pcid sse4_1 sse4_2 x2apic movbe popcnt tsc_deadline_timer aes xsave avx f16c rdrand hypervisor lahf_lm cmp_legacy svm cr8_legacy abm sse4a misalignsse 3dnowprefetch osvw perfctr_core invpcid_single ssbd ibrs ibpb stibp vmmcall fsgsbase tsc_adjust bmi1 avx2 smep bmi2 invpcid rdseed adx smap clflushopt clwb sha_ni xsaveopt xsavec xgetbv1 xsaves clzero xsaveerptr wbnoinvd arat npt nrip_save umip pku ospke vaes vpclmulqdq rdpid arch_capabilities

Versions of relevant libraries:
[pip3] mypy==1.7.0
[pip3] mypy-extensions==1.0.0
[pip3] numpy==1.26.4
[pip3] optree==0.9.1
[pip3] pytorch-triton-rocm==3.0.0+01cbe5045a
[pip3] torch==2.4.0.dev20240612+rocm6.1
[pip3] torchaudio==2.4.0.dev20240612+rocm6.1
[pip3] torchvision==0.19.0.dev20240612+rocm6.1
[pip3] transformers==4.42.3
[pip3] triton==3.0.0
[conda] No relevant packages
ROCM Version: 6.1.40093-bd86f1708
Neuron SDK Version: N/A
vLLM Version: 0.5.1
vLLM Build Flags:
CUDA Archs: Not Set; ROCm: Disabled; Neuron: Disabled
GPU Topology:
Could not collect

🐛 Describe the bug

Using gfx908 GPU, we recently updated the version to 0.5.1, but came across the issue:

# python -m vllm.entrypoints.openai.api_server --download-dir /models --model meta-llama/Meta-Llama-3-8B-Instruct
INFO 07-12 14:27:28 api_server.py:206] vLLM API server version 0.5.1
INFO 07-12 14:27:28 api_server.py:207] args: Namespace(host=None, port=8000, uvicorn_log_level='info', allow_credentials=False, allowed_origins=['*'], allowed_methods=['*'], allowed_headers=['*'], api_key=None, lora_modules=None, chat_template=None, response_role='assistant', ssl_keyfile=None, ssl_certfile=None, ssl_ca_certs=None, ssl_cert_reqs=0, root_path=None, middleware=[], model='meta-llama/Meta-Llama-3-8B-Instruct', tokenizer=None, skip_tokenizer_init=False, revision=None, code_revision=None, tokenizer_revision=None, tokenizer_mode='auto', trust_remote_code=False, download_dir='/models', load_format='auto', dtype='auto', kv_cache_dtype='auto', quantization_param_path=None, max_model_len=None, guided_decoding_backend='outlines', distributed_executor_backend=None, worker_use_ray=False, pipeline_parallel_size=1, tensor_parallel_size=1, max_parallel_loading_workers=None, ray_workers_use_nsight=False, block_size=16, enable_prefix_caching=False, disable_sliding_window=False, use_v2_block_manager=False, num_lookahead_slots=0, seed=0, swap_space=4, gpu_memory_utilization=0.9, num_gpu_blocks_override=None, max_num_batched_tokens=None, max_num_seqs=256, max_logprobs=20, disable_log_stats=False, quantization=None, rope_scaling=None, rope_theta=None, enforce_eager=False, max_context_len_to_capture=None, max_seq_len_to_capture=8192, disable_custom_all_reduce=False, tokenizer_pool_size=0, tokenizer_pool_type='ray', tokenizer_pool_extra_config=None, enable_lora=False, max_loras=1, max_lora_rank=16, lora_extra_vocab_size=256, lora_dtype='auto', long_lora_scaling_factors=None, max_cpu_loras=None, fully_sharded_loras=False, device='auto', scheduler_delay_factor=0.0, enable_chunked_prefill=False, speculative_model=None, num_speculative_tokens=None, speculative_draft_tensor_parallel_size=None, speculative_max_model_len=None, speculative_disable_by_batch_size=None, ngram_prompt_lookup_max=None, ngram_prompt_lookup_min=None, spec_decoding_acceptance_method='rejection_sampler', typical_acceptance_sampler_posterior_threshold=None, typical_acceptance_sampler_posterior_alpha=None, model_loader_extra_config=None, preemption_mode=None, served_model_name=None, qlora_adapter_name_or_path=None, otlp_traces_endpoint=None, engine_use_ray=False, disable_log_requests=False, max_log_len=None)
INFO 07-12 14:27:28 config.py:724] Disabled the custom all-reduce kernel because it is not supported on AMD GPUs.
INFO 07-12 14:27:28 llm_engine.py:169] Initializing an LLM engine (v0.5.1) with config: model='meta-llama/Meta-Llama-3-8B-Instruct', speculative_config=None, tokenizer='meta-llama/Meta-Llama-3-8B-Instruct', skip_tokenizer_init=False, tokenizer_mode=auto, revision=None, rope_scaling=None, rope_theta=None, tokenizer_revision=None, trust_remote_code=False, dtype=torch.bfloat16, max_seq_len=8192, download_dir='/models', load_format=LoadFormat.AUTO, tensor_parallel_size=1, pipeline_parallel_size=1, disable_custom_all_reduce=True, quantization=None, enforce_eager=False, kv_cache_dtype=auto, quantization_param_path=None, device_config=cuda, decoding_config=DecodingConfig(guided_decoding_backend='outlines'), observability_config=ObservabilityConfig(otlp_traces_endpoint=None), seed=0, served_model_name=meta-llama/Meta-Llama-3-8B-Instruct, use_v2_block_manager=False, enable_prefix_caching=False)
Special tokens have been added in the vocabulary, make sure the associated word embeddings are fine-tuned or trained.
INFO 07-12 14:27:30 selector.py:58] Using ROCmFlashAttention backend.
INFO 07-12 14:27:31 selector.py:58] Using ROCmFlashAttention backend.
INFO 07-12 14:27:31 weight_utils.py:218] Using model weights format ['*.safetensors']
INFO 07-12 14:27:42 model_runner.py:255] Loading model weights took 14.9595 GB

rocblaslt warning: No paths matched /opt/conda/envs/py_3.9/lib/python3.9/site-packages/torch/lib/hipblaslt/library/*gfx908*co. Make sure that HIPBLASLT_TENSILE_LIBPATH is set correctly.
[rank0]: Traceback (most recent call last):
[rank0]:   File "/opt/conda/envs/py_3.9/lib/python3.9/runpy.py", line 197, in _run_module_as_main
[rank0]:     return _run_code(code, main_globals, None,
[rank0]:   File "/opt/conda/envs/py_3.9/lib/python3.9/runpy.py", line 87, in _run_code
[rank0]:     exec(code, run_globals)
[rank0]:   File "/vllm-workspace/vllm/entrypoints/openai/api_server.py", line 216, in <module>
[rank0]:     engine = AsyncLLMEngine.from_engine_args(
[rank0]:   File "/vllm-workspace/vllm/engine/async_llm_engine.py", line 431, in from_engine_args
[rank0]:     engine = cls(
[rank0]:   File "/vllm-workspace/vllm/engine/async_llm_engine.py", line 360, in __init__
[rank0]:     self.engine = self._init_engine(*args, **kwargs)
[rank0]:   File "/vllm-workspace/vllm/engine/async_llm_engine.py", line 507, in _init_engine
[rank0]:     return engine_class(*args, **kwargs)
[rank0]:   File "/vllm-workspace/vllm/engine/llm_engine.py", line 256, in __init__
[rank0]:     self._initialize_kv_caches()
[rank0]:   File "/vllm-workspace/vllm/engine/llm_engine.py", line 353, in _initialize_kv_caches
[rank0]:     self.model_executor.determine_num_available_blocks())
[rank0]:   File "/vllm-workspace/vllm/executor/gpu_executor.py", line 76, in determine_num_available_blocks
[rank0]:     return self.driver_worker.determine_num_available_blocks()
[rank0]:   File "/opt/conda/envs/py_3.9/lib/python3.9/site-packages/torch/utils/_contextlib.py", line 116, in decorate_context
[rank0]:     return func(*args, **kwargs)
[rank0]:   File "/vllm-workspace/vllm/worker/worker.py", line 173, in determine_num_available_blocks
[rank0]:     self.model_runner.profile_run()
[rank0]:   File "/opt/conda/envs/py_3.9/lib/python3.9/site-packages/torch/utils/_contextlib.py", line 116, in decorate_context
[rank0]:     return func(*args, **kwargs)
[rank0]:   File "/vllm-workspace/vllm/worker/model_runner.py", line 874, in profile_run
[rank0]:     self.execute_model(model_input, kv_caches, intermediate_tensors)
[rank0]:   File "/opt/conda/envs/py_3.9/lib/python3.9/site-packages/torch/utils/_contextlib.py", line 116, in decorate_context
[rank0]:     return func(*args, **kwargs)
[rank0]:   File "/vllm-workspace/vllm/worker/model_runner.py", line 1243, in execute_model
[rank0]:     hidden_or_intermediate_states = model_executable(
[rank0]:   File "/opt/conda/envs/py_3.9/lib/python3.9/site-packages/torch/nn/modules/module.py", line 1553, in _wrapped_call_impl
[rank0]:     return self._call_impl(*args, **kwargs)
[rank0]:   File "/opt/conda/envs/py_3.9/lib/python3.9/site-packages/torch/nn/modules/module.py", line 1562, in _call_impl
[rank0]:     return forward_call(*args, **kwargs)
[rank0]:   File "/vllm-workspace/vllm/model_executor/models/llama.py", line 401, in forward
[rank0]:     model_output = self.model(input_ids, positions, kv_caches,
[rank0]:   File "/opt/conda/envs/py_3.9/lib/python3.9/site-packages/torch/nn/modules/module.py", line 1553, in _wrapped_call_impl
[rank0]:     return self._call_impl(*args, **kwargs)
[rank0]:   File "/opt/conda/envs/py_3.9/lib/python3.9/site-packages/torch/nn/modules/module.py", line 1562, in _call_impl
[rank0]:     return forward_call(*args, **kwargs)
[rank0]:   File "/vllm-workspace/vllm/model_executor/models/llama.py", line 306, in forward
[rank0]:     hidden_states, residual = layer(
[rank0]:   File "/opt/conda/envs/py_3.9/lib/python3.9/site-packages/torch/nn/modules/module.py", line 1553, in _wrapped_call_impl
[rank0]:     return self._call_impl(*args, **kwargs)
[rank0]:   File "/opt/conda/envs/py_3.9/lib/python3.9/site-packages/torch/nn/modules/module.py", line 1562, in _call_impl
[rank0]:     return forward_call(*args, **kwargs)
[rank0]:   File "/vllm-workspace/vllm/model_executor/models/llama.py", line 230, in forward
[rank0]:     hidden_states = self.self_attn(
[rank0]:   File "/opt/conda/envs/py_3.9/lib/python3.9/site-packages/torch/nn/modules/module.py", line 1553, in _wrapped_call_impl
[rank0]:     return self._call_impl(*args, **kwargs)
[rank0]:   File "/opt/conda/envs/py_3.9/lib/python3.9/site-packages/torch/nn/modules/module.py", line 1562, in _call_impl
[rank0]:     return forward_call(*args, **kwargs)
[rank0]:   File "/vllm-workspace/vllm/model_executor/models/llama.py", line 161, in forward
[rank0]:     qkv, _ = self.qkv_proj(hidden_states)
[rank0]:   File "/opt/conda/envs/py_3.9/lib/python3.9/site-packages/torch/nn/modules/module.py", line 1553, in _wrapped_call_impl
[rank0]:     return self._call_impl(*args, **kwargs)
[rank0]:   File "/opt/conda/envs/py_3.9/lib/python3.9/site-packages/torch/nn/modules/module.py", line 1562, in _call_impl
[rank0]:     return forward_call(*args, **kwargs)
[rank0]:   File "/vllm-workspace/vllm/model_executor/layers/linear.py", line 317, in forward
[rank0]:     output_parallel = self.quant_method.apply(self, input_, bias)
[rank0]:   File "/vllm-workspace/vllm/model_executor/layers/linear.py", line 134, in apply
[rank0]:     return F.linear(x, weight, bias)
[rank0]: RuntimeError: CUDA error: HIPBLAS_STATUS_NOT_SUPPORTED when calling `HIPBLAS_STATUS_NOT_SUPPORTED`
hongxiayang commented 2 months ago

@erkintelnyx Can you try now with current main branch as the torch now used is 2.5 which might fix your issue: see this PR https://github.com/vllm-project/vllm/pull/6352

hongxiayang commented 2 months ago

also, can someone add label "rocm" to this issue so that it is easier for us to track?

erkintelnyx commented 2 months ago

@erkintelnyx Can you try now with current main branch as the torch now used is 2.5 which might fix your issue: see this PR #6352

Working with the main branch as you suggested.