I have a 16GB GPU which is enough to handle 2 instances of 8B models using vLLM. But when I try to do so, even though there is a lot of VRAM left (according to nvidia-smi), the second container fails to start with a cuda error. Can anyone tell if it’s possible and if yes, how?
Docker Command -
docker run -d --name vllmeta --runtime=nvidia --gpus all \
-v ~/.cache/huggingface:/root/.cache/huggingface \
–env “HUGGING_FACE_HUB_TOKEN=” \
–env “VLLM_SERVER_DEV_MODE=1” \
-p 8000:8000 \
–ipc=host \
vllm/vllm-openai:latest \
–model deepseek-ai/DeepSeek-R1-Distill-Qwen-7B\
–gpu-memory-utilization 0.5 \
–quantization bitsandbytes \
–dtype float16 \
–enforce-eager \
–max-model-len 2048
Mon May 12 07:58:02 2025
+-----------------------------------------------------------------------------------------+
| NVIDIA-SMI 570.133.20 Driver Version: 570.133.20 CUDA Version: 12.8 |
|-----------------------------------------+------------------------+----------------------+
| GPU Name Persistence-M | Bus-Id Disp.A | Volatile Uncorr. ECC |
| Fan Temp Perf Pwr:Usage/Cap | Memory-Usage | GPU-Util Compute M. |
| | | MIG M. |
|========================================+========================+======================|
| 0 Tesla T4 Off | 00000000:00:04.0 Off | 0 |
| N/A 78C P0 33W / 70W | 6631MiB / 15360MiB | 0% Default |
| | | N/A |
+-----------------------------------------+------------------------+----------------------+
+-----------------------------------------------------------------------------------------+
| Processes: |
| GPU GI CI PID Type Process name GPU Memory |
| ID ID Usage |
|=========================================================================================|
| 0 N/A N/A 329374 C /usr/bin/python3 6620MiB |
+-----------------------------------------------------------------------------------------+
The error that I get after starting the second container.
INFO 05-12 00:40:44 [__init__.py:239] Automatically detected platform cuda.
INFO 05-12 00:40:47 [api_server.py:1043] vLLM API server version 0.8.5.post1
INFO 05-12 00:40:47 [api_server.py:1044] args: Namespace(host=None, port=8000, uvicorn_log_level='info', disable_uvicorn_access_log=False, allow_credentials=False, allowed_origins=['*'], allowed_methods=['*'], allowed_headers=['*'], api_key=None, lora_modules=None, prompt_adapters=None, chat_template=None, chat_template_content_format='auto', response_role='assistant', ssl_keyfile=None, ssl_certfile=None, ssl_ca_certs=None, enable_ssl_refresh=False, ssl_cert_reqs=0, root_path=None, middleware=[], return_tokens_as_token_ids=False, disable_frontend_multiprocessing=False, enable_request_id_headers=False, enable_auto_tool_choice=False, tool_call_parser=None, tool_parser_plugin='', model='deepseek-ai/DeepSeek-R1-Distill-Qwen-7B', task='auto', tokenizer=None, hf_config_path=None, skip_tokenizer_init=False, revision=None, code_revision=None, tokenizer_revision=None, tokenizer_mode='auto', trust_remote_code=False, allowed_local_media_path=None, load_format='auto', download_dir=None, model_loader_extra_config={}, use_tqdm_on_load=True, config_format=<ConfigFormat.AUTO: 'auto'>, dtype='float16', max_model_len=2048, guided_decoding_backend='auto', reasoning_parser=None, logits_processor_pattern=None, model_impl='auto', distributed_executor_backend=None, pipeline_parallel_size=1, tensor_parallel_size=1, data_parallel_size=1, enable_expert_parallel=False, max_parallel_loading_workers=None, ray_workers_use_nsight=False, disable_custom_all_reduce=False, block_size=None, gpu_memory_utilization=0.5, swap_space=4, kv_cache_dtype='auto', num_gpu_blocks_override=None, enable_prefix_caching=None, prefix_caching_hash_algo='builtin', cpu_offload_gb=0, calculate_kv_scales=False, disable_sliding_window=False, use_v2_block_manager=True, seed=None, max_logprobs=20, disable_log_stats=False, quantization='bitsandbytes', rope_scaling=None, rope_theta=None, hf_token=None, hf_overrides=None, enforce_eager=True, max_seq_len_to_capture=8192, tokenizer_pool_size=0, tokenizer_pool_type='ray', tokenizer_pool_extra_config={}, limit_mm_per_prompt={}, mm_processor_kwargs=None, disable_mm_preprocessor_cache=False, enable_lora=None, enable_lora_bias=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, enable_prompt_adapter=None, max_prompt_adapters=1, max_prompt_adapter_token=0, device='auto', speculative_config=None, ignore_patterns=[], served_model_name=None, qlora_adapter_name_or_path=None, show_hidden_metrics_for_version=None, otlp_traces_endpoint=None, collect_detailed_traces=None, disable_async_output_proc=False, max_num_batched_tokens=None, max_num_seqs=None, max_num_partial_prefills=1, max_long_partial_prefills=1, long_prefill_token_threshold=0, num_lookahead_slots=0, scheduler_delay_factor=0.0, preemption_mode=None, num_scheduler_steps=1, multi_step_stream_outputs=True, scheduling_policy='fcfs', enable_chunked_prefill=None, disable_chunked_mm_input=False, scheduler_cls='vllm.core.scheduler.Scheduler', override_neuron_config=None, override_pooler_config=None, compilation_config=None, kv_transfer_config=None, worker_cls='auto', worker_extension_cls='', generation_config='auto', override_generation_config=None, enable_sleep_mode=False, additional_config=None, enable_reasoning=False, disable_cascade_attn=False, disable_log_requests=False, max_log_len=None, disable_fastapi_docs=False, enable_prompt_tokens_details=False, enable_server_load_tracking=False)
WARNING 05-12 00:40:48 [config.py:2972] Casting torch.bfloat16 to torch.float16.
INFO 05-12 00:40:57 [config.py:717] This model supports multiple tasks: {'reward', 'generate', 'score', 'embed', 'classify'}. Defaulting to 'generate'.
WARNING 05-12 00:40:57 [config.py:830] bitsandbytes quantization is not fully optimized yet. The speed can be slower than non-quantized models.
WARNING 05-12 00:40:57 [arg_utils.py:1658] Compute Capability < 8.0 is not supported by the V1 Engine. Falling back to V0.
WARNING 05-12 00:40:57 [cuda.py:93] To see benefits of async output processing, enable CUDA graph. Since, enforce-eager is enabled, async output processor cannot be used
INFO 05-12 00:40:58 [api_server.py:246] Started engine process with PID 48
INFO 05-12 00:41:02 [__init__.py:239] Automatically detected platform cuda.
INFO 05-12 00:41:04 [llm_engine.py:240] Initializing a V0 LLM engine (v0.8.5.post1) with config: model='deepseek-ai/DeepSeek-R1-Distill-Qwen-7B', speculative_config=None, tokenizer='deepseek-ai/DeepSeek-R1-Distill-Qwen-7B', skip_tokenizer_init=False, tokenizer_mode=auto, revision=None, override_neuron_config=None, tokenizer_revision=None, trust_remote_code=False, dtype=torch.float16, max_seq_len=2048, download_dir=None, load_format=LoadFormat.BITSANDBYTES, tensor_parallel_size=1, pipeline_parallel_size=1, disable_custom_all_reduce=False, quantization=bitsandbytes, enforce_eager=True, kv_cache_dtype=auto, device_config=cuda, decoding_config=DecodingConfig(guided_decoding_backend='auto', reasoning_backend=None), observability_config=ObservabilityConfig(show_hidden_metrics=False, otlp_traces_endpoint=None, collect_model_forward_time=False, collect_model_execute_time=False), seed=None, served_model_name=deepseek-ai/DeepSeek-R1-Distill-Qwen-7B, num_scheduler_steps=1, multi_step_stream_outputs=True, enable_prefix_caching=None, chunked_prefill_enabled=False, use_async_output_proc=False, disable_mm_preprocessor_cache=False, mm_processor_kwargs=None, pooler_config=None, compilation_config={"splitting_ops":[],"compile_sizes":[],"cudagraph_capture_sizes":[],"max_capture_size":0}, use_cached_outputs=True,
INFO 05-12 00:41:06 [cuda.py:240] Cannot use FlashAttention-2 backend for Volta and Turing GPUs.
INFO 05-12 00:41:06 [cuda.py:289] Using XFormers backend.
INFO 05-12 00:41:07 [parallel_state.py:1004] rank 0 in world size 1 is assigned as DP rank 0, PP rank 0, TP rank 0
INFO 05-12 00:41:07 [model_runner.py:1108] Starting to load model deepseek-ai/DeepSeek-R1-Distill-Qwen-7B...
INFO 05-12 00:41:08 [loader.py:1187] Loading weights with BitsAndBytes quantization. May take a while ...
INFO 05-12 00:41:08 [weight_utils.py:265] Using model weights format ['*.safetensors']
Loading safetensors checkpoint shards: 0% Completed | 0/2 [00:00<?, ?it/s]
Loading safetensors checkpoint shards: 50% Completed | 1/2 [00:06<00:06, 6.23s/it]
Loading safetensors checkpoint shards: 100% Completed | 2/2 [00:08<00:00, 3.97s/it]
Loading safetensors checkpoint shards: 100% Completed | 2/2 [00:08<00:00, 4.31s/it]
INFO 05-12 00:41:18 [model_runner.py:1140] Model loading took 5.2273 GiB and 9.910612 seconds
INFO 05-12 00:41:30 [worker.py:287] Memory profiling takes 12.44 seconds
INFO 05-12 00:41:30 [worker.py:287] the current vLLM instance can use total_gpu_memory (14.56GiB) x gpu_memory_utilization (0.50) = 7.28GiB
INFO 05-12 00:41:30 [worker.py:287] model weights take 5.23GiB; non_torch_memory takes 0.05GiB; PyTorch activation peak memory takes 1.40GiB; the rest of the memory reserved for KV Cache is 0.61GiB.
INFO 05-12 00:41:30 [executor_base.py:112] # cuda blocks: 709, # CPU blocks: 4681
INFO 05-12 00:41:30 [executor_base.py:117] Maximum concurrency for 2048 tokens per request: 5.54x
ERROR 05-12 00:41:31 [engine.py:448] CUDA error: invalid argument
ERROR 05-12 00:41:31 [engine.py:448] CUDA kernel errors might be asynchronously reported at some other API call, so the stacktrace below might be incorrect.
ERROR 05-12 00:41:31 [engine.py:448] For debugging consider passing CUDA_LAUNCH_BLOCKING=1
ERROR 05-12 00:41:31 [engine.py:448] Compile with `TORCH_USE_CUDA_DSA` to enable device-side assertions.
ERROR 05-12 00:41:31 [engine.py:448] Traceback (most recent call last):
ERROR 05-12 00:41:31 [engine.py:448] File "/usr/local/lib/python3.12/dist-packages/vllm/engine/multiprocessing/engine.py", line 436, in run_mp_engine
ERROR 05-12 00:41:31 [engine.py:448] engine = MQLLMEngine.from_vllm_config(
ERROR 05-12 00:41:31 [engine.py:448] ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
ERROR 05-12 00:41:31 [engine.py:448] File "/usr/local/lib/python3.12/dist-packages/vllm/engine/multiprocessing/engine.py", line 128, in from_vllm_config
ERROR 05-12 00:41:31 [engine.py:448] return cls(
ERROR 05-12 00:41:31 [engine.py:448] ^^^^
ERROR 05-12 00:41:31 [engine.py:448] File "/usr/local/lib/python3.12/dist-packages/vllm/engine/multiprocessing/engine.py", line 82, in __init__
ERROR 05-12 00:41:31 [engine.py:448] self.engine = LLMEngine(*args, **kwargs)
ERROR 05-12 00:41:31 [engine.py:448] ^^^^^^^^^^^^^^^^^^^^^^^^^^
Process SpawnProcess-1:
ERROR 05-12 00:41:31 [engine.py:448] File "/usr/local/lib/python3.12/dist-packages/vllm/engine/llm_engine.py", line 278, in __init__
ERROR 05-12 00:41:31 [engine.py:448] self._initialize_kv_caches()
ERROR 05-12 00:41:31 [engine.py:448] File "/usr/local/lib/python3.12/dist-packages/vllm/engine/llm_engine.py", line 435, in _initialize_kv_caches
ERROR 05-12 00:41:31 [engine.py:448] self.model_executor.initialize_cache(num_gpu_blocks, num_cpu_blocks)
ERROR 05-12 00:41:31 [engine.py:448] File "/usr/local/lib/python3.12/dist-packages/vllm/executor/executor_base.py", line 123, in initialize_cache
ERROR 05-12 00:41:31 [engine.py:448] self.collective_rpc("initialize_cache",
ERROR 05-12 00:41:31 [engine.py:448] File "/usr/local/lib/python3.12/dist-packages/vllm/executor/uniproc_executor.py", line 56, in collective_rpc
ERROR 05-12 00:41:31 [engine.py:448] answer = run_method(self.driver_worker, method, args, kwargs)
ERROR 05-12 00:41:31 [engine.py:448] ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
ERROR 05-12 00:41:31 [engine.py:448] File "/usr/local/lib/python3.12/dist-packages/vllm/utils.py", line 2456, in run_method
ERROR 05-12 00:41:31 [engine.py:448] return func(*args, **kwargs)
ERROR 05-12 00:41:31 [engine.py:448] ^^^^^^^^^^^^^^^^^^^^^
ERROR 05-12 00:41:31 [engine.py:448] File "/usr/local/lib/python3.12/dist-packages/vllm/worker/worker.py", line 327, in initialize_cache
ERROR 05-12 00:41:31 [engine.py:448] self._init_cache_engine()
ERROR 05-12 00:41:31 [engine.py:448] File "/usr/local/lib/python3.12/dist-packages/vllm/worker/worker.py", line 333, in _init_cache_engine
ERROR 05-12 00:41:31 [engine.py:448] CacheEngine(self.cache_config, self.model_config,
ERROR 05-12 00:41:31 [engine.py:448] File "/usr/local/lib/python3.12/dist-packages/vllm/worker/cache_engine.py", line 66, in __init__
ERROR 05-12 00:41:31 [engine.py:448] self.cpu_cache = self._allocate_kv_cache(self.num_cpu_blocks, "cpu")
ERROR 05-12 00:41:31 [engine.py:448] ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
ERROR 05-12 00:41:31 [engine.py:448] File "/usr/local/lib/python3.12/dist-packages/vllm/worker/cache_engine.py", line 95, in _allocate_kv_cache
ERROR 05-12 00:41:31 [engine.py:448] layer_kv_cache = torch.zeros(
ERROR 05-12 00:41:31 [engine.py:448] ^^^^^^^^^^^^
ERROR 05-12 00:41:31 [engine.py:448] RuntimeError: CUDA error: invalid argument
ERROR 05-12 00:41:31 [engine.py:448] CUDA kernel errors might be asynchronously reported at some other API call, so the stacktrace below might be incorrect.
ERROR 05-12 00:41:31 [engine.py:448] For debugging consider passing CUDA_LAUNCH_BLOCKING=1
ERROR 05-12 00:41:31 [engine.py:448] Compile with `TORCH_USE_CUDA_DSA` to enable device-side assertions.
ERROR 05-12 00:41:31 [engine.py:448]
Traceback (most recent call last):
File "/usr/lib/python3.12/multiprocessing/process.py", line 314, in _bootstrap
self.run()
File "/usr/lib/python3.12/multiprocessing/process.py", line 108, in run
self._target(*self._args, **self._kwargs)
File "/usr/local/lib/python3.12/dist-packages/vllm/engine/multiprocessing/engine.py", line 450, in run_mp_engine
raise e
File "/usr/local/lib/python3.12/dist-packages/vllm/engine/multiprocessing/engine.py", line 436, in run_mp_engine
engine = MQLLMEngine.from_vllm_config(
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "/usr/local/lib/python3.12/dist-packages/vllm/engine/multiprocessing/engine.py", line 128, in from_vllm_config
return cls(
^^^^
File "/usr/local/lib/python3.12/dist-packages/vllm/engine/multiprocessing/engine.py", line 82, in __init__
self.engine = LLMEngine(*args, **kwargs)
^^^^^^^^^^^^^^^^^^^^^^^^^^
File "/usr/local/lib/python3.12/dist-packages/vllm/engine/llm_engine.py", line 278, in __init__
self._initialize_kv_caches()
File "/usr/local/lib/python3.12/dist-packages/vllm/engine/llm_engine.py", line 435, in _initialize_kv_caches
self.model_executor.initialize_cache(num_gpu_blocks, num_cpu_blocks)
File "/usr/local/lib/python3.12/dist-packages/vllm/executor/executor_base.py", line 123, in initialize_cache
self.collective_rpc("initialize_cache",
File "/usr/local/lib/python3.12/dist-packages/vllm/executor/uniproc_executor.py", line 56, in collective_rpc
answer = run_method(self.driver_worker, method, args, kwargs)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "/usr/local/lib/python3.12/dist-packages/vllm/utils.py", line 2456, in run_method
return func(*args, **kwargs)
^^^^^^^^^^^^^^^^^^^^^
File "/usr/local/lib/python3.12/dist-packages/vllm/worker/worker.py", line 327, in initialize_cache
self._init_cache_engine()
File "/usr/local/lib/python3.12/dist-packages/vllm/worker/worker.py", line 333, in _init_cache_engine
CacheEngine(self.cache_config, self.model_config,
File "/usr/local/lib/python3.12/dist-packages/vllm/worker/cache_engine.py", line 66, in __init__
self.cpu_cache = self._allocate_kv_cache(self.num_cpu_blocks, "cpu")
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "/usr/local/lib/python3.12/dist-packages/vllm/worker/cache_engine.py", line 95, in _allocate_kv_cache
layer_kv_cache = torch.zeros(
^^^^^^^^^^^^
RuntimeError: CUDA error: invalid argument
CUDA kernel errors might be asynchronously reported at some other API call, so the stacktrace below might be incorrect.
For debugging consider passing CUDA_LAUNCH_BLOCKING=1
Compile with `TORCH_USE_CUDA_DSA` to enable device-side assertions.
[rank0]:[W512 00:41:31.212053077 ProcessGroupNCCL.cpp:1496] Warning: WARNING: destroy_process_group() was not called before program exit, which can leak resources. For more info, please see https://pytorch.org/docs/stable/distributed.html#shutdown (function operator())
Traceback (most recent call last):
File "<frozen runpy>", line 198, in _run_module_as_main
File "<frozen runpy>", line 88, in _run_code
File "/usr/local/lib/python3.12/dist-packages/vllm/entrypoints/openai/api_server.py", line 1130, in <module>
uvloop.run(run_server(args))
File "/usr/local/lib/python3.12/dist-packages/uvloop/__init__.py", line 109, in run
return __asyncio.run(
^^^^^^^^^^^^^^
File "/usr/lib/python3.12/asyncio/runners.py", line 195, in run
return runner.run(main)
^^^^^^^^^^^^^^^^
File "/usr/lib/python3.12/asyncio/runners.py", line 118, in run
return self._loop.run_until_complete(task)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "uvloop/loop.pyx", line 1518, in uvloop.loop.Loop.run_until_complete
File "/usr/local/lib/python3.12/dist-packages/uvloop/__init__.py", line 61, in wrapper
return await main
^^^^^^^^^^
File "/usr/local/lib/python3.12/dist-packages/vllm/entrypoints/openai/api_server.py", line 1078, in run_server
async with build_async_engine_client(args) as engine_client:
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "/usr/lib/python3.12/contextlib.py", line 210, in __aenter__
return await anext(self.gen)
^^^^^^^^^^^^^^^^^^^^^
File "/usr/local/lib/python3.12/dist-packages/vllm/entrypoints/openai/api_server.py", line 146, in build_async_engine_client
async with build_async_engine_client_from_engine_args(
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "/usr/lib/python3.12/contextlib.py", line 210, in __aenter__
return await anext(self.gen)
^^^^^^^^^^^^^^^^^^^^^
File "/usr/local/lib/python3.12/dist-packages/vllm/entrypoints/openai/api_server.py", line 269, in build_async_engine_client_from_engine_args
raise RuntimeError(
RuntimeError: Engine process failed to start. See stack trace for the root cause.