search

vllm-project / vllm

A high-throughput and memory-efficient inference and serving engine for LLMs
https://docs.vllm.ai
Apache License 2.0
26.85k stars 3.94k forks source link

[Bug]: Failed to launch api_server with FP8D quantized gemma-2-27b-it on vllm 0.5.3post1 #6957

Closed bedovyy closed 1 month ago

bedovyy commented 1 month ago

Your current environment

Collecting environment information...
PyTorch version: 2.3.1+cu121
Is debug build: False
CUDA used to build PyTorch: 12.1
ROCM used to build PyTorch: N/A

OS: Ubuntu 22.04.4 LTS (x86_64)
GCC version: (Ubuntu 12.3.0-1ubuntu1~22.04) 12.3.0
Clang version: Could not collect
CMake version: version 3.30.1
Libc version: glibc-2.35

Python version: 3.10.12 (main, Mar 22 2024, 16:50:05) [GCC 11.4.0] (64-bit runtime)
Python platform: Linux-6.5.0-44-generic-x86_64-with-glibc2.35
Is CUDA available: True
CUDA runtime version: 12.1.105
CUDA_MODULE_LOADING set to: LAZY
GPU models and configuration:
GPU 0: NVIDIA GeForce RTX 3090
GPU 1: NVIDIA GeForce RTX 3090

Nvidia driver version: 550.90.07
cuDNN version: Could not collect
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:                      48 bits physical, 48 bits virtual
Byte Order:                         Little Endian
CPU(s):                             16
On-line CPU(s) list:                0-15
Vendor ID:                          AuthenticAMD
Model name:                         AMD Ryzen 7 5700X 8-Core Processor
CPU family:                         25
Model:                              33
Thread(s) per core:                 2
Core(s) per socket:                 8
Socket(s):                          1
Stepping:                           2
Frequency boost:                    enabled
CPU max MHz:                        4661.7178
CPU min MHz:                        2200.0000
BogoMIPS:                           6787.67
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 constant_tsc rep_good nopl nonstop_tsc cpuid extd_apicid aperfmperf rapl pni pclmulqdq monitor ssse3 fma cx16 sse4_1 sse4_2 x2apic movbe popcnt aes xsave avx f16c rdrand lahf_lm cmp_legacy svm extapic cr8_legacy abm sse4a misalignsse 3dnowprefetch osvw ibs skinit wdt tce topoext perfctr_core perfctr_nb bpext perfctr_llc mwaitx cpb cat_l3 cdp_l3 hw_pstate ssbd mba ibrs ibpb stibp vmmcall fsgsbase bmi1 avx2 smep bmi2 erms invpcid cqm rdt_a rdseed adx smap clflushopt clwb sha_ni xsaveopt xsavec xgetbv1 xsaves cqm_llc cqm_occup_llc cqm_mbm_total cqm_mbm_local clzero irperf xsaveerptr rdpru wbnoinvd arat npt lbrv svm_lock nrip_save tsc_scale vmcb_clean flushbyasid decodeassists pausefilter pfthreshold avic v_vmsave_vmload vgif v_spec_ctrl umip pku ospke vaes vpclmulqdq rdpid overflow_recov succor smca fsrm
Virtualization:                     AMD-V
L1d cache:                          256 KiB (8 instances)
L1i cache:                          256 KiB (8 instances)
L2 cache:                           4 MiB (8 instances)
L3 cache:                           32 MiB (1 instance)
NUMA node(s):                       1
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
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; Retpolines; IBPB conditional; IBRS_FW; STIBP always-on; RSB filling; PBRSB-eIBRS Not affected; BHI Not affected
Vulnerability Srbds:                Not affected
Vulnerability Tsx async abort:      Not affected

Versions of relevant libraries:
[pip3] flashinfer==0.1.0+cu121torch2.3
[pip3] numpy==1.26.4
[pip3] nvidia-nccl-cu12==2.20.5
[pip3] torch==2.3.1
[pip3] torchvision==0.18.1
[pip3] transformers==4.43.2
[pip3] triton==2.3.1
[pip3] vllm-nccl-cu12==2.18.1.0.4.0
[conda] Could not collect
ROCM Version: Could not collect
Neuron SDK Version: N/A
vLLM Version: 0.5.3.post1
vLLM Build Flags:
CUDA Archs: Not Set; ROCm: Disabled; Neuron: Disabled
GPU Topology:
GPU0    GPU1    CPU Affinity    NUMA Affinity   GPU NUMA ID
GPU0     X      PHB     0-15    0               N/A
GPU1    PHB      X      0-15    0               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

🐛 Describe the bug

I could not launch api_server with gemma-2-27b-it-FP8D on 0.5.3post1.

I could launch with,

Log is as below,

$  VLLM_LOGGING_LEVEL=DEBUG VLLM_ATTENTION_BACKEND=FLASHINFER python -m vllm.entrypoints.openai.api_server --gpu-memory-utilization 1.0 --model models/google_gemma-2-27b-it-FP8D --enforce-eager --disable-log-request -tp 2 --port 8000
INFO 07-31 04:04:06 api_server.py:219] vLLM API server version 0.5.3.post1
INFO 07-31 04:04:06 api_server.py:220] 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, prompt_adapters=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='models/google_gemma-2-27b-it-FP8D', tokenizer=None, skip_tokenizer_init=False, revision=None, code_revision=None, tokenizer_revision=None, tokenizer_mode='auto', trust_remote_code=False, download_dir=None, 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=2, 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, cpu_offload_gb=0, gpu_memory_utilization=1.0, 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=True, 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, enable_prompt_adapter=False, max_prompt_adapters=1, max_prompt_adapter_token=0, device='auto', scheduler_delay_factor=0.0, enable_chunked_prefill=None, 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, disable_logprobs_during_spec_decoding=None, model_loader_extra_config=None, ignore_patterns=[], preemption_mode=None, served_model_name=['[AI-45] models/google_gemma-2-27b-it-FP8D'], qlora_adapter_name_or_path=None, otlp_traces_endpoint=None, engine_use_ray=False, disable_log_requests=True, max_log_len=None)
WARNING 07-31 04:04:06 utils.py:569] Gemma 2 uses sliding window attention for every odd layer, which is currently not supported by vLLM. Disabling sliding window and capping the max length to the sliding window size (4096).
INFO 07-31 04:04:06 config.py:715] Defaulting to use mp for distributed inference
INFO 07-31 04:04:06 llm_engine.py:176] Initializing an LLM engine (v0.5.3.post1) with config: model='models/google_gemma-2-27b-it-FP8D', speculative_config=None, tokenizer='models/google_gemma-2-27b-it-FP8D', 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=4096, download_dir=None, load_format=LoadFormat.AUTO, tensor_parallel_size=2, pipeline_parallel_size=1, disable_custom_all_reduce=False, quantization=fp8, enforce_eager=True, 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=[AI-45] models/google_gemma-2-27b-it-FP8D, use_v2_block_manager=False, enable_prefix_caching=False)
INFO 07-31 04:04:07 custom_cache_manager.py:17] Setting Triton cache manager to: vllm.triton_utils.custom_cache_manager:CustomCacheManager
(VllmWorkerProcess pid=270647) INFO 07-31 04:04:07 selector.py:80] Using Flashinfer backend.
INFO 07-31 04:04:07 selector.py:80] Using Flashinfer backend.
(VllmWorkerProcess pid=270647) INFO 07-31 04:04:07 multiproc_worker_utils.py:215] Worker ready; awaiting tasks
DEBUG 07-31 04:04:08 parallel_state.py:803] world_size=2 rank=0 local_rank=0 distributed_init_method=tcp://127.0.0.1:33343 backend=nccl
(VllmWorkerProcess pid=270647) DEBUG 07-31 04:04:08 parallel_state.py:803] world_size=2 rank=1 local_rank=1 distributed_init_method=tcp://127.0.0.1:33343 backend=nccl
INFO 07-31 04:04:08 utils.py:784] Found nccl from library libnccl.so.2
(VllmWorkerProcess pid=270647) INFO 07-31 04:04:08 utils.py:784] Found nccl from library libnccl.so.2
INFO 07-31 04:04:08 pynccl.py:63] vLLM is using nccl==2.20.5
(VllmWorkerProcess pid=270647) INFO 07-31 04:04:08 pynccl.py:63] vLLM is using nccl==2.20.5
INFO 07-31 04:04:08 custom_all_reduce_utils.py:232] reading GPU P2P access cache from /cache/vllm/gpu_p2p_access_cache_for_0,1.json
(VllmWorkerProcess pid=270647) INFO 07-31 04:04:08 custom_all_reduce_utils.py:232] reading GPU P2P access cache from /cache/vllm/gpu_p2p_access_cache_for_0,1.json
INFO 07-31 04:04:08 shm_broadcast.py:241] vLLM message queue communication handle: Handle(connect_ip='127.0.0.1', local_reader_ranks=[1], buffer=<vllm.distributed.device_communicators.shm_broadcast.ShmRingBuffer object at 0x7c3b6cf06fe0>, local_subscribe_port=37909, local_sync_port=40267, remote_subscribe_port=None, remote_sync_port=None)
INFO 07-31 04:04:08 model_runner.py:680] Starting to load model models/google_gemma-2-27b-it-FP8D...
(VllmWorkerProcess pid=270647) INFO 07-31 04:04:08 model_runner.py:680] Starting to load model models/google_gemma-2-27b-it-FP8D...
WARNING 07-31 04:04:08 fp8.py:39] Detected fp8 checkpoint. Please note that the format is experimental and subject to change.
(VllmWorkerProcess pid=270647) WARNING 07-31 04:04:08 fp8.py:39] Detected fp8 checkpoint. Please note that the format is experimental and subject to change.
INFO 07-31 04:04:08 selector.py:80] Using Flashinfer backend.
(VllmWorkerProcess pid=270647) INFO 07-31 04:04:08 selector.py:80] Using Flashinfer backend.
Loading safetensors checkpoint shards:   0% Completed | 0/6 [00:00<?, ?it/s]
Loading safetensors checkpoint shards:  17% Completed | 1/6 [00:00<00:03,  1.51it/s]
Loading safetensors checkpoint shards:  33% Completed | 2/6 [00:01<00:03,  1.27it/s]
Loading safetensors checkpoint shards:  50% Completed | 3/6 [00:02<00:02,  1.31it/s]
Loading safetensors checkpoint shards:  67% Completed | 4/6 [00:03<00:01,  1.29it/s]
Loading safetensors checkpoint shards:  83% Completed | 5/6 [00:03<00:00,  1.24it/s]
Loading safetensors checkpoint shards: 100% Completed | 6/6 [00:04<00:00,  1.23it/s]
Loading safetensors checkpoint shards: 100% Completed | 6/6 [00:04<00:00,  1.27it/s]

[rank0]: Traceback (most recent call last):
[rank0]:   File "/usr/lib/python3.10/runpy.py", line 196, in _run_module_as_main
[rank0]:     return _run_code(code, main_globals, None,
[rank0]:   File "/usr/lib/python3.10/runpy.py", line 86, in _run_code
[rank0]:     exec(code, run_globals)
[rank0]:   File "/home/bedovyy/Projects/vllm/venv/lib/python3.10/site-packages/vllm/entrypoints/openai/api_server.py", line 317, in <module>
[rank0]:     run_server(args)
[rank0]:   File "/home/bedovyy/Projects/vllm/venv/lib/python3.10/site-packages/vllm/entrypoints/openai/api_server.py", line 231, in run_server
[rank0]:     if llm_engine is not None else AsyncLLMEngine.from_engine_args(
[rank0]:   File "/home/bedovyy/Projects/vllm/venv/lib/python3.10/site-packages/vllm/engine/async_llm_engine.py", line 466, in from_engine_args
[rank0]:     engine = cls(
[rank0]:   File "/home/bedovyy/Projects/vllm/venv/lib/python3.10/site-packages/vllm/engine/async_llm_engine.py", line 380, in __init__
[rank0]:     self.engine = self._init_engine(*args, **kwargs)
[rank0]:   File "/home/bedovyy/Projects/vllm/venv/lib/python3.10/site-packages/vllm/engine/async_llm_engine.py", line 547, in _init_engine
[rank0]:     return engine_class(*args, **kwargs)
[rank0]:   File "/home/bedovyy/Projects/vllm/venv/lib/python3.10/site-packages/vllm/engine/llm_engine.py", line 251, in __init__
[rank0]:     self.model_executor = executor_class(
[rank0]:   File "/home/bedovyy/Projects/vllm/venv/lib/python3.10/site-packages/vllm/executor/multiproc_gpu_executor.py", line 201, in __init__
[rank0]:     super().__init__(*args, **kwargs)
[rank0]:   File "/home/bedovyy/Projects/vllm/venv/lib/python3.10/site-packages/vllm/executor/distributed_gpu_executor.py", line 25, in __init__
[rank0]:     super().__init__(*args, **kwargs)
[rank0]:   File "/home/bedovyy/Projects/vllm/venv/lib/python3.10/site-packages/vllm/executor/executor_base.py", line 47, in __init__
[rank0]:     self._init_executor()
[rank0]:   File "/home/bedovyy/Projects/vllm/venv/lib/python3.10/site-packages/vllm/executor/multiproc_gpu_executor.py", line 124, in _init_executor
[rank0]:     self._run_workers("load_model",
[rank0]:   File "/home/bedovyy/Projects/vllm/venv/lib/python3.10/site-packages/vllm/executor/multiproc_gpu_executor.py", line 178, in _run_workers
[rank0]:     driver_worker_output = driver_worker_method(*args, **kwargs)
[rank0]:   File "/home/bedovyy/Projects/vllm/venv/lib/python3.10/site-packages/vllm/worker/worker.py", line 139, in load_model
[rank0]:     self.model_runner.load_model()
[rank0]:   File "/home/bedovyy/Projects/vllm/venv/lib/python3.10/site-packages/vllm/worker/model_runner.py", line 682, in load_model
[rank0]:     self.model = get_model(model_config=self.model_config,
[rank0]:   File "/home/bedovyy/Projects/vllm/venv/lib/python3.10/site-packages/vllm/model_executor/model_loader/__init__.py", line 21, in get_model
[rank0]:     return loader.load_model(model_config=model_config,
[rank0]:   File "/home/bedovyy/Projects/vllm/venv/lib/python3.10/site-packages/vllm/model_executor/model_loader/loader.py", line 283, in load_model
[rank0]:     model.load_weights(
[rank0]:   File "/home/bedovyy/Projects/vllm/venv/lib/python3.10/site-packages/vllm/model_executor/models/gemma2.py", line 393, in load_weights
[rank0]:     raise RuntimeError(
[rank0]: RuntimeError: Some weights are not initialized from checkpoints: {'model.layers.5.self_attn.attn.k_scale', 'model.layers.31.self_attn.attn.k_scale', 'model.layers.32.self_attn.attn.k_scale', 'model.layers.44.self_attn.attn.k_scale', 'model.layers.19.self_attn.attn.k_scale', 'model.layers.14.self_attn.attn.k_scale', 'model.layers.24.self_attn.attn.v_scale', 'model.layers.1.self_attn.attn.v_scale', 'model.layers.42.self_attn.attn.k_scale', 'model.layers.36.self_attn.attn.k_scale', 'model.layers.36.self_attn.attn.v_scale', 'model.layers.7.self_attn.attn.v_scale', 'model.layers.17.self_attn.attn.k_scale', 'model.layers.12.self_attn.attn.k_scale', 'model.layers.33.self_attn.attn.k_scale', 'model.layers.31.self_attn.attn.v_scale', 'model.layers.15.self_attn.attn.v_scale', 'model.layers.8.self_attn.attn.v_scale', 'model.layers.15.self_attn.attn.k_scale', 'model.layers.23.self_attn.attn.v_scale', 'model.layers.2.self_attn.attn.k_scale', 'model.layers.9.self_attn.attn.v_scale', 'model.layers.4.self_attn.attn.k_scale', 'model.layers.19.self_attn.attn.v_scale', 'model.layers.13.self_attn.attn.v_scale', 'model.layers.32.self_attn.attn.v_scale', 'model.layers.34.self_attn.attn.k_scale', 'model.layers.37.self_attn.attn.v_scale', 'model.layers.24.self_attn.attn.k_scale', 'model.layers.4.self_attn.attn.v_scale', 'model.layers.1.self_attn.attn.k_scale', 'model.layers.29.self_attn.attn.k_scale', 'model.layers.3.self_attn.attn.v_scale', 'model.layers.39.self_attn.attn.k_scale', 'model.layers.42.self_attn.attn.v_scale', 'model.layers.43.self_attn.attn.k_scale', 'model.layers.0.self_attn.attn.v_scale', 'model.layers.34.self_attn.attn.v_scale', 'model.layers.25.self_attn.attn.v_scale', 'model.layers.27.self_attn.attn.v_scale', 'model.layers.28.self_attn.attn.v_scale', 'model.layers.26.self_attn.attn.v_scale', 'model.layers.17.self_attn.attn.v_scale', 'model.layers.5.self_attn.attn.v_scale', 'model.layers.12.self_attn.attn.v_scale', 'model.layers.14.self_attn.attn.v_scale', 'model.layers.40.self_attn.attn.v_scale', 'model.layers.21.self_attn.attn.v_scale', 'model.layers.45.self_attn.attn.k_scale', 'model.layers.11.self_attn.attn.v_scale', 'model.layers.7.self_attn.attn.k_scale', 'model.layers.45.self_attn.attn.v_scale', 'model.layers.0.self_attn.attn.k_scale', 'model.layers.8.self_attn.attn.k_scale', 'model.layers.30.self_attn.attn.k_scale', 'model.layers.41.self_attn.attn.k_scale', 'model.layers.29.self_attn.attn.v_scale', 'model.layers.18.self_attn.attn.v_scale', 'model.layers.10.self_attn.attn.v_scale', 'model.layers.35.self_attn.attn.k_scale', 'model.layers.35.self_attn.attn.v_scale', 'model.layers.30.self_attn.attn.v_scale', 'model.layers.22.self_attn.attn.v_scale', 'model.layers.6.self_attn.attn.v_scale', 'model.layers.3.self_attn.attn.k_scale', 'model.layers.39.self_attn.attn.v_scale', 'model.layers.21.self_attn.attn.k_scale', 'model.layers.23.self_attn.attn.k_scale', 'model.layers.22.self_attn.attn.k_scale', 'model.layers.16.self_attn.attn.v_scale', 'model.layers.9.self_attn.attn.k_scale', 'model.layers.37.self_attn.attn.k_scale', 'model.layers.20.self_attn.attn.k_scale', 'model.layers.6.self_attn.attn.k_scale', 'model.layers.33.self_attn.attn.v_scale', 'model.layers.20.self_attn.attn.v_scale', 'model.layers.11.self_attn.attn.k_scale', 'model.layers.16.self_attn.attn.k_scale', 'model.layers.40.self_attn.attn.k_scale', 'model.layers.41.self_attn.attn.v_scale', 'model.layers.18.self_attn.attn.k_scale', 'model.layers.38.self_attn.attn.k_scale', 'model.layers.27.self_attn.attn.k_scale', 'model.layers.28.self_attn.attn.k_scale', 'model.layers.2.self_attn.attn.v_scale', 'model.layers.44.self_attn.attn.v_scale', 'model.layers.38.self_attn.attn.v_scale', 'model.layers.26.self_attn.attn.k_scale', 'model.layers.13.self_attn.attn.k_scale', 'model.layers.25.self_attn.attn.k_scale', 'model.layers.10.self_attn.attn.k_scale', 'model.layers.43.self_attn.attn.v_scale'}
[rank0]:[W CudaIPCTypes.cpp:16] Producer process has been terminated before all shared CUDA tensors released. See Note [Sharing CUDA tensors]
/usr/lib/python3.10/multiprocessing/resource_tracker.py:224: UserWarning: resource_tracker: There appear to be 1 leaked shared_memory objects to clean up at shutdown
  warnings.warn('resource_tracker: There appear to be %d '

I have quanitized FP8 dynamic using the below code.

from datasets import load_dataset
from transformers import AutoTokenizer
from auto_fp8 import AutoFP8ForCausalLM, BaseQuantizeConfig
import sys

if len(sys.argv) != 3:
    print(f"usage: python3 {sys.argv[0]} <base_model> <output_path>")
    exit(1)

pretrained_model_dir = sys.argv[1]
quantized_model_dir = sys.argv[2]

# Define quantization config with static activation scales
quantize_config = BaseQuantizeConfig(quant_method="fp8", activation_scheme="dynamic")
# For dynamic activation scales, there is no need for calbration examples
examples = []

# Load the model, quantize, and save checkpoint
model = AutoFP8ForCausalLM.from_pretrained(pretrained_model_dir, quantize_config, device_map="cpu")
model.quantize(examples)
model.save_quantized(quantized_model_dir)

you may also reproduce using nm-testing/gemma-2-27b-it-FP8 on huggingface.

orellavie1212 commented 1 month ago

So vllm 0 5.2 works, but 0.5.3p1 doesn't?

robertgshaw2-neuralmagic commented 1 month ago

@mgoin

mgoin commented 1 month ago

I see, thanks for reporting. This seems to be happening due to Gemma's more strict model loading in vLLM https://github.com/vllm-project/vllm/blob/d7a299edaa5d23f3d7d5c98b53872a8ced9aad80/vllm/model_executor/models/gemma.py#L405-L409 I will work on a fix.

linpan commented 1 month ago

gptq_marlin not found. image

mgoin commented 1 month ago

@linpan Please open a separate issue as this is unrelated. It seems you already have a gptq model, so you should not specify any --quantization tag. vLLM will automatically convert it to gptq_marlin if it is able to

bedovyy commented 1 month ago

@mgoin I could load gemma-2-27b FP8 quants successfully on latest main branch. The response seems corrupted tho but it may be different issue (because GPTQ quants of gemma-2-27b was corrupted too)

thank you for quick fix!