[Bug]: Different logprobs output behaviour under vllm 0.8.0 and 0.8.1 #15381
Description
Your current environment
The output of `python collect_env.py`
PyTorch version: 2.6.0+cu124
Is debug build: False
CUDA used to build PyTorch: 12.4
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-6.5.0-41-generic-x86_64-with-glibc2.35
Is CUDA available: True
CUDA runtime version: Could not collect
CUDA_MODULE_LOADING set to: LAZY
GPU models and configuration:
GPU 0: NVIDIA L40S
GPU 1: NVIDIA L40S
GPU 2: NVIDIA L40S
GPU 3: NVIDIA L40S
GPU 4: NVIDIA L40S
GPU 5: NVIDIA L40S
GPU 6: NVIDIA L40S
GPU 7: NVIDIA L40S
Nvidia driver version: 550.127.08
cuDNN version: Probably one of the following:
/usr/local/cuda-11.2/targets/x86_64-linux/lib/libcudnn.so.8.1.1
/usr/local/cuda-11.2/targets/x86_64-linux/lib/libcudnn_adv_infer.so.8.1.1
/usr/local/cuda-11.2/targets/x86_64-linux/lib/libcudnn_adv_train.so.8.1.1
/usr/local/cuda-11.2/targets/x86_64-linux/lib/libcudnn_cnn_infer.so.8.1.1
/usr/local/cuda-11.2/targets/x86_64-linux/lib/libcudnn_cnn_train.so.8.1.1
/usr/local/cuda-11.2/targets/x86_64-linux/lib/libcudnn_ops_infer.so.8.1.1
/usr/local/cuda-11.2/targets/x86_64-linux/lib/libcudnn_ops_train.so.8.1.1
/usr/local/cuda-12.2/targets/x86_64-linux/lib/libcudnn.so.8.9.3
/usr/local/cuda-12.2/targets/x86_64-linux/lib/libcudnn_adv_infer.so.8.9.3
/usr/local/cuda-12.2/targets/x86_64-linux/lib/libcudnn_adv_train.so.8.9.3
/usr/local/cuda-12.2/targets/x86_64-linux/lib/libcudnn_cnn_infer.so.8.9.3
/usr/local/cuda-12.2/targets/x86_64-linux/lib/libcudnn_cnn_train.so.8.9.3
/usr/local/cuda-12.2/targets/x86_64-linux/lib/libcudnn_ops_infer.so.8.9.3
/usr/local/cuda-12.2/targets/x86_64-linux/lib/libcudnn_ops_train.so.8.9.3
/usr/local/cuda-12.4/targets/x86_64-linux/lib/libcudnn.so.9.8.0
/usr/local/cuda-12.4/targets/x86_64-linux/lib/libcudnn_adv.so.9.8.0
/usr/local/cuda-12.4/targets/x86_64-linux/lib/libcudnn_cnn.so.9.8.0
/usr/local/cuda-12.4/targets/x86_64-linux/lib/libcudnn_engines_precompiled.so.9.8.0
/usr/local/cuda-12.4/targets/x86_64-linux/lib/libcudnn_engines_runtime_compiled.so.9.8.0
/usr/local/cuda-12.4/targets/x86_64-linux/lib/libcudnn_graph.so.9.8.0
/usr/local/cuda-12.4/targets/x86_64-linux/lib/libcudnn_heuristic.so.9.8.0
/usr/local/cuda-12.4/targets/x86_64-linux/lib/libcudnn_ops.so.9.8.0
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: 52 bits physical, 57 bits virtual
Byte Order: Little Endian
CPU(s): 256
On-line CPU(s) list: 0-255
Vendor ID: AuthenticAMD
Model name: AMD EPYC 9534 64-Core Processor
CPU family: 25
Model: 17
Thread(s) per core: 2
Core(s) per socket: 64
Socket(s): 2
Stepping: 1
Frequency boost: enabled
CPU max MHz: 3718.0659
CPU min MHz: 1500.0000
BogoMIPS: 4900.31
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 amd_lbr_v2 nopl nonstop_tsc cpuid extd_apicid aperfmperf rapl pni pclmulqdq monitor ssse3 fma cx16 pcid 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 invpcid_single hw_pstate ssbd mba perfmon_v2 ibrs ibpb stibp ibrs_enhanced vmmcall fsgsbase bmi1 avx2 smep bmi2 erms invpcid cqm rdt_a avx512f avx512dq rdseed adx smap avx512ifma clflushopt clwb avx512cd sha_ni avx512bw avx512vl xsaveopt xsavec xgetbv1 xsaves cqm_llc cqm_occup_llc cqm_mbm_total cqm_mbm_local avx512_bf16 clzero irperf xsaveerptr rdpru wbnoinvd amd_ppin cppc arat npt lbrv svm_lock nrip_save tsc_scale vmcb_clean flushbyasid decodeassists pausefilter pfthreshold avic v_vmsave_vmload vgif x2avic v_spec_ctrl vnmi avx512vbmi umip pku ospke avx512_vbmi2 gfni vaes vpclmulqdq avx512_vnni avx512_bitalg avx512_vpopcntdq la57 rdpid overflow_recov succor smca fsrm flush_l1d
Virtualization: AMD-V
L1d cache: 4 MiB (128 instances)
L1i cache: 4 MiB (128 instances)
L2 cache: 128 MiB (128 instances)
L3 cache: 512 MiB (16 instances)
NUMA node(s): 2
NUMA node0 CPU(s): 0-63,128-191
NUMA node1 CPU(s): 64-127,192-255
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; Enhanced / Automatic IBRS; IBPB conditional; 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] numpy==1.26.4
[pip3] nvidia-cublas-cu12==12.4.5.8
[pip3] nvidia-cuda-cupti-cu12==12.4.127
[pip3] nvidia-cuda-nvrtc-cu12==12.4.127
[pip3] nvidia-cuda-runtime-cu12==12.4.127
[pip3] nvidia-cudnn-cu12==9.1.0.70
[pip3] nvidia-cufft-cu12==11.2.1.3
[pip3] nvidia-curand-cu12==10.3.5.147
[pip3] nvidia-cusolver-cu12==11.6.1.9
[pip3] nvidia-cusparse-cu12==12.3.1.170
[pip3] nvidia-cusparselt-cu12==0.6.2
[pip3] nvidia-ml-py==12.570.86
[pip3] nvidia-nccl-cu12==2.21.5
[pip3] nvidia-nvjitlink-cu12==12.4.127
[pip3] nvidia-nvtx-cu12==12.4.127
[pip3] pyzmq==26.2.1
[pip3] torch==2.6.0
[pip3] torchaudio==2.6.0
[pip3] torchvision==0.21.0
[pip3] transformers==4.50.0.dev0
[pip3] triton==3.2.0
[conda] Could not collect
ROCM Version: Could not collect
Neuron SDK Version: N/A
vLLM Version: 0.8.1
vLLM Build Flags:
CUDA Archs: Not Set; ROCm: Disabled; Neuron: Disabled
GPU Topology:
GPU0 GPU1 GPU2 GPU3 GPU4 GPU5 GPU6 GPU7 CPU Affinity NUMA Affinity GPU NUMA ID
GPU0 X NODE NODE NODE SYS SYS SYS SYS 0-63,128-191 0 N/A
GPU1 NODE X NODE NODE SYS SYS SYS SYS 0-63,128-191 0 N/A
GPU2 NODE NODE X NODE SYS SYS SYS SYS 0-63,128-191 0 N/A
GPU3 NODE NODE NODE X SYS SYS SYS SYS 0-63,128-191 0 N/A
GPU4 SYS SYS SYS SYS X NODE NODE NODE 64-127,192-255 1 N/A
GPU5 SYS SYS SYS SYS NODE X NODE NODE 64-127,192-255 1 N/A
GPU6 SYS SYS SYS SYS NODE NODE X NODE 64-127,192-255 1 N/A
GPU7 SYS SYS SYS SYS NODE NODE NODE X 64-127,192-255 1 N/A
Legend:
X = Self
SYS = Connection traversing PCIe as well as the SMP interconnect between NUMA nodes (e.g., QPI/UPI)
NODE = Connection traversing PCIe as well as the interconnect between PCIe Host Bridges within a NUMA node
PHB = Connection traversing PCIe as well as a PCIe Host Bridge (typically the CPU)
PXB = Connection traversing multiple PCIe bridges (without traversing the PCIe Host Bridge)
PIX = Connection traversing at most a single PCIe bridge
NV# = Connection traversing a bonded set of # NVLinks
LD_LIBRARY_PATH=:/usr/local/cuda-11.7/extras/CUPTI/lib64:/usr/local/cuda-11.7/extras/CUPTI/lib64:/usr/local/cuda-11.7/lib64
NCCL_CUMEM_ENABLE=0
TORCHINDUCTOR_COMPILE_THREADS=1
CUDA_MODULE_LOADING=LAZY
🐛 Describe the bug
I use VLLM by starting it as an OpenAI-compatible server. When requesting logprobs from it, vllm in version 0.8.0 and 0.8.1 shows a different behaviour than before.
It can be reproduced with the following function:
from openai import OpenAI
def openAI_test(url,model):
client = OpenAI( base_url=url,api_key="dummy")
openai_response = client.chat.completions.create(
model=model,
messages=[{"role": "user", "content": "What is the capital of Germany! "}],
temperature=0.0,
logprobs=True,
#top_logprobs=1
)
print(openai_response.choices[0].logprobs)
I started a vllm server with the model Qwen2.5-VL-7B-Instruct and send with the above function a request to it.
Print output with vllm 0.8.0 or 0.8.1:
ChoiceLogprobs(content=[ChatCompletionTokenLogprob(token='The', bytes=[84, 104, 101], logprob=-0.0005716835148632526, top_logprobs=[]), ChatCompletionTokenLogprob(token='Ġcapital', bytes=[196, 160, 99, 97, 112, 105, 116, 97, 108], logprob=-8.630380034446716e-05, top_logprobs=[]), ChatCompletionTokenLogprob(token='Ġof', bytes=[196, 160, 111, 102], logprob=-0.10054320842027664, top_logprobs=[]), ChatCompletionTokenLogprob(token='ĠGermany', bytes=[196, 160, 71, 101, 114, 109, 97, 110, 121], logprob=-4.994744449504651e-05, top_logprobs=[]), ChatCompletionTokenLogprob(token='Ġis', bytes=[196, 160, 105, 115], logprob=-0.0004349001101218164, top_logprobs=[]), ChatCompletionTokenLogprob(token='ĠBerlin', bytes=[196, 160, 66, 101, 114, 108, 105, 110], logprob=-0.0014521064003929496, top_logprobs=[]), ChatCompletionTokenLogprob(token='.', bytes=[46], logprob=-0.015473785810172558, top_logprobs=[]), ChatCompletionTokenLogprob(token='<|im_end|>', bytes=[60, 124, 105, 109, 95, 101, 110, 100, 124, 62], logprob=-0.02971755340695381, top_logprobs=[])], refusal=None)
Print output with vllm 0.7.3 or lower:
ChoiceLogprobs(content=[ChatCompletionTokenLogprob(token='The', bytes=[84, 104, 101], logprob=-0.0006090931710787117, top_logprobs=[]), ChatCompletionTokenLogprob(token=' capital', bytes=[32, 99, 97, 112, 105, 116, 97, 108], logprob=-7.70062324590981e-05, top_logprobs=[]), ChatCompletionTokenLogprob(token=' of', bytes=[32, 111, 102], logprob=-0.1005404070019722, top_logprobs=[]), ChatCompletionTokenLogprob(token=' Germany', bytes=[32, 71, 101, 114, 109, 97, 110, 121], logprob=-4.6132929128361866e-05, top_logprobs=[]), ChatCompletionTokenLogprob(token=' is', bytes=[32, 105, 115], logprob=-0.0005459486856125295, top_logprobs=[]), ChatCompletionTokenLogprob(token=' Berlin', bytes=[32, 66, 101, 114, 108, 105, 110], logprob=-0.0014579391572624445, top_logprobs=[]), ChatCompletionTokenLogprob(token='.', bytes=[46], logprob=-0.013593370094895363, top_logprobs=[]), ChatCompletionTokenLogprob(token='', bytes=[], logprob=-0.0293378084897995, top_logprobs=[])], refusal=None)
It seems that in the newer vllm version the tokenwise output is given in the tokenizers internal representation of the used model (as in the vocab.json of the corresponding HF-model). New: Ġcapital . Old: capital. But this is a problem, because for an application on the client-site it is no longer possible to parse the output together with the logprob-positions without having special knowledge about the tokenizer. The official Open-AI server (see https://api.openai.com/v1/) behaves in the same way as the older VLLM versions.
Can this be fixed so that VLLM behaves as before? Or is there an easy way to convert the output to the old format?
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.