[Usage]: Difference in language model usage post updating versions form 0.2 to 0.4

[servient-ashwin]

trafficstars

Your current environment

The output of `python collect_env.py`

Collecting environment information...
PyTorch version: 2.2.1+cu121
Is debug build: False
CUDA used to build PyTorch: 12.1
ROCM used to build PyTorch: N/A

OS: Amazon Linux 2 (x86_64)
GCC version: (GCC) 7.3.1 20180712 (Red Hat 7.3.1-17)
Clang version: Could not collect
CMake version: version 3.27.7
Libc version: glibc-2.26

Python version: 3.10.9 | packaged by conda-forge | (main, Feb  2 2023, 20:20:04) [GCC 11.3.0] (64-bit runtime)
Python platform: Linux-5.10.213-201.855.amzn2.x86_64-x86_64-with-glibc2.26
Is CUDA available: True
CUDA runtime version: 12.1.105
CUDA_MODULE_LOADING set to: LAZY
GPU models and configuration: GPU 0: NVIDIA A10G
Nvidia driver version: 550.54.14
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
Byte Order:          Little Endian
CPU(s):              4
On-line CPU(s) list: 0-3
Thread(s) per core:  2
Core(s) per socket:  2
Socket(s):           1
NUMA node(s):        1
Vendor ID:           AuthenticAMD
CPU family:          23
Model:               49
Model name:          AMD EPYC 7R32
Stepping:            0
CPU MHz:             3293.001
BogoMIPS:            5600.00
Hypervisor vendor:   KVM
Virtualization type: full
L1d cache:           32K
L1i cache:           32K
L2 cache:            512K
L3 cache:            8192K
NUMA node0 CPU(s):   0-3
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 tsc_known_freq pni pclmulqdq ssse3 fma cx16 sse4_1 sse4_2 movbe popcnt aes xsave avx f16c rdrand hypervisor lahf_lm cmp_legacy cr8_legacy abm sse4a misalignsse 3dnowprefetch topoext ssbd ibrs ibpb stibp vmmcall fsgsbase bmi1 avx2 smep bmi2 rdseed adx smap clflushopt clwb sha_ni xsaveopt xsavec xgetbv1 clzero xsaveerptr rdpru wbnoinvd arat npt nrip_save rdpid

Versions of relevant libraries:
[pip3] mypy==1.9.0
[pip3] mypy-extensions==1.0.0
[pip3] numpy==1.26.2
[pip3] nvidia-nccl-cu11==2.14.3
[pip3] nvidia-nccl-cu12==2.19.3
[pip3] torch==2.2.1
[pip3] triton==2.2.0
[pip3] vllm-nccl-cu12==2.18.1.0.4.0
[conda] numpy                     1.26.2                   pypi_0    pypi
[conda] nvidia-nccl-cu11          2.14.3                   pypi_0    pypi
[conda] nvidia-nccl-cu12          2.19.3                   pypi_0    pypi
[conda] torch                     2.2.1                    pypi_0    pypi
[conda] triton                    2.2.0                    pypi_0    pypi
[conda] vllm-nccl-cu12            2.18.1.0.4.0             pypi_0    pypiROCM Version: Could not collect
Neuron SDK Version: N/A
vLLM Version: 0.4.1
vLLM Build Flags:
CUDA Archs: Not Set; ROCm: Disabled; Neuron: Disabled
GPU Topology:
GPU0	CPU Affinity	NUMA Affinity	GPU NUMA ID
GPU0	 X 	0-3	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

How would you like to use vllm

I am observing a change in the behavior of vLLM since updating the vLLm library from v ~2.0+ to the latest v 0.4.1 build.

What are the changes?

1. The same Mistral 7B that ran on both OpenAI API and "traditional" API doesn't function anymore as is in the new version that I upgraded to. For example for an A10 GPU the API now requires the max-model-len param to be set for every model because it can't load any 7B models. API requires a max_model_len param to be set which I assume is a part of the engine arguments and from the description is the context length for the model and is made compulsory so that solves it, but what I am unable to figure out is why doesn't it load the model with it's full length after updating?

Now since I admit this jump from 0.2 to 0.4 is far from an ideal update when there are so many versions that were released in between them and I tried looking into the change logs but, couldn't discern what was going on. since I get this as an error output.

ValueError: The model's max seq len (32768) is larger than the maximum number of tokens that can be stored in KV cache (16864). Try increasing `gpu_memory_utilization` or decreasing `max_model_len` when initializing the engine.

Can I get help in understanding what changed?

Also I keep seeing a burst of these messages and not sure what the model is trying to do here

INFO:    xx:xx - "GET / HTTP/1.1" 404 Not Found
INFO 05-03 18:07:24 metrics.py:229] Avg prompt throughput: 0.0 tokens/s, Avg generation throughput: 0.0 tokens/s, Running: 0 reqs, Swapped: 0 reqs, Pending: 0 reqs, GPU KV cache usage: 0.0%, CPU KV cache usage: 0.0%
INFO 05-03 18:07:34 metrics.py:229] Avg prompt throughput: 0.0 tokens/s, Avg generation throughput: 0.0 tokens/s, Running: 0 reqs, Swapped: 0 reqs, Pending: 0 reqs, GPU KV cache usage: 0.0%, CPU KV cache usage: 0.0%

[servient-ashwin]

I have done some sanity checks here:

https://github.com/vllm-project/vllm/pull/1395 https://github.com/vllm-project/vllm/pull/1796 https://github.com/vllm-project/vllm/pull/1858 https://github.com/vllm-project/vllm/pull/972

as I found these were the closest to what I am experiencing.