vllm-project
[Bug]: Can't use yarn rope config for long context in Qwen2 model
Your current environment
The output of `python collect_env.py`
Collecting environment information...
PyTorch version: 2.4.0+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 11.4.0-1ubuntu1~22.04) 11.4.0
Clang version: Could not collect
CMake version: Could not collect
Libc version: glibc-2.35
Python version: 3.12.7 (main, Oct 1 2024, 08:52:12) [GCC 11.4.0] (64-bit runtime)
Python platform: Linux-5.14.0-284.25.1.el9_2.x86_64-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 H100 80GB HBM3
GPU 1: NVIDIA H100 80GB HBM3
GPU 2: NVIDIA H100 80GB HBM3
GPU 3: NVIDIA H100 80GB HBM3
GPU 4: NVIDIA H100 80GB HBM3
GPU 5: NVIDIA H100 80GB HBM3
GPU 6: NVIDIA H100 80GB HBM3
GPU 7: NVIDIA H100 80GB HBM3
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: 46 bits physical, 57 bits virtual
Byte Order: Little Endian
CPU(s): 128
On-line CPU(s) list: 0-127
Vendor ID: GenuineIntel
Model name: Intel(R) Xeon(R) Platinum 8462Y+
CPU family: 6
Model: 143
Thread(s) per core: 2
Core(s) per socket: 32
Socket(s): 2
Stepping: 8
CPU max MHz: 4100.0000
CPU min MHz: 800.0000
BogoMIPS: 5600.00
Flags: fpu vme de pse tsc msr pae mce cx8 apic sep mtrr pge mca cmov pat pse36 clflush dts acpi mmx fxsr sse sse2 ss ht tm pbe syscall nx pdpe1gb rdtscp lm constant_tsc art arch_perfmon pebs bts rep_good nopl xtopology nonstop_tsc cpuid aperfmperf tsc_known_freq pni pclmulqdq dtes64 monitor ds_cpl vmx smx est tm2 ssse3 sdbg fma cx16 xtpr pdcm pcid dca sse4_1 sse4_2 x2apic movbe popcnt tsc_deadline_timer aes xsave avx f16c rdrand lahf_lm abm 3dnowprefetch cpuid_fault epb cat_l3 cat_l2 cdp_l3 invpcid_single intel_ppin cdp_l2 ssbd mba ibrs ibpb stibp ibrs_enhanced tpr_shadow vnmi flexpriority ept vpid ept_ad fsgsbase tsc_adjust bmi1 avx2 smep bmi2 erms invpcid cqm rdt_a avx512f avx512dq rdseed adx smap avx512ifma clflushopt clwb intel_pt avx512cd sha_ni avx512bw avx512vl xsaveopt xsavec xgetbv1 xsaves cqm_llc cqm_occup_llc cqm_mbm_total cqm_mbm_local split_lock_detect avx_vnni avx512_bf16 wbnoinvd dtherm ida arat pln pts hwp hwp_act_window hwp_epp hwp_pkg_req hfi avx512vbmi umip pku ospke waitpkg avx512_vbmi2 gfni vaes vpclmulqdq avx512_vnni avx512_bitalg tme avx512_vpopcntdq la57 rdpid bus_lock_detect cldemote movdiri movdir64b enqcmd fsrm md_clear serialize tsxldtrk pconfig arch_lbr ibt amx_bf16 avx512_fp16 amx_tile amx_int8 flush_l1d arch_capabilities
Virtualization: VT-x
L1d cache: 3 MiB (64 instances)
L1i cache: 2 MiB (64 instances)
L2 cache: 128 MiB (64 instances)
L3 cache: 120 MiB (2 instances)
NUMA node(s): 2
NUMA node0 CPU(s): 0-31,64-95
NUMA node1 CPU(s): 32-63,96-127
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 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 IBRS, IBPB conditional, RSB filling, PBRSB-eIBRS SW sequence
Vulnerability Srbds: Not affected
Vulnerability Tsx async abort: Not affected
Versions of relevant libraries:
[pip3] flashinfer==0.1.6+cu121torch2.4
[pip3] numpy==1.26.4
[pip3] nvidia-cublas-cu12==12.1.3.1
[pip3] nvidia-cuda-cupti-cu12==12.1.105
[pip3] nvidia-cuda-nvrtc-cu12==12.1.105
[pip3] nvidia-cuda-runtime-cu12==12.1.105
[pip3] nvidia-cudnn-cu12==9.1.0.70
[pip3] nvidia-cufft-cu12==11.0.2.54
[pip3] nvidia-curand-cu12==10.3.2.106
[pip3] nvidia-cusolver-cu12==11.4.5.107
[pip3] nvidia-cusparse-cu12==12.1.0.106
[pip3] nvidia-ml-py==12.560.30
[pip3] nvidia-nccl-cu12==2.20.5
[pip3] nvidia-nvjitlink-cu12==12.6.77
[pip3] nvidia-nvtx-cu12==12.1.105
[pip3] pyzmq==26.2.0
[pip3] torch==2.4.0
[pip3] torchvision==0.19.0
[pip3] transformers==4.45.2
[pip3] triton==3.0.0
[conda] Could not collect
ROCM Version: Could not collect
Neuron SDK Version: N/A
vLLM Version: 0.6.3.post1
vLLM Build Flags:
CUDA Archs: Not Set; ROCm: Disabled; Neuron: Disabled
GPU Topology:
[4mGPU0 GPU1 GPU2 GPU3 GPU4 GPU5 GPU6 GPU7 NIC0 NIC1 NIC2 NIC3 NIC4 NIC5 NIC6 NIC7 NIC8 NIC9 CPU Affinity NUMA Affinity GPU NUMA ID[0m
GPU0 X NV18 NV18 NV18 NV18 NV18 NV18 NV18 PXB PXB NODE NODE NODE NODE SYS SYS SYS SYS 0-31,64-95 0 N/A
GPU1 NV18 X NV18 NV18 NV18 NV18 NV18 NV18 PXB PXB NODE NODE NODE NODE SYS SYS SYS SYS 0-31,64-95 0 N/A
GPU2 NV18 NV18 X NV18 NV18 NV18 NV18 NV18 NODE NODE NODE NODE PXB PXB SYS SYS SYS SYS 0-31,64-95 0 N/A
GPU3 NV18 NV18 NV18 X NV18 NV18 NV18 NV18 NODE NODE NODE NODE PXB PXB SYS SYS SYS SYS 0-31,64-95 0 N/A
GPU4 NV18 NV18 NV18 NV18 X NV18 NV18 NV18 SYS SYS SYS SYS SYS SYS PXB PXB NODE NODE 32-63,96-127 1 N/A
GPU5 NV18 NV18 NV18 NV18 NV18 X NV18 NV18 SYS SYS SYS SYS SYS SYS PXB PXB NODE NODE 32-63,96-127 1 N/A
GPU6 NV18 NV18 NV18 NV18 NV18 NV18 X NV18 SYS SYS SYS SYS SYS SYS NODE NODE PXB PXB 32-63,96-127 1 N/A
GPU7 NV18 NV18 NV18 NV18 NV18 NV18 NV18 X SYS SYS SYS SYS SYS SYS NODE NODE PXB PXB 32-63,96-127 1 N/A
NIC0 PXB PXB NODE NODE SYS SYS SYS SYS X PXB NODE NODE NODE NODE SYS SYS SYS SYS
NIC1 PXB PXB NODE NODE SYS SYS SYS SYS PXB X NODE NODE NODE NODE SYS SYS SYS SYS
NIC2 NODE NODE NODE NODE SYS SYS SYS SYS NODE NODE X PIX NODE NODE SYS SYS SYS SYS
NIC3 NODE NODE NODE NODE SYS SYS SYS SYS NODE NODE PIX X NODE NODE SYS SYS SYS SYS
NIC4 NODE NODE PXB PXB SYS SYS SYS SYS NODE NODE NODE NODE X PXB SYS SYS SYS SYS
NIC5 NODE NODE PXB PXB SYS SYS SYS SYS NODE NODE NODE NODE PXB X SYS SYS SYS SYS
NIC6 SYS SYS SYS SYS PXB PXB NODE NODE SYS SYS SYS SYS SYS SYS X PXB NODE NODE
NIC7 SYS SYS SYS SYS PXB PXB NODE NODE SYS SYS SYS SYS SYS SYS PXB X NODE NODE
NIC8 SYS SYS SYS SYS NODE NODE PXB PXB SYS SYS SYS SYS SYS SYS NODE NODE X PXB
NIC9 SYS SYS SYS SYS NODE NODE PXB PXB SYS SYS SYS SYS SYS SYS NODE NODE PXB X
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
NIC Legend:
NIC0: mlx5_0
NIC1: mlx5_1
NIC2: mlx5_2
NIC3: mlx5_3
NIC4: mlx5_4
NIC5: mlx5_5
NIC6: mlx5_6
NIC7: mlx5_7
NIC8: mlx5_8
NIC9: mlx5_9
NVIDIA_VISIBLE_DEVICES=GPU-4874116f-3878-1e53-8e46-120150b0b458,GPU-f56ccdda-e68b-fe5e-c5bf-12e8f08225fe,GPU-209e16ec-a6b6-a4a1-fbdd-04d686dda7b5,GPU-7f17f885-b8df-830e-ddba-6a9245e12adb,GPU-cc477212-a933-02f2-181a-eb37183f118e,GPU-6a3c81de-e3ca-fa3d-007d-23c3028efcce,GPU-f11f6342-4207-db4b-d663-74abe28e4046,GPU-ec7601fc-87f5-1bfe-4fbe-1161dd1ae0e9
NVIDIA_REQUIRE_CUDA=cuda>=12.4 brand=tesla,driver>=470,driver<471 brand=unknown,driver>=470,driver<471 brand=nvidia,driver>=470,driver<471 brand=nvidiartx,driver>=470,driver<471 brand=geforce,driver>=470,driver<471 brand=geforcertx,driver>=470,driver<471 brand=quadro,driver>=470,driver<471 brand=quadrortx,driver>=470,driver<471 brand=titan,driver>=470,driver<471 brand=titanrtx,driver>=470,driver<471 brand=tesla,driver>=525,driver<526 brand=unknown,driver>=525,driver<526 brand=nvidia,driver>=525,driver<526 brand=nvidiartx,driver>=525,driver<526 brand=geforce,driver>=525,driver<526 brand=geforcertx,driver>=525,driver<526 brand=quadro,driver>=525,driver<526 brand=quadrortx,driver>=525,driver<526 brand=titan,driver>=525,driver<526 brand=titanrtx,driver>=525,driver<526 brand=tesla,driver>=535,driver<536 brand=unknown,driver>=535,driver<536 brand=nvidia,driver>=535,driver<536 brand=nvidiartx,driver>=535,driver<536 brand=geforce,driver>=535,driver<536 brand=geforcertx,driver>=535,driver<536 brand=quadro,driver>=535,driver<536 brand=quadrortx,driver>=535,driver<536 brand=titan,driver>=535,driver<536 brand=titanrtx,driver>=535,driver<536
NVIDIA_DRIVER_CAPABILITIES=compute,utility
VLLM_USAGE_SOURCE=production-docker-image
CUDA_VERSION=12.4.1
NCCL_IB_AR_THRESHOLD=0
NCCL_IB_TIMEOUT=22
LD_LIBRARY_PATH=/usr/local/lib/python3.12/dist-packages/cv2/../../lib64:/usr/local/nvidia/lib:/usr/local/nvidia/lib64
NCCL_IB_RETRY_CNT=13
CUDA_MODULE_LOADING=LAZY
Model Input Dumps
No response
🐛 Describe the bug
when I add rope config in Qwen/Qwen2-72B-Instruct/config.json
"rope_scaling": {
"factor": 4.0,
"original_max_position_embeddings": 32768,
"type": "yarn"
}
I get transformers warning:
"Unrecognized keys in `rope_scaling` for 'rope_type'='yarn': {'original_max_position_embeddings'}"
And, when LLM's input length is shorter than original_position_embedding_len, the response is OK. However, if input's len is larger than 32768(original_position_embedding_len), the model's output will be something confusing, similar to a kind of repetition of the input. this error happened in the version of 0.6.3.post1, but when I switch to v0.6.0, everything is OK.
I find that transformers's repo from recent versions don't accept "original_max_position_embeddings", but vllm need it. Maybe this is a confict between transformers and vllm ?
Does anyone know how to correctly enable the long context feature? Thanks ^_^
PS: I can't run collect_env.py script in v0.6.0's docker image, but 0.6.3.post1's docker image is OK. PPS: I just search issues about "original_max_position_embeddings", but got nothing releated.
Before submitting a new issue...
- [X] 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.
Hi @DarkLight1337 the config is listed on the Qwen documentation in guidance for configuring long context with vllm:
https://qwen.readthedocs.io/en/latest/deployment/vllm.html#extended-context-support
I can also report the presence of the bug on 0.6.3.post1. using vLLM 0.6.0 does not remove the warning:
Unrecognized keys in `rope_scaling` for 'rope_type'='yarn': {'original_max_position_embeddings'}
However, empirically, version 0.6.0 does allow for longer context with the configuration enabled. So thanks for that tip @FlyCarrot !
I think this is link maybe help. https://huggingface.co/meta-llama/Llama-3.1-8B-Instruct/discussions/15#670a76a4e3d216b4243b2d4d In short, I Guess : vllm nedd original_max_position_embeddings but transformers do not allow this key. So, anybody knows how to balance it? Thanks!
~~I just found the issue - you should pass "rope_type" key instead of "type".~~ Actually that doesn't seem to be it.
same issue with vllm 0.6.4.post1 and If this warning occurs, the model's generation gets cut off at a very early stage.
same issue with vllm 0.6.6.post1
Unrecognized keys in
rope_scalingfor 'rope_type'='yarn': {'original_max_position_embeddings'}
Have you tried passing --hf-overrides '{"rope_scaling": {"factor": 4.0, "original_max_position_embeddings": 32768, "type": "yarn"}}'?
Have you tried passing
--hf-overrides '{"rope_scaling": {"factor": 4.0, "original_max_position_embeddings": 32768, "type": "yarn"}}'?
It is useful to me, I try to add the params directly to the start command. like ' --rope-scaling '{"type":"yarn","factor":4.0, "original_max_position_embeddings":32768,"rope_type": "yarn"}', and the qwen 72B now can use yarn to infer very long token .I change the length to 131k. thank u ,DarkLight1337
Can you also open an issue on Qwen side so they can update their documentation?
use this way
python -m vllm.entrypoints.openai.api_server \
--model "path to model" \
--port 8459 \
--gpu-memory-utilization 0.95 \
--trust-remote-code \
--enforce-eager \
--max-model-len 65536 \
--served-model-name qwen \
--tensor-parallel-size 4 \
--rope-scaling '{"factor": 4.0, "original_max_position_embeddings": 32768, "type": "yarn"}'
work in v0.6.3.post.1
use this way
python -m vllm.entrypoints.openai.api_server \ --model "path to model" \ --port 8459 \ --gpu-memory-utilization 0.95 \ --trust-remote-code \ --enforce-eager \ --max-model-len 65536 \ --served-model-name qwen \ --tensor-parallel-size 4 \ --rope-scaling '{"factor": 4.0, "original_max_position_embeddings": 32768, "type": "yarn"}'work in v0.6.3.post.1
Offline infer still reports an error
Have you tried passing
--hf-overrides '{"rope_scaling": {"factor": 4.0, "original_max_position_embeddings": 32768, "type": "yarn"}}'?It is useful to me, I try to add the params directly to the start command. like ' --rope-scaling '{"type":"yarn","factor":4.0, "original_max_position_embeddings":32768,"rope_type": "yarn"}', and the qwen 72B now can use yarn to infer very long token .I change the length to 131k. thank u ,DarkLight1337
Offline infer still reports an error
What is your transformers and vLLM version? Can you show the stack trace of the error?
I think this is link maybe help. https://huggingface.co/meta-llama/Llama-3.1-8B-Instruct/discussions/15#670a76a4e3d216b4243b2d4d In short, I Guess : vllm nedd original_max_position_embeddings but transformers do not allow this key. So, anybody knows how to balance it? Thanks!
I finally found out how to solve this warning, downloading specific transfomers and vllm can solve it. transfomers==4.43.2 vllm==0.6.0
What is your transformers and vLLM version? Can you show the stack trace of the error?
I didn't try the server approach, and general vllm offline inference doesn't seem to be solved.
I finally found out how to solve this warning, downloading specific transfomers and vllm can solve it. transfomers==4.43.2 vllm==0.6.0
What is your transformers and vLLM version? Can you show the stack trace of the error?
I didn't try the server approach, and general vllm offline inference doesn't seem to be solved.
I finally found out how to solve this warning, downloading specific transfomers and vllm can solve it. transfomers==4.43.2 vllm==0.6.0
This also fixed the issue for me. Thanks!
I think this is link maybe help. https://huggingface.co/meta-llama/Llama-3.1-8B-Instruct/discussions/15#670a76a4e3d216b4243b2d4d In short, I Guess : vllm nedd original_max_position_embeddings but transformers do not allow this key. So, anybody knows how to balance it? Thanks!
I finally found out how to solve this warning, downloading specific transfomers and vllm can solve it. transfomers==4.43.2 vllm==0.6.0
hi, I use vllm by docker image, so I do not control specific transformers' version. Official docker image is really recommended. : )
The following would resolve the error for you.
python -m vllm.entrypoints.openai.api_server \
--model "path to model" \
--port 8459 \
--gpu-memory-utilization 0.95 \
--trust-remote-code \
--enforce-eager \
--max-model-len 65536 \
--served-model-name qwen \
--tensor-parallel-size 4 \
--rope-scaling '{"factor": 4.0, "original_max_position_embeddings": 32768, "rope_type": "yarn"}'
Use rope_type param instead of legacy param rope.
https://github.com/vllm-project/vllm/blob/bd56c983d6fe8ff93bddd5faaf8d96e01c90fd83/vllm/transformers_utils/config.py#L198-L199
It needs to be using dynamic Yarn, else short context would be ruined. Is dynamic yarn still not possible in vllm?
vLLM 0.7.2 user here. For me, using "rope_type" instead of "type" helped. But now I am not sure if context is processed correctly, as Qwen instruction page (maybe outdated?) tells to use "type". And that page also conirms YaRN should be used only with long contexts.
+1. Changing key from type to rope_type matches to what transformers expect and it removes warning. But does this confirms that the YaRN is implemented and working?
Changing key from type to rope_type doesn't remove warning for me, I'm using vllm==v0.7.3 and transformers==4.45.1
To my understanding this warning doesn't influence yarn computation in vllm. It is a warning caused by transformers being used to initialize hf config [code that causes warning], while actual YARN computation is performed in [vllm code]. The config that worked for me for Qwen-72b-Instruct (with good needle-in-the-haystack scores) is:
{
"architectures": [
"Qwen2ForCausalLM"
],
"attention_dropout": 0.0,
"bos_token_id": 151643,
"eos_token_id": 151645,
"hidden_act": "silu",
"hidden_size": 8192,
"initializer_range": 0.02,
"intermediate_size": 29568,
"max_position_embeddings": 32768,
"max_window_layers": 70,
"model_type": "qwen2",
"num_attention_heads": 64,
"num_hidden_layers": 80,
"num_key_value_heads": 8,
"rms_norm_eps": 1e-06,
"rope_scaling": {
"type": "yarn",
"original_max_position_embeddings": 32768,
"factor": 4.0
},
"rope_theta": 1000000.0,
"sliding_window": 131072,
"tie_word_embeddings": false,
"torch_dtype": "bfloat16",
"transformers_version": "4.45.1",
"use_cache": true,
"use_sliding_window": false,
"vocab_size": 152064
}
max_positions_embeddings seems to be ignored according to this [vllm code], so to infer with seq_len=65536, I had to explicitly pass --max-model-len 65536 via command line while serving vllm.
