Skip to content

Error when using Vidur profiling for Llama-2-70b attention core #55

Description

time python ../vidur/profiling/mlp/main.py --models meta-llama/Llama-2-70b-hf --num_gpus 8 > mlp.txt && time python ../vidur/profiling/attention/main.py --models meta-llama/Llama-2-70b-hf --num_gpus 8 > att.txt
0%| | 0/1044 [00:00<?, ?it/s]2025-03-17 02:10:22,262 INFO worker.py:1832 -- Started a local Ray instance. View the dashboard at 127.0.0.1:8265
100%|████████████████████████████████████████████████████████████████████████| 1044/1044 [02:23<00:00, 7.26it/s]

real 2m31.669s
user 0m55.454s
sys 1m13.662s
0%| | 0/43888 [00:00<?, ?it/s]2025-03-17 02:12:55,162 INFO worker.py:1832 -- Started a local Ray instance. View the dashboard at 127.0.0.1:8265
0%| | 1/43888 [00:08<105:25:58, 8.65s/it](AttentionWrapper pid=149911) 2025-03-17 02:13:07,351 - INFO - flashinfer.jit: Loading JIT ops: batch_prefill_with_kv_cache_dtype_q_f16_dtype_kv_f16_dtype_o_f16_dtype_idx_i32_head_dim_128_posenc_0_use_swa_False_use_logits_cap_False_f16qk_False
(AttentionWrapper pid=149911) /home/huday/vidur_test/.venv/lib/python3.10/site-packages/torch/utils/cpp_extension.py:1967: UserWarning: TORCH_CUDA_ARCH_LIST is not set, all archs for visible cards are included for compilation.
(AttentionWrapper pid=149911) If this is not desired, please set os.environ['TORCH_CUDA_ARCH_LIST'].
(AttentionWrapper pid=149911) warnings.warn(
(AttentionWrapper pid=149913) 2025-03-17 02:13:07,363 - INFO - flashinfer.jit: Finished loading JIT ops: batch_prefill_with_kv_cache_dtype_q_f16_dtype_kv_f16_dtype_o_f16_dtype_idx_i32_head_dim_128_posenc_0_use_swa_False_use_logits_cap_False_f16qk_False
Traceback (most recent call last):
File "/home/huday/vidur_test/vidur/data/../vidur/profiling/attention/main.py", line 240, in
main()
File "/home/huday/vidur_test/vidur/data/../vidur/profiling/attention/main.py", line 223, in main
profile_model(
File "/home/huday/vidur_test/vidur/data/../vidur/profiling/attention/main.py", line 157, in profile_model
results = ray.get(promises)
File "/home/huday/vidur_test/.venv/lib/python3.10/site-packages/ray/_private/auto_init_hook.py", line 21, in auto_init_wrapper
return fn(*args, **kwargs)
File "/home/huday/vidur_test/.venv/lib/python3.10/site-packages/ray/_private/client_mode_hook.py", line 103, in wrapper
return func(*args, **kwargs)
File "/home/huday/vidur_test/.venv/lib/python3.10/site-packages/ray/_private/worker.py", line 2771, in get
values, debugger_breakpoint = worker.get_objects(object_refs, timeout=timeout)
File "/home/huday/vidur_test/.venv/lib/python3.10/site-packages/ray/_private/worker.py", line 919, in get_objects
raise value.as_instanceof_cause()
ray.exceptions.RayTaskError(TypeError): ray::AttentionWrapper.profile() (pid=149913, ip=172.17.0.2, actor_id=533c75b2787547df5b4810eb01000000, repr=<vidur.profiling.attention.attention_wrapper.AttentionWrapper object at 0x7f181e045870>)
File "/home/huday/vidur_test/.venv/lib/python3.10/site-packages/torch/utils/_contextlib.py", line 115, in decorate_context
return func(*args, **kwargs)
File "/home/huday/vidur_test/vidur/vidur/profiling/attention/attention_wrapper.py", line 131, in profile
get_attention_wrapper().forward(query, key, value, kv_cache)
File "/home/huday/vidur_test/sarathi-serve/sarathi/model_executor/attention/flashinfer_attention_wrapper.py", line 224, in forward
append_paged_kv_cache(
TypeError: append_paged_kv_cache() missing 1 required positional argument: 'kv_last_page_len'
(AttentionWrapper pid=149914) 2025-03-17 02:13:07,832 - INFO - flashinfer.jit: Loading JIT ops: batch_prefill_with_kv_cache_dtype_q_f16_dtype_kv_f16_dtype_o_f16_dtype_idx_i32_head_dim_128_posenc_0_use_swa_False_use_logits_cap_False_f16qk_False [repeated 6x across cluster] (Ray deduplicates logs by default. Set RAY_DEDUP_LOGS=0 to disable log deduplication, or see https://docs.ray.io/en/master/ray-observability/user-guides/configure-logging.html#log-deduplication for more options.)
(AttentionWrapper pid=149915) /home/huday/vidur_test/.venv/lib/python3.10/site-packages/torch/utils/cpp_extension.py:1967: UserWarning: TORCH_CUDA_ARCH_LIST is not set, all archs for visible cards are included for compilation. [repeated 12x across cluster]
(AttentionWrapper pid=149915) If this is not desired, please set os.environ['TORCH_CUDA_ARCH_LIST']. [repeated 12x across cluster]
(AttentionWrapper pid=149915) warnings.warn( [repeated 12x across cluster]
(AttentionWrapper pid=149915) 2025-03-17 02:13:07,911 - INFO - flashinfer.jit: Finished loading JIT ops: batch_prefill_with_kv_cache_dtype_q_f16_dtype_kv_f16_dtype_o_f16_dtype_idx_i32_head_dim_128_posenc_0_use_swa_False_use_logits_cap_False_f16qk_False [repeated 5x across cluster]
0%| | 7/43888 [00:17<31:07:03, 2.55s/it]

real 0m23.165s
user 0m33.359s
sys 0m17.043s

Metadata

Metadata

Assignees

No one assigned

    Labels

    No labels
    No labels

    Type

    No type

    Projects

    No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions