xinyun/vllm
1
0
Fork 0
mirror of https://git.datalinker.icu/vllm-project/vllm.git synced 2026-09-09 09:37:01 +08:00
Code Issues Packages Projects Releases Wiki Activity

Add timer

Browse Source
This commit is contained in:
Woosuk Kwon 2024-04-25 05:06:11 +00:00
parent 81b8b813f1
commit 98eda57899
1 changed files with 14 additions and 0 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

14
vllm/worker/tpu_model_runner.py
View File

@ -1,3 +1,4 @@
import time
from typing import Any, Dict, List, Optional, Tuple from typing import Any, Dict, List, Optional, Tuple
import numpy as np import numpy as np
@ -213,9 +214,22 @@ class TPUModelRunner:
) -> Tuple[Optional[SamplerOutput], List[jax.Array]]: ) -> Tuple[Optional[SamplerOutput], List[jax.Array]]:
from vllm.sequence import SequenceOutput, SequenceGroupOutput, Logprob from vllm.sequence import SequenceOutput, SequenceGroupOutput, Logprob
start = time.time()
inputs = self.prepare_input_arrays(seq_group_metadata_list) inputs = self.prepare_input_arrays(seq_group_metadata_list)
end = time.time()
print(f"prepare_input_arrays: {(end - start) * 1000:.2f} ms")
start = time.time()
next_token_ids, new_kv_caches = self.compiled_fn(self.params, *inputs, kv_caches) next_token_ids, new_kv_caches = self.compiled_fn(self.params, *inputs, kv_caches)
next_token_ids.block_until_ready()
end = time.time()
print(f"compiled_fn: {(end - start) * 1000:.2f} ms")
start = time.time()
next_token_ids = jax.device_put(next_token_ids, jax.devices("cpu")[0]) next_token_ids = jax.device_put(next_token_ids, jax.devices("cpu")[0])
end = time.time()
print(f"jax.device_put: {(end - start) * 1000:.2f} ms")
next_token_ids = next_token_ids.tolist() next_token_ids = next_token_ids.tolist()
i = 0 i = 0