Skip to content

Commit

Permalink
[torch.compile] remove reset (vllm-project#7975)
Browse files Browse the repository at this point in the history
  • Loading branch information
youkaichao authored Aug 29, 2024
1 parent 4289cad commit a7f65c2
Show file tree
Hide file tree
Showing 3 changed files with 28 additions and 15 deletions.
35 changes: 28 additions & 7 deletions tests/tpu/test_compilation.py
Original file line number Diff line number Diff line change
Expand Up @@ -5,6 +5,10 @@

import depyf

# disable custom dispatcher, let Dynamo takes over
# all the control
os.environ['VLLM_DYNAMO_USE_CUSTOM_DISPATCHER'] = "0"

temp_dir = tempfile.mkdtemp()
with depyf.prepare_debug(temp_dir):
cur_dir = os.path.dirname(__file__)
Expand All @@ -16,19 +20,36 @@

compiled_code = sorted(
glob.glob(os.path.join(temp_dir, "__transformed_code*.py")))
full_code = glob.glob(os.path.join(temp_dir, "full_code*.py"))[0]

# we should only trigger Dynamo compilation three times:
# one for the profiling phase (and the compiled artifact will be discarded)
# one for the profiling phase without kv cache
# one for the prefill phase with symbolic shapes
# one for the decode phase with symbolic shapes
# and later calls should not trigger Dynamo compilation again.
# NOTE: it might still trigger XLA compilation.

# check we have three compiled code
# this is the assumption when we use the custom dispatcher
assert len(compiled_code) == 3

# check the first compilation is discarded
with open(full_code) as f:
full_code_content = f.read()
profile_function = compiled_code[0].split(".")[0]
assert profile_function not in full_code_content
# check all the compilations are as expected
compiled_fn = sorted(
glob.glob(os.path.join(temp_dir, "__compiled_fn*Captured*.py")))

# the first compilation is the profiling phase,
# it should not have any kv cache
with open(compiled_fn[0]) as f:
content = f.read()
assert "kv_caches" not in content

# the second compilation is the prefill phase,
# it should have kv cache and the flash_attention op
with open(compiled_fn[1]) as f:
content = f.read()
assert "kv_caches" in content and "torch.ops.xla.flash_attention" in content

# the third compilation is the decode phase,
# it should have kv cache and the paged_attention op
with open(compiled_fn[2]) as f:
content = f.read()
assert "kv_caches" in content and "torch.ops.xla.paged_attention" in content
4 changes: 0 additions & 4 deletions vllm/worker/model_runner.py
Original file line number Diff line number Diff line change
Expand Up @@ -1123,10 +1123,6 @@ def profile_run(self) -> None:
device=self.device)
self.execute_model(model_input, kv_caches, intermediate_tensors)
torch.cuda.synchronize()

# reset and discard the guard and compiled bytecode for profiling runs
torch._dynamo.reset()

return

def remove_all_loras(self):
Expand Down
4 changes: 0 additions & 4 deletions vllm/worker/tpu_worker.py
Original file line number Diff line number Diff line change
Expand Up @@ -143,10 +143,6 @@ def determine_num_available_blocks(self) -> Tuple[int, int]:
num_cpu_blocks = int(self.cache_config.swap_space_bytes //
block_size_bytes)
num_cpu_blocks = (num_cpu_blocks // 8) * 8 # Round down to 8.

# reset and discard the guard and compiled bytecode for profiling runs
torch._dynamo.reset()

return num_tpu_blocks, num_cpu_blocks

def initialize_cache(
Expand Down

0 comments on commit a7f65c2

Please sign in to comment.