Skip to content

Michalkuligowski patch 7 #542

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Draft
wants to merge 3 commits into
base: main
Choose a base branch
from
Draft

Michalkuligowski patch 7 #542

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
3 commits
Select commit Hold shift + click to select a range
2cac21a
Update config.py
michalkuligowski Nov 6, 2025
ab0eb41
Update features.py
michalkuligowski Nov 6, 2025
84f0e03
Update config.py
michalkuligowski Nov 6, 2025
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
13 changes: 12 additions & 1 deletion vllm_gaudi/extension/config.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -143,8 +143,12 @@ def __init__(self, name: str, value_type: Constructor, check: Checker = skip_val
self.value_type = value_type
self.check = check

@cache
Copy link
Collaborator Author

@michalkuligowski michalkuligowski Nov 6, 2025

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

cache reading from the system for performance

def get_from_env(self, n):
return os.environ.get(n)

def __call__(self, _):
value = os.environ.get(self.name)
value = self.get_from_env(self.name)
if value is not None:
try:
value = self.value_type(value)
Expand All @@ -157,6 +161,13 @@ def __call__(self, _):
return None


class ExperimentalEnv(Env):
def __call__(self, _):
if Env('VLLM_ENABLE_EXPERIMENTAL_FLAGS', boolean)():
return super(ExperimentalEnv, self).__call__()
return None


class Value:
"""A callable that returns the value calculated through its dependencies or overriden by an associated experimental flag"""

Expand Down
56 changes: 28 additions & 28 deletions vllm_gaudi/extension/features.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -15,29 +15,29 @@ def get_user_flags():
Env('VLLM_USE_V1', boolean),
Env('VLLM_ENABLE_EXPERIMENTAL_FLAGS', boolean),
Env('VLLM_EXPONENTIAL_BUCKETING', boolean),
Env('VLLM_PROMPT_BS_BUCKET_MIN', int),
Env('VLLM_PROMPT_BS_BUCKET_STEP', int),
Env('VLLM_PROMPT_BS_BUCKET_MAX', int),
Env('VLLM_PROMPT_QUERY_BUCKET_MIN', int),
Env('VLLM_PROMPT_QUERY_BUCKET_STEP', int),
Env('VLLM_PROMPT_QUERY_BUCKET_MAX', int),
Env('VLLM_PROMPT_SEQ_BUCKET_MIN', int),
Env('VLLM_PROMPT_SEQ_BUCKET_STEP', int),
Env('VLLM_PROMPT_SEQ_BUCKET_MAX', int),
Env('VLLM_PROMPT_CTX_BUCKET_MIN', int),
Env('VLLM_PROMPT_CTX_BUCKET_STEP', int),
Env('VLLM_PROMPT_CTX_BUCKET_MAX', int),
Env('VLLM_DECODE_BS_BUCKET_MIN', int),
Env('VLLM_DECODE_BS_BUCKET_STEP', int),
Env('VLLM_DECODE_BS_BUCKET_MAX', int),
Env('VLLM_DECODE_BLOCK_BUCKET_MIN', int),
Env('VLLM_DECODE_BLOCK_BUCKET_STEP', int),
Env('VLLM_DECODE_BLOCK_BUCKET_MAX', int),
Env('VLLM_DECODE_BLOCK_BUCKET_LIMIT', int),
Env('VLLM_BUCKETING_FROM_FILE', str),
ExperimentalEnv('VLLM_PROMPT_BS_BUCKET_MIN', int),
ExperimentalEnv('VLLM_PROMPT_BS_BUCKET_STEP', int),
ExperimentalEnv('VLLM_PROMPT_BS_BUCKET_MAX', int),
ExperimentalEnv('VLLM_PROMPT_QUERY_BUCKET_MIN', int),
ExperimentalEnv('VLLM_PROMPT_QUERY_BUCKET_STEP', int),
ExperimentalEnv('VLLM_PROMPT_QUERY_BUCKET_MAX', int),
ExperimentalEnv('VLLM_PROMPT_SEQ_BUCKET_MIN', int),
ExperimentalEnv('VLLM_PROMPT_SEQ_BUCKET_STEP', int),
ExperimentalEnv('VLLM_PROMPT_SEQ_BUCKET_MAX', int),
ExperimentalEnv('VLLM_PROMPT_CTX_BUCKET_MIN', int),
ExperimentalEnv('VLLM_PROMPT_CTX_BUCKET_STEP', int),
ExperimentalEnv('VLLM_PROMPT_CTX_BUCKET_MAX', int),
ExperimentalEnv('VLLM_DECODE_BS_BUCKET_MIN', int),
ExperimentalEnv('VLLM_DECODE_BS_BUCKET_STEP', int),
ExperimentalEnv('VLLM_DECODE_BS_BUCKET_MAX', int),
ExperimentalEnv('VLLM_DECODE_BLOCK_BUCKET_MIN', int),
ExperimentalEnv('VLLM_DECODE_BLOCK_BUCKET_STEP', int),
ExperimentalEnv('VLLM_DECODE_BLOCK_BUCKET_MAX', int),
ExperimentalEnv('VLLM_DECODE_BLOCK_BUCKET_LIMIT', int),
ExperimentalEnv('VLLM_BUCKETING_FROM_FILE', str),

# Non-vllm flags that are also important to print
Env('EXPERIMENTAL_WEIGHT_SHARING', str),
ExperimentalEnv('EXPERIMENTAL_WEIGHT_SHARING', str),
Env('PT_HPU_WEIGHT_SHARING', str),
Env('RUNTIME_SCALE_PATCHING', str),

Expand All @@ -51,13 +51,13 @@ def get_user_flags():

def get_experimental_flags():
Copy link
Collaborator Author

@michalkuligowski michalkuligowski Nov 6, 2025

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

No need for thia method, it is covered by the metaclass

flags = [
Env('VLLM_PT_PROFILE', str),
Env('VLLM_PROFILE_PROMPT', str),
Env('VLLM_PROFILE_DECODE', str),
Env('VLLM_PROFILE_STEPS', list_of(int)),
Env('VLLM_DEFRAG_THRESHOLD', int),
Env('VLLM_DEFRAG_WITH_GRAPHS', boolean),
Env('VLLM_DEBUG', list_of(str), check=for_all(choice('steps', 'defrag', 'fwd'))),
ExperimentalEnv('VLLM_PT_PROFILE', str),
ExperimentalEnv('VLLM_PROFILE_PROMPT', str),
ExperimentalEnv('VLLM_PROFILE_DECODE', str),
ExperimentalEnv('VLLM_PROFILE_STEPS', list_of(int)),
ExperimentalEnv('VLLM_DEFRAG_THRESHOLD', int),
ExperimentalEnv('VLLM_DEFRAG_WITH_GRAPHS', boolean),
ExperimentalEnv('VLLM_DEBUG', list_of(str), check=for_all(choice('steps', 'defrag', 'fwd'))),
]
return to_dict(flags)

Expand Down
Loading