Skip to content

[SB][CB] mark input tensors refactoring #198

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

Closed
wants to merge 2 commits into from
Closed

[SB][CB] mark input tensors refactoring #198

Changes from all commits
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
820f053
refactoring tensor marking into _mark_input_tensors as in static batc…
yannicks1 Jun 3, 2025
e6154be
[SB] update states before returning empty model runner (as done for CB)
yannicks1 Jun 3, 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
75 changes: 38 additions & 37 deletions vllm_spyre/v1/worker/spyre_model_runner.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -432,6 +432,7 @@ def execute_model(

t0 = time.time()

self._update_states(scheduler_output)
# TODO: change to EMPTY_MODEL_RUNNER_OUTPUT, right now this
# will be a breaking change, or clumsy to make retrocompatible
# with conditional import
Expand All @@ -446,8 +447,6 @@ def execute_model(
prompt_logprobs_dict={},
)

self._update_states(scheduler_output)

model_input = self.prepare_model_input(scheduler_output)
self._mark_input_tensors(model_input)

Expand Down Expand Up @@ -916,41 +915,7 @@ def execute_model(
)

model_input = self.prepare_model_input(scheduler_output)

# Marking dimensions static/dynamic
if model_input.is_prompt:

# batch static (batch size 1)
torch._dynamo.mark_static(model_input.input_tokens, 0)
torch._dynamo.mark_static(model_input.slot_mapping, 0)
torch._dynamo.mark_static(model_input.input_positions, 0)
torch._dynamo.mark_static(model_input.input_masks, 0)

# sequence dynamic
torch._dynamo.mark_dynamic(model_input.input_tokens, 1)
torch._dynamo.mark_dynamic(model_input.slot_mapping, 1)
torch._dynamo.mark_dynamic(model_input.input_positions, 1)
torch._dynamo.mark_dynamic(model_input.input_masks, 2)
torch._dynamo.mark_dynamic(model_input.input_masks, 3)

# decode
else:
# mask is no longer used here

# batch dynamic
torch._dynamo.mark_dynamic(model_input.input_tokens, 0)
torch._dynamo.mark_dynamic(model_input.block_table, 0)
torch._dynamo.mark_dynamic(model_input.slot_mapping, 0)
torch._dynamo.mark_dynamic(model_input.input_positions, 0)
torch._dynamo.mark_dynamic(model_input.current_tkv_mask, 0)
torch._dynamo.mark_dynamic(model_input.left_padded_prompt_mask, 0)

# sequence
torch._dynamo.mark_static(model_input.input_tokens, 1) # always 1
torch._dynamo.mark_dynamic(model_input.block_table, 1)
torch._dynamo.mark_static(model_input.slot_mapping, 1) # always 1
torch._dynamo.mark_static(model_input.input_positions,
1) # always 1
self._mark_input_tensors(model_input)
Copy link
Collaborator

@prashantgupta24 prashantgupta24 Jun 3, 2025

Choose a reason for hiding this comment

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

do we really need this function to be using self? It only interacts with the model_input param

Copy link
Collaborator Author

@yannicks1 yannicks1 Jun 4, 2025

Choose a reason for hiding this comment

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

you are totally right, I copied it from the static batching code, where it also uses self which can be removed. FYI: @wallashss


# Execute the model
hidden_states = self.model(
Expand Down Expand Up @@ -1001,3 +966,39 @@ def execute_model(
tkv=self.tkv,
)
return model_output

def _mark_input_tensors(self, model_input: ModelForwardInputs) -> None:
# Marking dimensions static/dynamic
if model_input.is_prompt:

# batch static (batch size 1)
torch._dynamo.mark_static(model_input.input_tokens, 0)
torch._dynamo.mark_static(model_input.slot_mapping, 0)
torch._dynamo.mark_static(model_input.input_positions, 0)
torch._dynamo.mark_static(model_input.input_masks, 0)

# sequence dynamic
torch._dynamo.mark_dynamic(model_input.input_tokens, 1)
torch._dynamo.mark_dynamic(model_input.slot_mapping, 1)
torch._dynamo.mark_dynamic(model_input.input_positions, 1)
torch._dynamo.mark_dynamic(model_input.input_masks, 2)
torch._dynamo.mark_dynamic(model_input.input_masks, 3)

# decode
else:
# mask is no longer used here

# batch dynamic
torch._dynamo.mark_dynamic(model_input.input_tokens, 0)
torch._dynamo.mark_dynamic(model_input.block_table, 0)
torch._dynamo.mark_dynamic(model_input.slot_mapping, 0)
torch._dynamo.mark_dynamic(model_input.input_positions, 0)
torch._dynamo.mark_dynamic(model_input.current_tkv_mask, 0)
torch._dynamo.mark_dynamic(model_input.left_padded_prompt_mask, 0)

# sequence
torch._dynamo.mark_static(model_input.input_tokens, 1) # always 1
torch._dynamo.mark_dynamic(model_input.block_table, 1)
torch._dynamo.mark_static(model_input.slot_mapping, 1) # always 1
torch._dynamo.mark_static(model_input.input_positions,
1) # always 1