Skip to content
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

Fix p2p pushing in rpc_inference, support transformers 4.38.2 #563

Merged
merged 4 commits into from
Mar 17, 2024
Merged

Fix p2p pushing in rpc_inference, support transformers 4.38.2 #563

Show file tree
Hide file tree
Changes from 1 commit
Commits
Show all changes
4 commits
Select commit Hold shift + click to select a range
ac48d66
Forward step_metadata to fix p2p pushing in rpc_inference; found by @…
YingtongDou Mar 17, 2024
6dc7794
try bump transformers version
justheuristic Mar 17, 2024
6278395
support transformers 4.38.2
justheuristic Mar 17, 2024
18ec785
test cache_position safely
justheuristic Mar 17, 2024
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
4 changes: 2 additions & 2 deletions src/petals/server/block_functions.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -153,7 +153,7 @@ async def iterate_rpc_inference(
points: int,
quant_type: QuantType,
args_structure: Any = None,
) -> AsyncIterator[Tuple[Sequence[runtime_pb2.Tensor], bool]]:
) -> AsyncIterator[Tuple[Sequence[runtime_pb2.Tensor], bool, Dict]]:
assert len(cache_handles) == len(requested_backends)

prefix_length = 0
Expand Down Expand Up @@ -224,7 +224,7 @@ async def iterate_rpc_inference(
for result, proto in zip((hidden_states,), nested_flatten(requested_backends[-1].outputs_schema))
]
can_push = not has_prompts
yield output_tensors, can_push
yield output_tensors, can_push, step_metadata

# prepare for next step
prefix_length += length_increment
4 changes: 2 additions & 2 deletions src/petals/server/handler.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -171,7 +171,7 @@ async def rpc_inference(
requested_backends, batch_size=batch_size, max_length=max_length, timeout=alloc_timeout
) as cache_handles:
background_tasks = set()
async for output_tensors, can_push in iterate_rpc_inference(
async for output_tensors, can_push, step_metadata in iterate_rpc_inference(
requested_uids=requested_uids,
requested_backends=requested_backends,
active_adapter=self._get_active_adapter(metadata),
Expand All @@ -186,7 +186,7 @@ async def rpc_inference(
args_structure=args_structure,
):
if can_push:
task = asyncio.create_task(self._push_outputs(request, output_tensors[0], metadata))
task = asyncio.create_task(self._push_outputs(request, output_tensors[0], step_metadata))
background_tasks.add(task) # Keep reference until it is done to save it from GC
task.add_done_callback(background_tasks.discard)
yield runtime_pb2.ExpertResponse(tensors=output_tensors)
Expand Down
Loading