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

Improve recovery from KV cache starvation #403

Merged
merged 3 commits into from
Feb 2, 2024
Merged

Improve recovery from KV cache starvation #403

Changes from all commits
Commits
Show all changes
3 commits
Select commit Hold shift + click to select a range
ad51254
improve deadlock recovery
tohtana Feb 2, 2024
052a5f1
Merge branch 'main' into tohtana/deadlock_recovery
mrwyattii Feb 2, 2024
21f4681
add message to assersion
tohtana Feb 2, 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
49 changes: 36 additions & 13 deletions mii/batching/ragged_batching.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -273,9 +273,10 @@ def _schedule_prompts(self, requests: List[Request]) -> None:

max_blocks = free_blocks - self.scheduled_req_blocks

# Check capacity to mitigate the deadlock risk
# We don't schedule requests when we find that a prompt is too long to fit to the KV cache
if len(r.input_tokens) > 1:
# When the KV cache is out of capacity, we release KV cache blocks for a request.
# However, we can immediately schedule the request again if we split the request.
# So we make sure that we have capacity for the entire prompt (+tokens already generated).
req_tokens, _ = self.inference_engine.query(r.uid, len(r.input_tokens), max_blocks)
if req_tokens < len(r.input_tokens):
break
Expand Down Expand Up @@ -329,9 +330,6 @@ def schedule_requests(self) -> None:
self._schedule_prompts(prompt_reqs)

if len(self.buffer) > 0 and len(self.scheduled_requests) == 0:
print(
"Deadlock detected. Resetting KV cache and recomputing requests. Consider limiting number of concurrent requests or decreasing max lengths of prompts/generations."
)
self.scheduled_requests = RequestBatch()
self.reset_request_status()
else:
Expand All @@ -353,20 +351,45 @@ def _queue_flush_request(self, uid: int) -> None:
))

def reset_request_status(self):
## Get the last request that consumes KV cache
last_r = None
for r in self.buffer:
if r.seq_length > 0:
self._queue_flush_request(r.uid)
last_r = r
assert last_r is not None, "Function to clear the KV cache is invoked, but no request consumes KV cache"

## Schedule flushing r
self.scheduled_requests.append(
Request(
tid=None,
uid=last_r.uid,
input_tokens=None,
prompt_tokens=None,
seq_length=None,
last_in_prompt=None,
post_processing=None,
generate_params=None,
))

## Rebuild the request
new_req = copy.copy(last_r)
new_req.prompt_tokens = new_req.input_tokens = torch.concat(
[last_r.prompt_tokens] + [t.unsqueeze(0) for t in last_r.generated_tokens])
new_req.seq_length = 0
new_req.max_new_tokens = last_r.max_new_tokens - len(last_r.generated_tokens)
new_req.clear_generated_token()

## Remove the requests from buffer and queue
new_buffer = deque()
for r in self.buffer:
new_req = copy.copy(r)
new_req.prompt_tokens = new_req.input_tokens = torch.concat(
[r.prompt_tokens] + [t.unsqueeze(0) for t in r.generated_tokens])
new_req.seq_length = 0
new_req.max_new_tokens = r.max_new_tokens - len(r.generated_tokens)
new_req.clear_generated_token()
new_buffer.append(new_req)
if r.uid != last_r.uid:
new_buffer.append(r)

while not self.request_queue.empty():
r = self.request_queue.get_nowait()
if r.uid != last_r.uid:
new_buffer.append(r)
new_buffer.append(new_req)
self.buffer = new_buffer

def make_request(self,
Expand Down
Loading