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

Jump to bottom

fix: Reduce GPU to CPU copies at inference #3127

Merged
merged 5 commits into from
Sep 7, 2022
Merged

fix: Reduce GPU to CPU copies at inference #3127

Changes from all commits
Commits
Show all changes
5 commits
Select commit Hold shift + click to select a range
da8e5e6
Send matrix from gpu to cpu once instead of individual elements
sjrl Aug 31, 2022
fb296a0
Added pydantic pin until it is merged into main
sjrl Aug 31, 2022
2a97b92
Merge branch 'main' of github.com:deepset-ai/haystack into issue-2926
sjrl Aug 31, 2022
bd7d18f
Moved location of if statement so it would be triggered only when
sjrl Sep 5, 2022
7d4e326
Merge branch 'main' of github.com:deepset-ai/haystack into issue-2926
sjrl Sep 5, 2022
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
25 changes: 14 additions & 11 deletions haystack/modeling/model/prediction_head.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -495,6 +495,8 @@ def logits_to_preds(
sorted_candidates = torch.cat((start_indices, end_indices), dim=2)

# Get the n_best candidate answers for each sample
sorted_candidates = sorted_candidates.cpu().numpy()
start_end_matrix = start_end_matrix.cpu().numpy()
for sample_idx in range(batch_size):
sample_top_n = self.get_top_candidates(
sorted_candidates[sample_idx],
Expand All @@ -519,24 +521,21 @@ def get_top_candidates(self, sorted_candidates, start_end_matrix, sample_idx: in
start_idx_candidates = set()
end_idx_candidates = set()

start_matrix_softmax_start = torch.softmax(start_matrix[:, 0], dim=-1)
end_matrix_softmax_end = torch.softmax(end_matrix[0, :], dim=-1)
start_matrix_softmax_start = torch.softmax(start_matrix[:, 0], dim=-1).cpu().numpy()
end_matrix_softmax_end = torch.softmax(end_matrix[0, :], dim=-1).cpu().numpy()
# Iterate over all candidates and break when we have all our n_best candidates
for candidate_idx in range(n_candidates):
if len(top_candidates) == self.n_best_per_sample:
break

# Retrieve candidate's indices
start_idx = sorted_candidates[candidate_idx, 0].item()
end_idx = sorted_candidates[candidate_idx, 1].item()
start_idx = sorted_candidates[candidate_idx, 0]
end_idx = sorted_candidates[candidate_idx, 1]
# Ignore no_answer scores which will be extracted later in this method
if start_idx == 0 and end_idx == 0:
continue
if self.duplicate_filtering > -1 and (start_idx in start_idx_candidates or end_idx in end_idx_candidates):
continue
score = start_end_matrix[start_idx, end_idx].item()
score = start_end_matrix[start_idx, end_idx]
confidence = (
(start_matrix_softmax_start[start_idx].item() + end_matrix_softmax_end[end_idx].item()) / 2
(start_matrix_softmax_start[start_idx] + end_matrix_softmax_end[end_idx]) / 2
if score > -500
else np.exp(score / 10) # disqualify answers according to scores in logits_to_preds()
)
Expand All @@ -559,8 +558,12 @@ def get_top_candidates(self, sorted_candidates, start_end_matrix, sample_idx: in
end_idx_candidates.add(end_idx + i)
end_idx_candidates.add(end_idx - i)

no_answer_score = start_end_matrix[0, 0].item()
no_answer_confidence = (start_matrix_softmax_start[0].item() + end_matrix_softmax_end[0].item()) / 2
# Only check if we have enough candidates after adding new candidate to the list
if len(top_candidates) == self.n_best_per_sample:
break

no_answer_score = start_end_matrix[0, 0]
no_answer_confidence = (start_matrix_softmax_start[0] + end_matrix_softmax_end[0]) / 2
top_candidates.append(
QACandidate(
offset_answer_start=0,
Expand Down