Skip to content

[Backend] Fix device mismatch for NLI model in AnswerPredictor #441

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

Open
Siddhazntx wants to merge 2 commits into
base: main
Choose a base branch
from
Open

[Backend] Fix device mismatch for NLI model in AnswerPredictor #441

Changes from all commits
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
68b44bc
fix: push NLI model and input tensors to device in AnswerPredictor
Siddhazntx Feb 19, 2026
13cad6e
Optimize NLI model with eval(), no_grad(), and device-aware inputs
Siddhazntx Mar 1, 2026
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
14 changes: 13 additions & 1 deletion backend/Generator/main.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -23,6 +23,9 @@
import fitz
import mammoth




class MCQGenerator:

def __init__(self):
Expand Down Expand Up @@ -251,6 +254,10 @@ def __init__(self):
self.nli_tokenizer = AutoTokenizer.from_pretrained(self.nli_model_name)
self.nli_model = AutoModelForSequenceClassification.from_pretrained(self.nli_model_name)

# Explicitly push the NLI model to the detected hardware (GPU or CPU)
self.nli_model.to(self.device)
self.nli_model.eval()

self.set_seed(42)

def set_seed(self, seed):
Expand Down Expand Up @@ -286,7 +293,8 @@ def predict_answer(self, payload):
torch.cuda.empty_cache()

return answers


@torch.no_grad()
def predict_boolean_answer(self, payload):
input_text = payload.get("input_text", "")
input_questions = payload.get("input_question", [])
Expand All @@ -296,6 +304,10 @@ def predict_boolean_answer(self, payload):
for question in input_questions:
hypothesis = question
inputs = self.nli_tokenizer.encode_plus(input_text, hypothesis, return_tensors="pt")

# Push the input tensors to the same device as the model
inputs = {key: value.to(self.device) for key, value in inputs.items()}

outputs = self.nli_model(**inputs)
logits = outputs.logits
probabilities = torch.softmax(logits, dim=1)
Expand Down