From 3e680db4f5469cdaeb4b2fc27f2d399568b39f3e Mon Sep 17 00:00:00 2001
From: SparshM8 <188700067+SparshM8@users.noreply.github.com>
Date: Fri, 21 Aug 2026 13:52:16 +0000
Subject: [PATCH] Optimize memory for 512MB free tier: tune ONNX session
 options and reduce TF-IDF features

---
 backend/app/services/ai_detection.py | 3 +++
 backend/app/services/plagiarism.py   | 2 +-
 2 files changed, 4 insertions(+), 1 deletion(-)

diff --git a/backend/app/services/ai_detection.py b/backend/app/services/ai_detection.py
index 2b5de419..f9c32e9a 100644
--- a/backend/app/services/ai_detection.py
+++ b/backend/app/services/ai_detection.py
@@ -233,6 +233,9 @@ def _onnx_session(model_path: str) -> "Optional[InferenceSession]":
     # concurrent inference instead of ballooning and triggering a fallback.
     sess_options.enable_mem_pattern = False
     sess_options.enable_cpu_mem_arena = False
+    # Further memory reduction: disable profiling and memory-heavy optimizations
+    sess_options.graph_optimization_level = ort.GraphOptimizationLevel.ORT_ENABLE_BASIC
+    sess_options.add_session_config_entry("session.use_device_allocator_for_initializers", "1")
     try:
         return ort.InferenceSession(str(onnx_path), sess_options, providers=providers)
     except Exception as exc:
diff --git a/backend/app/services/plagiarism.py b/backend/app/services/plagiarism.py
index f5603476..12a84737 100644
--- a/backend/app/services/plagiarism.py
+++ b/backend/app/services/plagiarism.py
@@ -66,7 +66,7 @@ def _load_vectorizer():
             from sklearn.feature_extraction.text import TfidfVectorizer
 
             _vectorizer = TfidfVectorizer(
-                max_features=20000,
+                max_features=10000,
                 min_df=2,
                 max_df=0.95,
                 sublinear_tf=True,