From 09bf71b4f9892be4bb83cf78e1697485a235bd2f Mon Sep 17 00:00:00 2001
From: Nicolas Mowen <nickmowen213@gmail.com>
Date: Thu, 30 Oct 2025 07:09:03 -0600
Subject: [PATCH] Adjustments to context sizing

---
 .../post/review_descriptions.py               | 36 +++++++++++--------
 frigate/genai/ollama.py                       |  2 +-
 2 files changed, 22 insertions(+), 16 deletions(-)

diff --git a/frigate/data_processing/post/review_descriptions.py b/frigate/data_processing/post/review_descriptions.py
index 35b6a55e6..b49e3018b 100644
--- a/frigate/data_processing/post/review_descriptions.py
+++ b/frigate/data_processing/post/review_descriptions.py
@@ -53,32 +53,38 @@ class ReviewDescriptionProcessor(PostProcessorApi):
     ) -> int:
         """Calculate optimal number of frames based on context size and image source.
 
-        Recordings (480p): ~250 tokens/image, capped at 20 frames
-        Previews (180p): ~100 tokens/image, capped at 20 frames
-        Targets 70-80% context utilization while keeping inference time reasonable.
+        Recordings (480p): ~500 tokens/image, capped at 20 frames
+        Previews (180p): ~170 tokens/image, capped at 20 frames
+        Targets 75% context utilization while keeping inference time reasonable.
         """
         context_size = self.genai_client.get_context_size()
 
         if image_source == ImageSourceEnum.recordings:
+            if context_size > 16000:
+                return 20
+            elif context_size > 14000:
+                return 18
+            elif context_size > 12000:
+                return 14
+            elif context_size > 10000:
+                return 10
+            elif context_size > 8000:
+                return 8
+            elif context_size > 6000:
+                return 6
+            else:
+                return 4
+        else:
             if context_size > 12000:
                 return 20
-            elif context_size > 10000:
-                return 18
+            elif context_size > 8000:
+                return 16
             elif context_size > 6000:
-                return 14
+                return 12
             elif context_size > 4000:
                 return 10
             else:
                 return 6
-        else:
-            if context_size > 10000:
-                return 20
-            elif context_size > 6000:
-                return 16
-            elif context_size > 4000:
-                return 12
-            else:
-                return 8
 
     def process_data(self, data, data_type):
         self.metrics.review_desc_dps.value = self.review_descs_dps.eps()
diff --git a/frigate/genai/ollama.py b/frigate/genai/ollama.py
index 374698b8e..9f9c8a750 100644
--- a/frigate/genai/ollama.py
+++ b/frigate/genai/ollama.py
@@ -68,7 +68,7 @@ class OllamaClient(GenAIClient):
                 f"Ollama tokens used: eval_count={result.get('eval_count')}, prompt_eval_count={result.get('prompt_eval_count')}"
             )
             return result["response"].strip()
-        except (TimeoutException, ResponseError) as e:
+        except (TimeoutException, ResponseError, ConnectionError) as e:
             logger.warning("Ollama returned an error: %s", str(e))
             return None