Fix: 8192 ctx window + trim safety context to 3000 chars

safety_checker.py

@@ -110,7 +110,7 @@ def run_safety_check(context: str, call_model_fn) -> SafetyReport:
         "<|im_start|>system\n"
         "Extract medications and allergies from clinical documents. Output ONLY valid JSON.<|im_end|>\n"
         "<|im_start|>user\n"
-        + EXTRACTION_PROMPT.format(context=context[:5000])
+        + EXTRACTION_PROMPT.format(context=context[:3000])
         + "\n<|im_end|>\n<|im_start|>assistant\n"
     )