Bump evaluation reqs (#144)

* fix eval sanity check

* bump reqs

---------

Signed-off-by: Jack Luar <jluar@precisioninno.com>

Author: luarss

evaluation/auto_evaluation/eval_main.py

@@ -55,20 +55,31 @@ def __init__(self, base_url: str, dataset: str, reranker_base_url: str = ""):
 
     def sanity_check(self):
         cur_time = time.time()
+        if not os.path.exists(self.dataset):
+            raise ValueError("Dataset path does not exist")
         while time.time() - cur_time < RETRY_TIMEOUT:
-            if not requests.get(f"{self.base_url}/healthcheck").status_code == 200:
-                raise ValueError("Endpoint is not running")
-            if not os.path.exists(self.dataset):
-                raise ValueError("Dataset path does not exist")
-            if (
-                self.reranker_base_url
-                and not requests.get(
-                    f"{self.reranker_base_url}/healthcheck"
-                ).status_code
-                == 200
-            ):
-                raise ValueError("Reranker endpoint is not running")
-            time.sleep(RETRY_INTERVAL)
+            try:
+                if not requests.get(f"{self.base_url}/healthcheck").status_code == 200:
+                    print("Endpoint not ready, retrying...")
+                    time.sleep(RETRY_INTERVAL)
+                    continue
+                if (
+                    self.reranker_base_url
+                    and not requests.get(
+                        f"{self.reranker_base_url}/healthcheck"
+                    ).status_code
+                    == 200
+                ):
+                    print("Reranker endpoint not ready, retrying...")
+                    time.sleep(RETRY_INTERVAL)
+                    continue
+                # All checks passed
+                return
+            except requests.exceptions.RequestException:
+                print("Connection failed, retrying...")
+                time.sleep(RETRY_INTERVAL)
+                continue
+        raise ValueError("Sanity check failed after timeout")
 
     def evaluate(self, retriever: str):
         retrieval_tcs = []

evaluation/requirements-test.txt

@@ -1,5 +1,5 @@
 mypy==1.10.1
 ruff==0.5.1
-types-requests==2.32.0.20240622
+types-requests==2.32.0.20250602
 google-api-python-client-stubs==1.28.0
 types-tqdm==4.67.0.20241221

evaluation/requirements.txt

@@ -4,8 +4,7 @@ google-auth-httplib2==0.2.0
 google-auth-oauthlib==1.2.0
 gspread==6.1.2
 python-dotenv==1.0.1
-requests==2.32.3
-requests-oauthlib==2.0.0
+requests==2.32.4
 streamlit==1.37.0
 gspread==6.1.2
 deepeval==2.6.8