Avoid direct comparison of floating point numbers (#21002)

Signed-off-by: Max de Bayser <mbayser@br.ibm.com>

Author: maxdebayser

tests/entrypoints/openai/test_classification.py

@@ -176,4 +176,8 @@ async def test_invocations(server: RemoteOpenAIServer):
     invocation_output = invocation_response.json()
 
     assert classification_output.keys() == invocation_output.keys()
-    assert classification_output["data"] == invocation_output["data"]
+    for classification_data, invocation_data in zip(
+            classification_output["data"], invocation_output["data"]):
+        assert classification_data.keys() == invocation_data.keys()
+        assert classification_data["probs"] == pytest.approx(
+            invocation_data["probs"], rel=0.01)

tests/entrypoints/openai/test_embedding.py

@@ -14,6 +14,7 @@
 
 from ...models.language.pooling.embed_utils import (
     run_embedding_correctness_test)
+from ...models.utils import check_embeddings_close
 from ...utils import RemoteOpenAIServer
 
 MODEL_NAME = "intfloat/multilingual-e5-small"
@@ -321,7 +322,13 @@ async def test_invocations(server: RemoteOpenAIServer,
     invocation_output = invocation_response.json()
 
     assert completion_output.keys() == invocation_output.keys()
-    assert completion_output["data"] == invocation_output["data"]
+    for completion_data, invocation_data in zip(completion_output["data"],
+                                                invocation_output["data"]):
+        assert completion_data.keys() == invocation_data.keys()
+        check_embeddings_close(embeddings_0_lst=[completion_data["embedding"]],
+                               embeddings_1_lst=[invocation_data["embedding"]],
+                               name_0="completion",
+                               name_1="invocation")
 
 
 @pytest.mark.asyncio
@@ -355,4 +362,10 @@ async def test_invocations_conversation(server: RemoteOpenAIServer):
     invocation_output = invocation_response.json()
 
     assert chat_output.keys() == invocation_output.keys()
-    assert chat_output["data"] == invocation_output["data"]
+    for chat_data, invocation_data in zip(chat_output["data"],
+                                          invocation_output["data"]):
+        assert chat_data.keys() == invocation_data.keys()
+        check_embeddings_close(embeddings_0_lst=[chat_data["embedding"]],
+                               embeddings_1_lst=[invocation_data["embedding"]],
+                               name_0="chat",
+                               name_1="invocation")

tests/entrypoints/openai/test_pooling.py

@@ -281,7 +281,13 @@ async def test_invocations(server: RemoteOpenAIServer):
     invocation_output = invocation_response.json()
 
     assert completion_output.keys() == invocation_output.keys()
-    assert completion_output["data"] == invocation_output["data"]
+    for completion_data, invocation_data in zip(completion_output["data"],
+                                                invocation_output["data"]):
+        assert completion_data.keys() == invocation_data.keys()
+        check_embeddings_close(embeddings_0_lst=completion_data["data"],
+                               embeddings_1_lst=invocation_data["data"],
+                               name_0="completion",
+                               name_1="invocation")
 
 
 @pytest.mark.asyncio
@@ -314,4 +320,10 @@ async def test_invocations_conversation(server: RemoteOpenAIServer):
     invocation_output = invocation_response.json()
 
     assert chat_output.keys() == invocation_output.keys()
-    assert chat_output["data"] == invocation_output["data"]
+    for chat_data, invocation_data in zip(chat_output["data"],
+                                          invocation_output["data"]):
+        assert chat_data.keys() == invocation_data.keys()
+        check_embeddings_close(embeddings_0_lst=chat_data["data"],
+                               embeddings_1_lst=invocation_data["data"],
+                               name_0="chat",
+                               name_1="invocation")

tests/entrypoints/openai/test_rerank.py

@@ -120,4 +120,8 @@ def test_invocations(server: RemoteOpenAIServer):
     invocation_output = invocation_response.json()
 
     assert rerank_output.keys() == invocation_output.keys()
-    assert rerank_output["results"] == invocation_output["results"]
+    for rerank_result, invocations_result in zip(rerank_output["results"],
+                                                 invocation_output["results"]):
+        assert rerank_result.keys() == invocations_result.keys()
+        assert rerank_result["relevance_score"] == pytest.approx(
+            invocations_result["relevance_score"], rel=0.01)

tests/entrypoints/openai/test_score.py

@@ -215,4 +215,8 @@ def test_invocations(self, server: RemoteOpenAIServer, model: dict[str,
         invocation_output = invocation_response.json()
 
         assert score_output.keys() == invocation_output.keys()
-        assert score_output["data"] == invocation_output["data"]
+        for score_data, invocation_data in zip(score_output["data"],
+                                               invocation_output["data"]):
+            assert score_data.keys() == invocation_data.keys()
+            assert score_data["score"] == pytest.approx(
+                invocation_data["score"], rel=0.01)