[CI/Build] Replace math.isclose with pytest.approx (#18703)

Signed-off-by: DarkLight1337 <tlleungac@connect.ust.hk>

tests/entrypoints/openai/correctness/test_mteb.py

@@ -1,5 +1,4 @@
 # SPDX-License-Identifier: Apache-2.0
-import math
 import os
 
 import pytest
@@ -39,4 +38,4 @@ def test_mteb(server):
     print("SentenceTransformer main score: ", st_main_score)
     print("Difference: ", st_main_score - vllm_main_score)
 
-    assert math.isclose(st_main_score, vllm_main_score, rel_tol=1e-4)
+    assert st_main_score == pytest.approx(vllm_main_score, rel=1e-4)

tests/entrypoints/openai/test_score.py

@@ -1,6 +1,4 @@
 # SPDX-License-Identifier: Apache-2.0
-
-import math
 from typing import Any
 
 import pytest
@@ -92,7 +90,7 @@ class TestModel:
         hf_outputs = run_transformers(runner, model, text_pairs)
 
         for i in range(len(vllm_outputs)):
-            assert math.isclose(hf_outputs[i], vllm_outputs[i], rel_tol=0.01)
+            assert hf_outputs[i] == pytest.approx(vllm_outputs[i], rel=0.01)
 
     def test_text_1_list_text_2_list(self, server: RemoteOpenAIServer,
                                      model: dict[str, Any], runner):
@@ -124,7 +122,7 @@ class TestModel:
         hf_outputs = run_transformers(runner, model, text_pairs)
 
         for i in range(len(vllm_outputs)):
-            assert math.isclose(hf_outputs[i], vllm_outputs[i], rel_tol=0.01)
+            assert hf_outputs[i] == pytest.approx(vllm_outputs[i], rel=0.01)
 
     def test_text_1_str_text_2_str(self, server: RemoteOpenAIServer,
                                    model: dict[str, Any], runner):
@@ -150,7 +148,7 @@ class TestModel:
         hf_outputs = run_transformers(runner, model, text_pairs)
 
         for i in range(len(vllm_outputs)):
-            assert math.isclose(hf_outputs[i], vllm_outputs[i], rel_tol=0.01)
+            assert hf_outputs[i] == pytest.approx(vllm_outputs[i], rel=0.01)
 
     def test_score_max_model_len(self, server: RemoteOpenAIServer,
                                  model: dict[str, Any]):

tests/models/language/pooling/mteb_utils.py

@@ -1,5 +1,4 @@
 # SPDX-License-Identifier: Apache-2.0
-import math
 from collections.abc import Sequence
 
 import mteb
@@ -115,4 +114,4 @@ def mteb_test_embed_models(hf_runner,
     print("SentenceTransformer:", model_dtype, st_main_score)
     print("Difference:", st_main_score - vllm_main_score)
 
-    assert math.isclose(st_main_score, vllm_main_score, rel_tol=MTEB_EMBED_TOL)
+    assert st_main_score == pytest.approx(vllm_main_score, rel=MTEB_EMBED_TOL)

tests/models/language/pooling/test_gritlm.py

@@ -2,7 +2,6 @@
 from __future__ import annotations
 
 import importlib.util
-import math
 from array import array
 
 import openai
@@ -104,16 +103,16 @@ def get_test_data():
 
 def validate_embed_output(q_rep: list[list[float]], d_rep: list[list[float]]):
     cosine_sim_q0_d0 = 1 - cosine(q_rep[0], d_rep[0])
-    assert math.isclose(cosine_sim_q0_d0, 0.609, abs_tol=0.001)
+    assert cosine_sim_q0_d0 == pytest.approx(0.609, abs=0.001)
 
     cosine_sim_q0_d1 = 1 - cosine(q_rep[0], d_rep[1])
-    assert math.isclose(cosine_sim_q0_d1, 0.101, abs_tol=0.001)
+    assert cosine_sim_q0_d1 == pytest.approx(0.101, abs=0.001)
 
     cosine_sim_q1_d0 = 1 - cosine(q_rep[1], d_rep[0])
-    assert math.isclose(cosine_sim_q1_d0, 0.120, abs_tol=0.001)
+    assert cosine_sim_q1_d0 == pytest.approx(0.120, abs=0.001)
 
     cosine_sim_q1_d1 = 1 - cosine(q_rep[1], d_rep[1])
-    assert math.isclose(cosine_sim_q1_d1, 0.534, abs_tol=0.001)
+    assert cosine_sim_q1_d1 == pytest.approx(0.534, abs=0.001)
 
 
 def test_gritlm_offline_embedding(vllm_runner):

tests/models/language/pooling/test_jina.py

@@ -1,6 +1,4 @@
 # SPDX-License-Identifier: Apache-2.0
-import math
-
 import pytest
 
 from vllm import PoolingParams
@@ -60,7 +58,7 @@ def test_llm_1_to_1(vllm_runner, hf_runner, model_name, dtype: str):
     assert len(vllm_outputs) == 1
     assert len(hf_outputs) == 1
 
-    assert math.isclose(hf_outputs[0], vllm_outputs[0], rel_tol=0.01)
+    assert hf_outputs[0] == pytest.approx(vllm_outputs[0], rel=0.01)
 
 
 @pytest.mark.parametrize("dtype", ["half"])
@@ -78,8 +76,8 @@ def test_llm_1_to_N(vllm_runner, hf_runner, model_name, dtype: str):
     assert len(vllm_outputs) == 10
     assert len(hf_outputs) == 10
 
-    assert math.isclose(hf_outputs[0], vllm_outputs[0], rel_tol=0.01)
-    assert math.isclose(hf_outputs[1], vllm_outputs[1], rel_tol=0.01)
+    assert hf_outputs[0] == pytest.approx(vllm_outputs[0], rel=0.01)
+    assert hf_outputs[1] == pytest.approx(vllm_outputs[1], rel=0.01)
 
 
 @pytest.fixture(scope="module", params=EMBEDDING_MODELS)

tests/models/language/pooling/test_scoring.py

@@ -1,6 +1,4 @@
 # SPDX-License-Identifier: Apache-2.0
-import math
-
 import pytest
 import torch
 import torch.nn.functional as F
@@ -45,7 +43,7 @@ def test_cross_encoder_1_to_1(vllm_runner, hf_runner, model_name):
     assert len(vllm_outputs) == 1
     assert len(hf_outputs) == 1
 
-    assert math.isclose(hf_outputs[0], vllm_outputs[0], rel_tol=0.01)
+    assert hf_outputs[0] == pytest.approx(vllm_outputs[0], rel=0.01)
 
 
 def test_cross_encoder_1_to_N(vllm_runner, hf_runner, model_name):
@@ -64,8 +62,8 @@ def test_cross_encoder_1_to_N(vllm_runner, hf_runner, model_name):
     assert len(vllm_outputs) == 2
     assert len(hf_outputs) == 2
 
-    assert math.isclose(hf_outputs[0], vllm_outputs[0], rel_tol=0.01)
-    assert math.isclose(hf_outputs[1], vllm_outputs[1], rel_tol=0.01)
+    assert hf_outputs[0] == pytest.approx(vllm_outputs[0], rel=0.01)
+    assert hf_outputs[1] == pytest.approx(vllm_outputs[1], rel=0.01)
 
 
 def test_cross_encoder_N_to_N(vllm_runner, hf_runner, model_name):
@@ -84,8 +82,8 @@ def test_cross_encoder_N_to_N(vllm_runner, hf_runner, model_name):
     assert len(vllm_outputs) == 2
     assert len(hf_outputs) == 2
 
-    assert math.isclose(hf_outputs[0], vllm_outputs[0], rel_tol=0.01)
-    assert math.isclose(hf_outputs[1], vllm_outputs[1], rel_tol=0.01)
+    assert hf_outputs[0] == pytest.approx(vllm_outputs[0], rel=0.01)
+    assert hf_outputs[1] == pytest.approx(vllm_outputs[1], rel=0.01)
 
 
 @pytest.fixture(scope="module", params=EMBEDDING_MODELS)
@@ -112,7 +110,7 @@ def test_embedding_1_to_1(vllm_runner, hf_runner, emb_model_name):
     assert len(vllm_outputs) == 1
     assert len(hf_outputs) == 1
 
-    assert math.isclose(hf_outputs[0], vllm_outputs[0], rel_tol=0.01)
+    assert hf_outputs[0] == pytest.approx(vllm_outputs[0], rel=0.01)
 
 
 def test_embedding_1_to_N(vllm_runner, hf_runner, emb_model_name):
@@ -140,8 +138,8 @@ def test_embedding_1_to_N(vllm_runner, hf_runner, emb_model_name):
     assert len(vllm_outputs) == 2
     assert len(hf_outputs) == 2
 
-    assert math.isclose(hf_outputs[0], vllm_outputs[0], rel_tol=0.01)
-    assert math.isclose(hf_outputs[1], vllm_outputs[1], rel_tol=0.01)
+    assert hf_outputs[0] == pytest.approx(vllm_outputs[0], rel=0.01)
+    assert hf_outputs[1] == pytest.approx(vllm_outputs[1], rel=0.01)
 
 
 def test_embedding_N_to_N(vllm_runner, hf_runner, emb_model_name):
@@ -169,5 +167,5 @@ def test_embedding_N_to_N(vllm_runner, hf_runner, emb_model_name):
     assert len(vllm_outputs) == 2
     assert len(hf_outputs) == 2
 
-    assert math.isclose(hf_outputs[0], vllm_outputs[0], rel_tol=0.01)
-    assert math.isclose(hf_outputs[1], vllm_outputs[1], rel_tol=0.01)
+    assert hf_outputs[0] == pytest.approx(vllm_outputs[0], rel=0.01)
+    assert hf_outputs[1] == pytest.approx(vllm_outputs[1], rel=0.01)