Add e5-mistral embedding model - step 3/3 (#988)

test/srt/models/test_embedding_models.py

@@ -0,0 +1,69 @@
+"""
+Copyright 2023-2024 SGLang Team
+Licensed under the Apache License, Version 2.0 (the "License");
+you may not use this file except in compliance with the License.
+You may obtain a copy of the License at
+
+    http://www.apache.org/licenses/LICENSE-2.0
+
+Unless required by applicable law or agreed to in writing, software
+distributed under the License is distributed on an "AS IS" BASIS,
+WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+See the License for the specific language governing permissions and
+limitations under the License.
+"""
+
+import unittest
+
+import torch
+
+from sglang.test.runners import DEFAULT_PROMPTS, HFRunner, SRTRunner
+from sglang.test.test_utils import get_similarities
+
+MODELS = [("intfloat/e5-mistral-7b-instruct", 1)]
+TORCH_DTYPES = [torch.float16]
+
+
+class TestEmbeddingModels(unittest.TestCase):
+
+    def assert_close_prefill_logits(
+        self,
+        prompts,
+        model_path,
+        tp_size,
+        torch_dtype,
+    ) -> None:
+        with HFRunner(
+            model_path, torch_dtype=torch_dtype, is_generation_model=False
+        ) as hf_runner:
+            hf_outputs = hf_runner.forward(prompts)
+
+        with SRTRunner(
+            model_path,
+            tp_size=tp_size,
+            torch_dtype=torch_dtype,
+            is_generation_model=False,
+        ) as srt_runner:
+            srt_outputs = srt_runner.forward(prompts)
+
+        for i in range(len(prompts)):
+            hf_logits = torch.Tensor(hf_outputs.embed_logits[i])
+            srt_logits = torch.Tensor(srt_outputs.embed_logits[i])
+
+            similarities = torch.tensor(get_similarities(hf_logits, srt_logits))
+
+            tolerance = 1e-2
+            assert torch.all(
+                abs(similarities - 1) < tolerance
+            ), f"embeddings not all close"
+
+    def test_prefill_logits(self):
+        for model, tp_size in MODELS:
+            for torch_dtype in TORCH_DTYPES:
+                self.assert_close_prefill_logits(
+                    DEFAULT_PROMPTS, model, tp_size, torch_dtype
+                )
+
+
+if __name__ == "__main__":
+    unittest.main(warnings="ignore")

test/srt/models/test_generation_models.py

@@ -3,7 +3,9 @@ Copyright 2023-2024 SGLang Team
 Licensed under the Apache License, Version 2.0 (the "License");
 you may not use this file except in compliance with the License.
 You may obtain a copy of the License at
+
     http://www.apache.org/licenses/LICENSE-2.0
+
 Unless required by applicable law or agreed to in writing, software
 distributed under the License is distributed on an "AS IS" BASIS,
 WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
@@ -33,7 +35,7 @@ class TestCausalModels(unittest.TestCase):
         torch_dtype,
     ) -> None:
         with HFRunner(
-            model_path, torch_dtype=torch_dtype, is_embedding_model=False
+            model_path, torch_dtype=torch_dtype, is_generation_model=True
         ) as hf_runner:
             hf_outputs = hf_runner.forward(prompts)
 
@@ -41,7 +43,7 @@ class TestCausalModels(unittest.TestCase):
             model_path,
             tp_size=tp_size,
             torch_dtype=torch_dtype,
-            is_embedding_model=False,
+            is_generation_model=True,
         ) as srt_runner:
             srt_outputs = srt_runner.forward(prompts)
 

test/srt/run_suite.py

@@ -10,7 +10,8 @@ suites = {
         "test_vision_openai_server.py",
         "test_chunked_prefill.py",
         "test_torch_compile.py",
-        "models/test_causal_models.py",
+        "models/test_generation_models.py",
+        "models/test_embedding_models.py",
         "sampling/penaltylib",
     ],
     "sampling/penaltylib": glob.glob(