fix circular import (#309)

XuhuiZhou · web-flow · commit 23ea384d22f2 · 2025-04-05T17:23:00.000-04:00
* fix circular import

* update generate

* update generate
diff --git a/examples/experiment_eval.py b/examples/experiment_eval.py
@@ -17,12 +17,12 @@
     EnvironmentProfile,
     EpisodeLog,
     EvaluationDimensionBuilder,
+    SotopiaDimensions,
 )
 from sotopia.envs.evaluators import (
     EvaluationForTwoAgents,
     EpisodeLLMEvaluator,
     RuleBasedTerminatedEvaluator,
-    SotopiaDimensions,
 )
 from sotopia.envs.parallel import ParallelSotopiaEnv
 from sotopia.messages import AgentAction, Observation
diff --git a/examples/fix_missing_episodes.py b/examples/fix_missing_episodes.py
@@ -13,7 +13,7 @@
 from sotopia.database.env_agent_combo_storage import (
     EnvAgentComboStorage,
 )
-from sotopia.database.logs import EpisodeLog
+from sotopia.database import EpisodeLog, SotopiaDimensions
 from sotopia.database.persistent_profile import (
     AgentProfile,
     EnvironmentProfile,
@@ -22,7 +22,6 @@
     EvaluationForTwoAgents,
     EpisodeLLMEvaluator,
     RuleBasedTerminatedEvaluator,
-    SotopiaDimensions,
 )
 from sotopia.envs.parallel import ParallelSotopiaEnv
 from sotopia.messages.message_classes import AgentAction, Observation
diff --git a/examples/fix_missing_episodes_with_tag.py b/examples/fix_missing_episodes_with_tag.py
@@ -29,7 +29,7 @@
 from sotopia.database.env_agent_combo_storage import (
     EnvAgentComboStorage,
 )
-from sotopia.database.logs import EpisodeLog
+from sotopia.database import EpisodeLog, SotopiaDimensions
 from sotopia.database.persistent_profile import (
     AgentProfile,
     EnvironmentProfile,
@@ -38,7 +38,6 @@
     EvaluationForTwoAgents,
     EpisodeLLMEvaluator,
     RuleBasedTerminatedEvaluator,
-    SotopiaDimensions,
 )
 from sotopia.envs.parallel import ParallelSotopiaEnv
 from sotopia.messages.message_classes import AgentAction, Observation
diff --git a/pyproject.toml b/pyproject.toml
@@ -22,7 +22,7 @@ dependencies = [
     "together>=0.2.4,<1.5.0",
     "pydantic>=2.5.0,<3.0.0",
     "hiredis>=3.0.0",
-    "litellm>=0.1.1",
+    "litellm>=1.65.0",
     "aact"
 ]
 
diff --git a/sotopia-chat/chat_server.py b/sotopia-chat/chat_server.py
@@ -25,8 +25,8 @@
 )
 from sotopia.envs.parallel import ParallelSotopiaEnv
 from sotopia.server import arun_one_episode
-
-from sotopia.envs.evaluators import SotopiaDimensions, EvaluationForTwoAgents
+from sotopia.envs.evaluators import EvaluationForTwoAgents
+from sotopia.database import SotopiaDimensions
 from sotopia.logging import FileHandler
 
 process = subprocess.Popen(
diff --git a/sotopia/api/fastapi_server.py b/sotopia/api/fastapi_server.py
@@ -20,13 +20,13 @@
     BaseEnvironmentProfile,
     BaseAgentProfile,
     BaseRelationshipProfile,
+    SotopiaDimensions,
 )
 from sotopia.envs.parallel import ParallelSotopiaEnv
 from sotopia.envs.evaluators import (
     RuleBasedTerminatedEvaluator,
     EpisodeLLMEvaluator,
     EvaluationForTwoAgents,
-    SotopiaDimensions,
 )
 from sotopia.server import arun_one_episode
 from sotopia.agents import LLMAgent, Agents
diff --git a/sotopia/cli/benchmark/benchmark.py b/sotopia/cli/benchmark/benchmark.py
@@ -21,13 +21,13 @@
     EnvAgentComboStorage,
     EnvironmentProfile,
     EpisodeLog,
+    SotopiaDimensions,
 )
 from sotopia.database.serialization import get_rewards_from_episode
 from sotopia.envs.evaluators import (
     EvaluationForTwoAgents,
     EpisodeLLMEvaluator,
     RuleBasedTerminatedEvaluator,
-    SotopiaDimensions,
 )
 from sotopia.envs.parallel import ParallelSotopiaEnv
 from sotopia.messages import AgentAction, Observation
diff --git a/sotopia/database/__init__.py b/sotopia/database/__init__.py
@@ -48,6 +48,9 @@
     BaseCustomEvaluationDimension,
     CustomEvaluationDimensionList,
     BaseCustomEvaluationDimensionList,
+    GoalDimension,
+    SotopiaDimensions,
+    SotopiaDimensionsPlus,
 )
 
 __all__ = [
@@ -91,6 +94,9 @@
     "CustomEvaluationDimensionList",
     "BaseCustomEvaluationDimensionList",
     "NonStreamingSimulationStatus",
+    "GoalDimension",
+    "SotopiaDimensions",
+    "SotopiaDimensionsPlus",
 ]
 
 InheritedJsonModel = TypeVar("InheritedJsonModel", bound="JsonModel")
diff --git a/sotopia/database/evaluation_dimensions.py b/sotopia/database/evaluation_dimensions.py
diff --git a/sotopia/envs/__init__.py b/sotopia/envs/__init__.py
@@ -1,4 +1,3 @@
 from .parallel import ParallelSotopiaEnv
-from .evaluators import SotopiaDimensions
 
-__all__ = ["ParallelSotopiaEnv", "SotopiaDimensions"]
+__all__ = ["ParallelSotopiaEnv"]
diff --git a/sotopia/envs/evaluators.py b/sotopia/envs/evaluators.py
diff --git a/sotopia/experimental/agents/evaluators.py b/sotopia/experimental/agents/evaluators.py
@@ -9,7 +9,7 @@
 from typing import Generic, TypeVar, Type, Any
 from pydantic import BaseModel, Field
 
-from sotopia.envs.evaluators import GoalDimension
+from sotopia.database import GoalDimension
 from sotopia.generation_utils.generate import agenerate
 from sotopia.generation_utils.output_parsers import PydanticOutputParser
 
diff --git a/sotopia/generation_utils/generate.py b/sotopia/generation_utils/generate.py
@@ -1,6 +1,10 @@
 import logging
 import os
 from litellm import acompletion
+from litellm.utils import supports_response_schema
+from litellm.litellm_core_utils.get_supported_openai_params import (
+    get_supported_openai_params,
+)
 from typing import cast
 
 import gin
@@ -115,11 +119,15 @@ async def agenerate(
         api_key = None
 
     if structured_output:
-        assert (
-            model_name.startswith("gpt-4o")
-            or model_name.startswith("openai/")
-            or model_name.startswith("o1")
-        ), "Structured output is only supported in limited models"
+        if not base_url:
+            params = get_supported_openai_params(model=model_name)
+            assert params is not None
+            assert (
+                "response_format" in params
+            ), "response_format is not supported in this model"
+            assert supports_response_schema(
+                model=model_name
+            ), "response_schema is not supported in this model"
         messages = [{"role": "user", "content": template}]
 
         assert isinstance(
diff --git a/sotopia/server.py b/sotopia/server.py
@@ -15,13 +15,12 @@
     ScriptWritingAgent,
 )
 from sotopia.agents.base_agent import BaseAgent
-from sotopia.database import EpisodeLog, NonStreamingSimulationStatus
+from sotopia.database import EpisodeLog, NonStreamingSimulationStatus, SotopiaDimensions
 from sotopia.envs import ParallelSotopiaEnv
 from sotopia.envs.evaluators import (
     EvaluationForTwoAgents,
     EpisodeLLMEvaluator,
     RuleBasedTerminatedEvaluator,
-    SotopiaDimensions,
     unweighted_aggregate_evaluate,
 )
 from sotopia.generation_utils.generate import agenerate_script
diff --git a/tests/integration/test_benchmark.py b/tests/integration/test_benchmark.py
@@ -13,7 +13,7 @@
 from unittest import mock
 from unittest.mock import create_autospec
 from sotopia.cli.benchmark.benchmark import initialize_benchmark_combo
-from sotopia.database import EnvAgentComboStorage
+from sotopia.database import EnvAgentComboStorage, SotopiaDimensions
 import pytest
 
 from sotopia.envs.parallel import ParallelSotopiaEnv
@@ -25,7 +25,6 @@
     EvaluationForTwoAgents,
     EpisodeLLMEvaluator,
     RuleBasedTerminatedEvaluator,
-    SotopiaDimensions,
 )
 from sotopia.agents import LLMAgent
 
diff --git a/uv.lock b/uv.lock

Original file line number	Diff line number	Diff line change
`@@ -17,12 +17,12 @@`
`17`	`17`	`EnvironmentProfile,`
`18`	`18`	`EpisodeLog,`
`19`	`19`	`EvaluationDimensionBuilder,`
	`20`	`+ SotopiaDimensions,`
`20`	`21`	`)`
`21`	`22`	`from sotopia.envs.evaluators import (`
`22`	`23`	`EvaluationForTwoAgents,`
`23`	`24`	`EpisodeLLMEvaluator,`
`24`	`25`	`RuleBasedTerminatedEvaluator,`
`25`		`- SotopiaDimensions,`
`26`	`26`	`)`
`27`	`27`	`from sotopia.envs.parallel import ParallelSotopiaEnv`
`28`	`28`	`from sotopia.messages import AgentAction, Observation`
Original file line number	Diff line number	Diff line change
`@@ -22,7 +22,7 @@ dependencies = [`
`22`	`22`	`"together>=0.2.4,<1.5.0",`
`23`	`23`	`"pydantic>=2.5.0,<3.0.0",`
`24`	`24`	`"hiredis>=3.0.0",`
`25`		`- "litellm>=0.1.1",`
	`25`	`+ "litellm>=1.65.0",`
`26`	`26`	`"aact"`
`27`	`27`	`]`
`28`	`28`