Adding a devcontainer with Redis and a Ollama server (#217)

* minimal demo example of running custom model

* devcontainer setup and example

* remove default_bad_process_model to allow using custom model entirely

* improve the demo to show parallel execution

* CI: update tests trigger from pull request target to pull request

* fix mypy errors

* adding stubs to pyproject.toml

* poetry lock

* install all extras in the devcontainer start script

* add dev containers instruction

* add a link to command palette wiki

* adding instructions for dev containers

.devcontainer/devcontainer.json

@@ -0,0 +1,26 @@
+// For format details, see https://aka.ms/devcontainer.json. For config options, see the
+// README at: https://github.com/devcontainers/templates/tree/main/src/python
+{
+	"name": "Sotopia Place",
+	// Or use a Dockerfile or Docker Compose file. More info: https://containers.dev/guide/dockerfile
+	"image": "mcr.microsoft.com/devcontainers/python:1-3.12-bullseye",
+	"features": {
+		"ghcr.io/itsmechlark/features/redis-server:1": {},
+		"ghcr.io/prulloac/devcontainer-features/ollama:1": {}
+	},
+
+	// Features to add to the dev container. More info: https://containers.dev/features.
+	// "features": {},
+
+	// Use 'forwardPorts' to make a list of ports inside the container available locally.
+	// "forwardPorts": [],
+
+	// Use 'postCreateCommand' to run commands after the container is created.
+	"postCreateCommand": "pipx install poetry; poetry install --all-extras; ollama pull llama3.2:1b"
+
+	// Configure tool-specific properties.
+	// "customizations": {},
+
+	// Uncomment to connect as root instead. More info: https://aka.ms/dev-containers-non-root.
+	// "remoteUser": "root"
+}

.github/workflows/tests.yml

@@ -10,7 +10,7 @@ on:
       - main
       - release
       - dev
-  pull_request_target:
+  pull_request:
     branches:
       - main
       - release

docs/pages/contribution/contribution.md

@@ -60,7 +60,21 @@ git push origin main
 
 ### 4. Set up the Development Environment
 
-?? @ProKil
+We recommend using Dev Containers to set up your development environment, but note that Docker is required for install Dev Containers.
+
+#### Using VSCode
+
+If you use VSCode, you can install the Dev Containers extension, and then in [Command Palette](https://code.visualstudio.com/docs/getstarted/userinterface#_command-palette), run `Dev Containers: Open Folder in Container`.
+After the container is built (the first time it may take 10+ minutes to build), you have a Redis server and local Llama server running.
+
+#### Other IDEs or Editors
+
+Please refer to [Dev Containers](https://containers.dev/supporting#editors) to see how to set up DevContainers in other editors.
+
+#### Without Dev Containers
+
+You can also set up the development environment without Dev Containers. Please manually install Redis, and if you want to use a local model, you can use Ollama, Llama.cpp,  vLLM or many others which support OpenAI compatible endpoints.
+
 
 ### 5. Write Code and Commit It
 

examples/generation_api/custom_model.py

@@ -0,0 +1,41 @@
+# Example: Generate (2~7) random numbers using Llama3.2 model served by ollama
+# To run this example, you can either
+# 1. Use the sotopia devcontainer and run the following command in the terminal:
+#        poetry run python examples/generation_api/custom_model.py
+#    This example can also serve a sanity check for your devcontainer setup.
+# OR 2. after installing sotopia, install ollama, and then:
+#        ollama pull llama3.2:1b; python examples/generation_api/custom_model.py
+# OR 3. after installing sotopia, serve your desired model on your desired port,
+#    and then change the model_name of `agenerate` to point to your desired model.
+#    Finally:
+#        python examples/generation_api/custom_model.py
+# Expected output for (1 and 2): a bunch of logs and an output [[14, 67], [6, 8, 3], [6, 8, 3, 9], [6, 8, 3, 9, 7], [7, 9, 6, 8, 4, 1]]
+
+from sotopia.generation_utils.generate import ListOfIntOutputParser, agenerate
+import logging
+
+# Set logging to the lowest level to show all logs
+logging.basicConfig(level=0)
+
+
+async def generate_n_random_numbers(n: int) -> list[int]:
+    return await agenerate(
+        model_name="custom/llama3.2:1b@http://localhost:11434/v1",
+        template="Generate {n} random integer numbers. {format_instructions}",
+        input_values={"n": str(n)},
+        temperature=0.0,
+        output_parser=ListOfIntOutputParser(n),
+    )
+
+
+async def main() -> None:
+    random_numbers = await asyncio.gather(
+        *[generate_n_random_numbers(n) for n in range(2, 7)]
+    )
+    print(random_numbers)
+
+
+if __name__ == "__main__":
+    import asyncio
+
+    asyncio.run(main())

pyproject.toml

@@ -44,6 +44,8 @@ nbmake = "*"
 types-setuptools = "*"
 types-requests = "*"
 types-tqdm = "*"
+lxml-stubs = "*"
+pandas-stubs = "*"
 ruff = "*"
 
 [tool.poetry.group.test.dependencies]

sotopia/generation_utils/generate.py

@@ -1,4 +1,3 @@
-import json
 import logging
 import os
 import re
@@ -57,11 +56,8 @@
     "redis",
     "groq/llama3-70b-8192",
 ]
-# subject to future OpenAI changes
-DEFAULT_BAD_OUTPUT_PROCESS_MODEL = "gpt-4o-mini"
 
 OutputType = TypeVar("OutputType", bound=object)
-client = OpenAI()
 
 
 class EnvResponse(BaseModel):
@@ -108,14 +104,14 @@ def __init__(
         self.range_of_int = range_of_int
 
     def _get_description_text(self) -> str:
-        return f"a list of {' ' + str(self.number_of_int) if self.number_of_int else ''} intergers{' within the range of' + str(self.range_of_int) if self.range_of_int else ''}. No code block is needed."
+        return f"a list of{' ' + str(self.number_of_int) if self.number_of_int else ''} intergers{' within the range of' + str(self.range_of_int) if self.range_of_int else ''} separated by spaces. Don't output anything else. Format example: 1 2 3 4 5"
 
     def get_format_instructions(self) -> str:
         return "Please output " + self._get_description_text()
 
     def parse(self, output: str) -> list[int]:
         try:
-            output_loaded = json.loads(output)
+            output_loaded = output.split(" ")
             result = [int(x) for x in output_loaded]
             if self.number_of_int and len(result) != self.number_of_int:
                 msg = f"Expect {self.number_of_int} integers, got {len(result)}"
@@ -261,18 +257,9 @@ def parse(self, output: str) -> ScriptInteractionReturnType:
             )
             return parsed_interaction
         except Exception as e:
-            print(f"Exception {e}: the output format is not correct. Reformatting ")
-            reformat_parsed_result = format_bad_output_for_script(
-                ill_formed_output=output,
-                format_instructions=self.get_format_instructions(),
-                agents=agent_names,
+            raise OutputParserException(
+                f"Failed to parse the output: {output}. Encounter Exception {e}"
             )
-            print("Reformatted output: ", reformat_parsed_result)
-            interaction = ScriptInteraction(interactions=reformat_parsed_result)
-            parsed_interaction = interaction.parse(
-                agent_names=agent_names, background=self.background
-            )
-            return parsed_interaction
 
     @property
     def _type(self) -> str:
@@ -404,7 +391,7 @@ def format_bad_output_for_script(
     ill_formed_output: str,
     format_instructions: str,
     agents: list[str],
-    model_name: str = DEFAULT_BAD_OUTPUT_PROCESS_MODEL,
+    model_name: str,
     use_fixed_model_version: bool = True,
 ) -> BaseMessage:
     template = """
@@ -440,7 +427,7 @@ def format_bad_output_for_script(
 def format_bad_output(
     ill_formed_output: BaseMessage,
     format_instructions: str,
-    model_name: str = DEFAULT_BAD_OUTPUT_PROCESS_MODEL,
+    model_name: str,
     use_fixed_model_version: bool = True,
 ) -> BaseMessage:
     template = """
@@ -475,7 +462,7 @@ async def agenerate(
     output_parser: BaseOutputParser[OutputType],
     temperature: float = 0.7,
     structured_output: bool = False,
-    bad_output_process_model: str = DEFAULT_BAD_OUTPUT_PROCESS_MODEL,
+    bad_output_process_model: str | None = None,
     use_fixed_model_version: bool = True,
 ) -> OutputType:
     input_variables = re.findall(
@@ -514,6 +501,7 @@ async def agenerate(
         instantiated_prompt = prompt_result.messages[0].content
         assert isinstance(output_parser, PydanticOutputParser)
         assert isinstance(instantiated_prompt, str)
+        client = OpenAI()
         completion = client.beta.chat.completions.parse(
             model=model_name,
             messages=[
@@ -538,7 +526,7 @@ async def agenerate(
         reformat_parsed_result = format_bad_output(
             result,
             format_instructions=output_parser.get_format_instructions(),
-            model_name=bad_output_process_model,
+            model_name=bad_output_process_model or model_name,
             use_fixed_model_version=use_fixed_model_version,
         )
         parsed_result = output_parser.invoke(reformat_parsed_result)
@@ -553,7 +541,7 @@ async def agenerate_env_profile(
     inspiration_prompt: str = "asking my boyfriend to stop being friends with his ex",
     examples: str = "",
     temperature: float = 0.7,
-    bad_output_process_model: str = DEFAULT_BAD_OUTPUT_PROCESS_MODEL,
+    bad_output_process_model: str | None = None,
     use_fixed_model_version: bool = True,
 ) -> tuple[EnvironmentProfile, str]:
     """
@@ -583,7 +571,7 @@ async def agenerate_env_profile(
 async def agenerate_relationship_profile(
     model_name: str,
     agents_profiles: list[str],
-    bad_output_process_model: str = DEFAULT_BAD_OUTPUT_PROCESS_MODEL,
+    bad_output_process_model: str | None = None,
     use_fixed_model_version: bool = True,
 ) -> tuple[RelationshipProfile, str]:
     """
@@ -617,7 +605,7 @@ async def agenerate_action(
     goal: str,
     temperature: float = 0.7,
     script_like: bool = False,
-    bad_output_process_model: str = DEFAULT_BAD_OUTPUT_PROCESS_MODEL,
+    bad_output_process_model: str | None = None,
     use_fixed_model_version: bool = True,
 ) -> AgentAction:
     """
@@ -685,7 +673,7 @@ async def agenerate_script(
     agent_name: str = "",
     history: str = "",
     single_step: bool = False,
-    bad_output_process_model: str = DEFAULT_BAD_OUTPUT_PROCESS_MODEL,
+    bad_output_process_model: str | None = None,
     use_fixed_model_version: bool = True,
 ) -> tuple[ScriptInteractionReturnType, str]:
     """
@@ -777,7 +765,7 @@ def process_history(
 async def agenerate_init_profile(
     model_name: str,
     basic_info: dict[str, str],
-    bad_output_process_model: str = DEFAULT_BAD_OUTPUT_PROCESS_MODEL,
+    bad_output_process_model: str | None = None,
     use_fixed_model_version: bool = True,
 ) -> str:
     """
@@ -823,7 +811,7 @@ async def convert_narratives(
     model_name: str,
     narrative: str,
     text: str,
-    bad_output_process_model: str = DEFAULT_BAD_OUTPUT_PROCESS_MODEL,
+    bad_output_process_model: str | None = None,
     use_fixed_model_version: bool = True,
 ) -> str:
     if narrative == "first":
@@ -856,7 +844,7 @@ async def convert_narratives(
 async def agenerate_goal(
     model_name: str,
     background: str,
-    bad_output_process_model: str = DEFAULT_BAD_OUTPUT_PROCESS_MODEL,
+    bad_output_process_model: str | None = None,
     use_fixed_model_version: bool = True,
 ) -> str:
     """