Update sampler in server configuration (#33)

* Update sampler in server configuration

* Add minimalist_demo.py to examples

* Update server configuration in minimalist_demo.py

* 📝 add minimalist_demo.py

* fix to max_token limit

---------

Co-authored-by: clementou <[email protected]>

README.md

@@ -55,9 +55,26 @@ mkdir logs
 ```
 
 ## Easy Sample Server
-You can view an episode demo with default parameters using the following command:
+You can view an episode demo with default parameters with the following:
 ```python
-python -m sotopia_conf.server --gin_file="sotopia_conf/server_conf/server.gin" --gin_file="sotopia_conf/generation_utils_conf/generate.gin"
+import asyncio
+from sotopia.samplers import UniformSampler
+from sotopia.server import run_async_server
+
+asyncio.run(
+    run_async_server(
+        model_dict={
+            "env": "gpt-4",
+            "agent1": "gpt-3.5-turbo",
+            "agent2": "gpt-3.5-turbo",
+        },
+        sampler=UniformSampler(),
+    )
+)
+```
+or run
+```bash
+python examples/minimalist_demo.py
 ```
 
 ## Contribution

examples/minimalist_demo.py

@@ -0,0 +1,23 @@
+# This demo servers as a minimal example of how to use the sotopia library.
+
+# 1. Import the sotopia library
+# 1.1. Import the `run_async_server` function: In sotopia, we use Python Async
+#     API to optimize the throughput.
+import asyncio
+
+# 1.2. Import the `UniformSampler` class: In sotopia, we use samplers to sample
+#     the social tasks.
+from sotopia.samplers import UniformSampler
+from sotopia.server import run_async_server
+
+# 2. Run the server
+asyncio.run(
+    run_async_server(
+        model_dict={
+            "env": "gpt-4",
+            "agent1": "gpt-3.5-turbo",
+            "agent2": "gpt-3.5-turbo",
+        },
+        sampler=UniformSampler(),
+    )
+)

sotopia/generation_utils/generate.py

@@ -321,7 +321,7 @@ def obtain_chain(
     chat = ChatLiteLLM(
         model=model_name,
         temperature=temperature,
-        max_tokens=3072,  # tweak as needed
+        max_tokens=2700,  # tweak as needed
         max_retries=max_retries,
     )
     human_message_prompt = HumanMessagePromptTemplate(

sotopia/server.py

@@ -317,7 +317,7 @@ def get_agent_class(
     if env_agent_combo_list:
         assert (
             type(sampler) is BaseSampler
-        ), "No sampler should be used when `env_agent_combo_list` is empty"
+        ), "No sampler should be used when `env_agent_combo_list` is not empty"
         env_agent_combo_iter = iter(env_agent_combo_list)
     else:
         env_agent_combo_iter = sampler.sample(

sotopia_conf/server_conf/server.gin

@@ -1,6 +1,7 @@
 from __gin__ import dynamic_registration
 import __main__ as train_script
 import sotopia.server as server
+import sotopia.samplers as samplers
 
 ACTION_ORDER="round-robin"
 TAG=None
@@ -14,3 +15,4 @@ server.run_async_server:
     using_async=True
     tag=%TAG
     omniscient=%OMNISCIENT
+    [email protected]()