[wip] client w/ stainless sdk

llama_toolchain/agentic_system/client_sdk.py

@@ -0,0 +1,49 @@
+import fire
+from llama_stack import LlamaStack
+from llama_stack.types import UserMessage
+
+def main(host: str, port: int):
+    client = LlamaStack(
+        base_url=f"http://{host}:{port}",
+    )
+
+    # Need smt like this to work w/ server, however this is not what was generated by SDK
+    # response = client.inference.chat_completion(
+    #     request={
+    #         "messages": [
+    #             UserMessage(content="hello world, troll me in two-paragraphs about 42", role="user"),
+    #         ],
+    #         "model": "Meta-Llama3.1-8B-Instruct",
+    #         "stream": True,
+    #     },
+    # )
+
+    agentic_system_create_response = client.agentic_system.create(
+        agent_config={
+            "instructions": "You are a helpful assistant",
+            "model": "Meta-Llama3.1-8B-Instruct",
+        },
+    )
+    print(agentic_system_create_response)
+
+    agentic_system_create_session_response = client.agentic_system.sessions.create(
+        agent_id=agentic_system_create_response.agent_id,
+        session_name="test_session",
+    )
+    print(agentic_system_create_session_response)
+
+    # TODO(xiyan): This does not work with current server, need to wrap it in a request (similar to AgentConfig?)
+    response = client.agentic_system.turns.create(
+        agent_id=agentic_system_create_response.agent_id,
+        session_id=agentic_system_create_session_response.session_id,
+        messages=[
+            UserMessage(content="What is the capital of France?", role="user"),
+        ],
+        stream=True,
+    )
+
+    # print(response)
+
+
+if __name__ == "__main__":
+    fire.Fire(main)

llama_toolchain/inference/client_sdk.py

@@ -0,0 +1,32 @@
+import fire
+from llama_stack import LlamaStack
+from llama_stack.types import UserMessage
+
+def main(host: str, port: int):
+    client = LlamaStack(
+        base_url=f"http://{host}:{port}",
+    )
+
+    # Need smt like this to work w/ server, however this is not what was generated by SDK (?)
+    response = client.inference.chat_completion(
+        request={
+            "messages": [
+                UserMessage(content="hello world, troll me in two-paragraphs about 42", role="user"),
+            ],
+            "model": "Meta-Llama3.1-8B-Instruct",
+            "stream": False,
+        },
+    )
+
+    print(response)
+    # This does not work with current server
+    # response = client.inference.chat_completion(
+    #     messages=[
+    #         UserMessage(content="hello world, troll me in two-paragraphs about 42", role="user"),
+    #     ],
+    #     model="Meta-Llama3.1-8B-Instruct",
+    #     stream=True,
+    # )
+
+if __name__ == "__main__":
+    fire.Fire(main)