Fix React (#41)

* add log mode, refine doc

* fix react

---------

Co-authored-by: Satan <liuxk2019@mail.sustech.edu.cn>

Author: liuxukun2000

Makefile

@@ -5,7 +5,7 @@
 # from the environment for the first two.
 SPHINXOPTS    ?=
 SPHINXBUILD   ?= sphinx-build
-SOURCEDIR     = source
+SOURCEDIR     = doc/source
 BUILDDIR      = build
 
 # Put it first so that "make" without argument is like "make help".

gentopia/agent/react/agent.py

@@ -148,6 +148,7 @@ def run(self, instruction, max_iterations=10):
         :return: AgentOutput object.
         :rtype: AgentOutput
         """
+        self.intermediate_steps.clear()
         logging.info(f"Running {self.name + ':' + self.version} with instruction: {instruction}")
         total_cost = 0.0
         total_token = 0
@@ -158,8 +159,8 @@ def run(self, instruction, max_iterations=10):
             logging.info(f"Prompt: {prompt}")
             response = self.llm.completion(prompt, stop=["Observation:"])
             if response.state == "error":
-                print("Planner failed to retrieve response from LLM")
-                raise ValueError("Planner failed to retrieve response from LLM")
+                print("Failed to retrieve response from LLM")
+                raise ValueError("Failed to retrieve response from LLM")
 
             logging.info(f"Response: {response.content}")
             total_cost += calculate_cost(self.llm.model_name, response.prompt_token,
@@ -180,7 +181,7 @@ def run(self, instruction, max_iterations=10):
             self.intermediate_steps[-1].append(result)
         return AgentOutput(output=response.content, cost=total_cost, token_usage=total_token)
 
-    def stream(self, instruction: Optional[str] = None, output: Optional[BaseOutput] = None):
+    def stream(self, instruction: Optional[str] = None, output: Optional[BaseOutput] = None, max_iterations: int = 10):
         """
         Stream output the agent with the given instruction.
 
@@ -197,7 +198,7 @@ def stream(self, instruction: Optional[str] = None, output: Optional[BaseOutput]
         if output is None:
             output = BaseOutput()
         output.thinking(self.name)
-        for _ in range(10):
+        for _ in range(max_iterations):
 
             prompt = self._compose_prompt(instruction)
             logging.info(f"Prompt: {prompt}")
@@ -208,8 +209,6 @@ def stream(self, instruction: Optional[str] = None, output: Optional[BaseOutput]
             for i in response:
                 content += i.content
                 output.panel_print(i.content, self.name, True)
-
-                # print(i.content)
             output.clear()
 
             logging.info(f"Response: {content}")

test.py

@@ -16,8 +16,11 @@
     enable_log(log_level='debug')
     dotenv.load_dotenv(".env")
 
-    assembler = AgentAssembler(file='configs/memory.yaml')
+    assembler = AgentAssembler(file='configs/react.yaml')
 
     # # assembler.manager = LocalLLMManager()
     agent = assembler.get_agent()
-    chat(agent)
+
+    # chat(agent)
+    print(agent.run("1+sqrt(33)=?"))
+    print(agent.run("1+sqrt(35)=?"))