Merge branch 'main' of https://github.com/tangledup-ai/langchain-agent

lang_agent/config.py

@@ -50,8 +50,42 @@ class InstantiateConfig(PrintableConfig):
     
     def save_config(self, filename: str) -> None:
         """Save the config to a YAML file."""
+        def mask_value(key, value):
+            # Apply masking if key is secret-like
+            if isinstance(value, str) and self.is_secrete(key):
+                sval = str(value)
+                return sval[:3] + "*" * (len(sval) - 6) + sval[-3:]
+            return value
+
+        def to_masked_serializable(obj):
+            # Recursively convert dataclasses and containers to serializable with masked secrets
+            if is_dataclass(obj):
+                out = {}
+                for k, v in vars(obj).items():
+                    if is_dataclass(v) or isinstance(v, (dict, list, tuple)):
+                        out[k] = to_masked_serializable(v)
+                    else:
+                        out[k] = mask_value(k, v)
+                return out
+            if isinstance(obj, dict):
+                out = {}
+                for k, v in obj.items():
+                    if is_dataclass(v) or isinstance(v, (dict, list, tuple)):
+                        out[k] = to_masked_serializable(v)
+                    else:
+                        # k might be a non-string; convert to str for is_secrete check consistency
+                        key_str = str(k)
+                        out[k] = mask_value(key_str, v)
+                return out
+            if isinstance(obj, list):
+                return [to_masked_serializable(v) for v in obj]
+            if isinstance(obj, tuple):
+                return tuple(to_masked_serializable(v) for v in obj)
+            return obj
+
+        masked = to_masked_serializable(self)
         with open(filename, 'w') as f:
-            yaml.dump(self, f)
+            yaml.dump(masked, f)
         logger.info(f"[yellow]config saved to: {filename}[/yellow]")
     
     def get_name(self):

lang_agent/eval/evaluator.py

@@ -25,7 +25,7 @@ class EvaluatorConfig(InstantiateConfig):
     experiment_desc:str = "testing if this works or not"
     """describe the experiment"""
 
-    dataset_name:Literal["Toxic Queries"] = "dev_langagent"
+    dataset_name:Literal["Toxic Queries"] = "xiao_zhan"
     """name of the dataset to evaluate"""
 
     log_dir:str = "logs"
@@ -73,7 +73,7 @@ class Evaluator:
 
         head_path = osp.join(self.config.log_dir, f"{self.dataset.name}-{self.config.experiment_prefix}")
         n_exp = len(glob.glob(f"{head_path}*"))
-        exp_save_f = osp.join(f"{head_path}-{n_exp}.csv")
+        exp_save_f = f"{head_path}-{n_exp}.csv"
 
         df = self.result.to_pandas()
         logger.info(f"saving experiment results to: {exp_save_f}")
@@ -93,4 +93,6 @@ class Evaluator:
         
         df_m.to_csv(metric_f)
 
+        self.config.save_config(f"{head_path}-{n_exp}.yml")
+
 

lang_agent/eval/validator.py

@@ -51,6 +51,9 @@ class Validator:
         actual_answer = outputs["output"][-1].content
         expected_answer = reference_outputs["answer"]
 
+        if expected_answer is None:
+            return True
+
         user_msg = (
             f"ACTUAL ANSWER: {actual_answer}"
             f"\n\nEXPECTED ANSWER: {expected_answer}"
@@ -77,7 +80,7 @@ class Validator:
             st_cond = False
             ref_tool = ref_tool.lower()
             for msg in tool_msgs:
-                st_cond = ref_tool in msg.name.lower()
+                st_cond = msg.name.lower() in ref_tool
                 if st_cond:
                     break
             tool_used.append(st_cond)

scripts/make_eval_dataset.py

@@ -2,23 +2,51 @@ from langsmith import Client
 from loguru import logger
 
 
-DATASET_NAME = "dev_langagent"
+DATASET_NAME = "xiao_zhan"
 
 examples = [
     {
-        "inputs": {"text": "介绍一下自己"},
+        "inputs": {"text": "今天有点热，推荐点茶喝一下"},
-        "outputs": {"answer": "我是小盏，是一个点餐助手"}
+        "outputs": {"answer": None,
-    },
-    {
-        "inputs": {"text": "用retrieve查询光予尘然后介绍"},
-        "outputs": {"answer": "茉莉绿茶为底，清冽茶香中漫出玫珑蜜瓜的绵甜与凤梨的明亮果香，层次鲜活；顶部白柚茉莉泡沫轻盈漫过舌尖，带着微酸的清新感，让整体风味更显灵动",
                     "tool_use": ["retrieve"]}
     },
     {
-        "inputs": {"text": ["我叫Steve",
+        "inputs": {"text": "有没有光予尘？"},
-                            "我叫什么名字?"]},   # list for conversation
+        "outputs": {"answer": "有的",
-        "outputs": {"answer": "你叫Steve"}
+                    "tool_use": ["retrieve|get_resource"]}
-    }
+    },
+    {
+        "inputs": {"text": "有没有关羽尘？"},
+        "outputs": {"answer": "有的",
+                    "tool_use": ["retrieve|get_resource"]}
+    },
+    {
+        "inputs": {"text": ["我要购买一杯野星星",
+                            "我要再加一杯"]},
+        "outputs": {"answer": "你的野星星已经下单成功",
+                    "tool_use": ["retrieve|get_resources",
+                                 "start_shopping_session",
+                                 "add_to_cart",
+                                 "create_wechat_pay",
+                                 "create_order_from_cart",
+                                 "update_cart_item"]}
+    },
+    {
+        "inputs": {"text": ["我要购买一杯野星星",
+                            "我现在点了些什么？"]},
+        "outputs": {"answer": "一杯野星星",
+                    "tool_use": ["query_wechat_order"]}
+    },
+    {
+        "inputs": {"text": ["我要购买三杯野星星",
+                            "现在取消所有我定了的饮品"]},
+        "outputs": {"answer": "取消成功",
+                    "tool_use": ["clear_cart"]}
+    },
+    {
+        "inputs": {"text": "你是谁？"},
+        "outputs": {"answer": "小盏"}
+    },
 ]
 
 cli = Client()