add promptfoo

promptfoo/conversation/tests.csv

@@ -1,9 +0,0 @@
-question,regex,llm-rubric,,,,,,,,,,,,,,,,,
-清水太郎在哪里,\[TOOL_CALL\].*find_employee_location,,,,,,,,,,,,,,,,,,
-通知他明天上午8点开会,,询问是否确认发送,,,,,,,,,,,,,,,,,
-确认,\[TOOL_CALL\].*wowtalk_send_message_to_member,,,,,,,,,,,,,,,,,,
-把DefineRoom 4的灯光状态发给他,,调用find_devices_by_room和dxcore_get_device_status获取灯光状态，并询问是否确认发送。,,,,,,,,,,,,,,,,,
-确认,\[TOOL_CALL\].*wowtalk_send_message_to_member,,,,,,,,,,,,,,,,,,
-900142の稼働状況,\[TOOL_CALL\].*dxcore_get_device_status,,,,,,,,,,,,,,,,,,
-关闭设备900142的灯光,,询问是否确认关闭,,,,,,,,,,,,,,,,,
-确认,\[TOOL_CALL\].*dxcore_update_device_status,,,,,,,,,,,,,,,,,,

promptfoo/csv_to_yaml.py

@@ -15,7 +15,8 @@ def csv_to_yaml(csv_file, yaml_file):
             if row['question']:
                 test_case = {
                     'vars': {
-                        'question': row['question'].strip()
+                        'question': row['question'].strip(),
+                        'use_history': True if row['use_history'] == "1" else False,
                     },
                     'assert':[]
                 }
@@ -41,5 +42,4 @@ def csv_to_yaml(csv_file, yaml_file):
     print(f"Converted {len(tests)} test cases from {csv_file} to {yaml_file}")
 
 if __name__ == '__main__':
-    csv_to_yaml("conversation/tests.csv", "conversation/tests.yaml")
-    csv_to_yaml("query/tests.csv", "query/tests.yaml")
+    csv_to_yaml("novare/novare.csv", "novare/tests.yaml")

promptfoo/novare/config.yaml

@@ -1,5 +1,5 @@
 env: {}
-description: Novare Test
+description: Novare Test - Unified Config for Both Single and Conversation
 providers:
   - id: openai:chat:qwen3
     config:
@@ -25,4 +25,4 @@ defaultTest:
           apiVersion: '2024-02-01'
 evaluateOptions: {}
 writeLatestResults: true
-sharing: true
+sharing: true

promptfoo/novare/novare.csv

@@ -0,0 +1,14 @@
+question,use_history,regex,llm-rubric,,,,,,,,,,,,,,,,,
+清水太郎在哪里,1,\[TOOL_CALL\].*find_employee_location,,,,,,,,,,,,,,,,,,
+通知他明天上午8点开会,1,,询问是否确认发送,,,,,,,,,,,,,,,,,
+确认,1,\[TOOL_CALL\].*wowtalk_send_message_to_member,,,,,,,,,,,,,,,,,,
+把DefineRoom 4的灯光状态发给他,1,,调用find_devices_by_room和dxcore_get_device_status获取灯光状态，并询问是否确认发送。,,,,,,,,,,,,,,,,,
+确认,1,\[TOOL_CALL\].*wowtalk_send_message_to_member,,,,,,,,,,,,,,,,,,
+关闭设备900142的灯光,1,,询问是否确认关闭,,,,,,,,,,,,,,,,,
+确认,1,\[TOOL_CALL\].*dxcore_update_device_status,,,,,,,,,,,,,,,,,,
+Define Room1 的灯光状态,0,\[TOOL_CALL\].*find_devices_by_room,,,,,,,,,,,,,,,,,,
+900142の稼働状況,0,\[TOOL_CALL\].*dxcore_get_device_status,,,,,,,,,,,,,,,,,,
+卫生间在哪里,0,\[TOOL_CALL\].*rag_retrieve,,,,,,,,,,,,,,,,,,
+我丢了物品怎么办,0,\[TOOL_CALL\].*rag_retrieve,,,,,,,,,,,,,,,,,,
+咖啡多少钱一杯,0,\[TOOL_CALL\].*rag_retrieve,,,,,,,,,,,,,,,,,,
+东京明天的天气,0,\[TOOL_CALL\].*weather_get_by_location,,,,,,,,,,,,,,,,,,

promptfoo/novare/prompt.json

@@ -1,4 +1,5 @@
 [
+{% if use_history %}
 {% for completion in _conversation %}
     {
       "role": "user",
@@ -13,4 +14,10 @@
       "role": "user",
       "content": "{{ question | encode }}"
     }
-]
+{% else %}
+  {
+      "role": "user",
+      "content": "{{ question }}"
+  }
+{% endif %}
+]

promptfoo/novare/tests.yaml

@@ -1,40 +1,78 @@
 - vars:
     question: 清水太郎在哪里
+    use_history: true
   assert:
   - type: regex
     value: \[TOOL_CALL\].*find_employee_location
 - vars:
     question: 通知他明天上午8点开会
+    use_history: true
   assert:
   - type: llm-rubric
     value: 询问是否确认发送
 - vars:
     question: 确认
+    use_history: true
   assert:
   - type: regex
     value: \[TOOL_CALL\].*wowtalk_send_message_to_member
 - vars:
     question: 把DefineRoom 4的灯光状态发给他
+    use_history: true
   assert:
   - type: llm-rubric
     value: 调用find_devices_by_room和dxcore_get_device_status获取灯光状态，并询问是否确认发送。
 - vars:
     question: 确认
+    use_history: true
   assert:
   - type: regex
     value: \[TOOL_CALL\].*wowtalk_send_message_to_member
-- vars:
-    question: 900142の稼働状況
-  assert:
-  - type: regex
-    value: \[TOOL_CALL\].*dxcore_get_device_status
 - vars:
     question: 关闭设备900142的灯光
+    use_history: true
   assert:
   - type: llm-rubric
     value: 询问是否确认关闭
 - vars:
     question: 确认
+    use_history: true
   assert:
   - type: regex
     value: \[TOOL_CALL\].*dxcore_update_device_status
+- vars:
+    question: Define Room1 的灯光状态
+    use_history: false
+  assert:
+  - type: regex
+    value: \[TOOL_CALL\].*find_devices_by_room
+- vars:
+    question: 900142の稼働状況
+    use_history: false
+  assert:
+  - type: regex
+    value: \[TOOL_CALL\].*dxcore_get_device_status
+- vars:
+    question: 卫生间在哪里
+    use_history: false
+  assert:
+  - type: regex
+    value: \[TOOL_CALL\].*rag_retrieve
+- vars:
+    question: 我丢了物品怎么办
+    use_history: false
+  assert:
+  - type: regex
+    value: \[TOOL_CALL\].*rag_retrieve
+- vars:
+    question: 咖啡多少钱一杯
+    use_history: false
+  assert:
+  - type: regex
+    value: \[TOOL_CALL\].*rag_retrieve
+- vars:
+    question: 东京明天的天气
+    use_history: false
+  assert:
+  - type: regex
+    value: \[TOOL_CALL\].*weather_get_by_location

promptfoo/query/config.yaml

@@ -1,26 +0,0 @@
-env: {}
-description: Novare Test
-providers:
-  - id: openai:chat:qwen3
-    config:
-      apiBaseUrl: https://catalog-agent-dev.gbase.ai/api/v2
-      apiKey: a21c99620a8ef61d69563afe05ccce89
-      passthrough:
-        bot_id: 63069654-7750-409d-9a58-a0960d899a20
-        tool_response: true
-        language: zh
-prompts:
-  - file://prompt.json
-tests: file://tests.yaml
-defaultTest:
-  options:
-    provider:
-      text:
-        id: openai:chat:qwen/qwen3-next-80b-a3b-instruct
-        config:
-          apiKey: sk-hsKClH0Z695EkK5fDdB2Ec2fE13f4fC1B627BdBb8e554b5b-26
-          apiBaseUrl: https://one.felo.me/v1
-          apiVersion: '2024-02-01'
-evaluateOptions: {}
-writeLatestResults: true
-sharing: true

promptfoo/query/prompt.json

@@ -1,6 +0,0 @@
-[
-  {
-      "role": "user",
-      "content": "{{ question }}"
-  }
-]

promptfoo/query/tests.csv

@@ -1,6 +0,0 @@
-question,regex,llm-rubric,,,,,,,,,,,,,,,,,
-Define Room1 的灯光状态,\[TOOL_CALL\].*find_devices_by_room,,,,,,,,,,,,,,,,,,
-卫生间在哪里,\[TOOL_CALL\].*rag_retrieve,,,,,,,,,,,,,,,,,,
-我丢了物品怎么办,\[TOOL_CALL\].*rag_retrieve,,,,,,,,,,,,,,,,,,
-咖啡多少钱一杯,\[TOOL_CALL\].*rag_retrieve,,,,,,,,,,,,,,,,,,
-东京明天的天气,\[TOOL_CALL\].*weather_get_by_location,,,,,,,,,,,,,,,,,,

promptfoo/query/tests.yaml

@@ -1,25 +0,0 @@
-- vars:
-    question: Define Room1 的灯光状态
-  assert:
-  - type: regex
-    value: \[TOOL_CALL\].*find_devices_by_room
-- vars:
-    question: 卫生间在哪里
-  assert:
-  - type: regex
-    value: \[TOOL_CALL\].*rag_retrieve
-- vars:
-    question: 我丢了物品怎么办
-  assert:
-  - type: regex
-    value: \[TOOL_CALL\].*rag_retrieve
-- vars:
-    question: 咖啡多少钱一杯
-  assert:
-  - type: regex
-    value: \[TOOL_CALL\].*rag_retrieve
-- vars:
-    question: 东京明天的天气
-  assert:
-  - type: regex
-    value: \[TOOL_CALL\].*weather_get_by_location