jupyter-agent 2025.6.104__py3-none-any.whl → 2025.7.100__py3-none-any.whl
This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
- jupyter_agent/bot_actions.py +270 -0
- jupyter_agent/bot_agents/__init__.py +0 -42
- jupyter_agent/bot_agents/base.py +89 -45
- jupyter_agent/bot_agents/master_planner.py +1 -0
- jupyter_agent/bot_agents/output_task_result.py +6 -7
- jupyter_agent/bot_agents/prepare_next_cell.py +52 -0
- jupyter_agent/bot_agents/request_user_supply.py +186 -0
- jupyter_agent/bot_agents/task_code_executor.py +3 -2
- jupyter_agent/bot_agents/task_planner_v3.py +16 -13
- jupyter_agent/bot_agents/task_reasoner.py +3 -2
- jupyter_agent/bot_agents/task_structrue_reasoner.py +22 -12
- jupyter_agent/bot_agents/task_structrue_summarier.py +22 -18
- jupyter_agent/bot_agents/task_summarier.py +3 -2
- jupyter_agent/bot_agents/task_verifier.py +2 -1
- jupyter_agent/bot_agents/task_verify_summarier.py +6 -6
- jupyter_agent/bot_chat.py +2 -2
- jupyter_agent/bot_contexts.py +37 -29
- jupyter_agent/bot_evaluation.py +262 -143
- jupyter_agent/bot_evaluators/__init__.py +0 -0
- jupyter_agent/bot_evaluators/base.py +42 -0
- jupyter_agent/bot_evaluators/dummy_flow.py +20 -0
- jupyter_agent/bot_evaluators/dummy_global.py +20 -0
- jupyter_agent/bot_evaluators/dummy_task.py +20 -0
- jupyter_agent/bot_evaluators/flow_global_planning.py +88 -0
- jupyter_agent/bot_evaluators/flow_task_executor.py +152 -0
- jupyter_agent/bot_flows/__init__.py +0 -4
- jupyter_agent/bot_flows/base.py +120 -41
- jupyter_agent/bot_flows/master_planner.py +15 -4
- jupyter_agent/bot_flows/task_executor_v3.py +57 -38
- jupyter_agent/bot_magics.py +119 -69
- jupyter_agent/bot_outputs.py +37 -43
- jupyter_agent/utils.py +20 -31
- {jupyter_agent-2025.6.104.dist-info → jupyter_agent-2025.7.100.dist-info}/METADATA +56 -4
- jupyter_agent-2025.7.100.dist-info/RECORD +41 -0
- jupyter_agent/bot_agents/task_planner_v1.py +0 -158
- jupyter_agent/bot_agents/task_planner_v2.py +0 -172
- jupyter_agent/bot_flows/task_executor_v1.py +0 -86
- jupyter_agent/bot_flows/task_executor_v2.py +0 -84
- jupyter_agent-2025.6.104.dist-info/RECORD +0 -35
- {jupyter_agent-2025.6.104.dist-info → jupyter_agent-2025.7.100.dist-info}/WHEEL +0 -0
- {jupyter_agent-2025.6.104.dist-info → jupyter_agent-2025.7.100.dist-info}/entry_points.txt +0 -0
- {jupyter_agent-2025.6.104.dist-info → jupyter_agent-2025.7.100.dist-info}/licenses/LICENSE +0 -0
- {jupyter_agent-2025.6.104.dist-info → jupyter_agent-2025.7.100.dist-info}/top_level.txt +0 -0
@@ -0,0 +1,152 @@
|
|
1
|
+
"""
|
2
|
+
Copyright (c) 2025 viewstar000
|
3
|
+
|
4
|
+
This software is released under the MIT License.
|
5
|
+
https://opensource.org/licenses/MIT
|
6
|
+
"""
|
7
|
+
|
8
|
+
import time
|
9
|
+
|
10
|
+
from enum import Enum
|
11
|
+
from typing import Optional, List
|
12
|
+
from pydantic import BaseModel, Field
|
13
|
+
from IPython.display import Markdown
|
14
|
+
from .base import BaseEvaluator
|
15
|
+
from ..bot_outputs import _D, _I, _W, _E, _F, _A, _O, _C, _M, _B
|
16
|
+
from ..bot_evaluation import FlowEvaluationRecord
|
17
|
+
|
18
|
+
|
19
|
+
FLOW_TASK_EXEC_EVAL_PROMPT = """\
|
20
|
+
**角色定义**:
|
21
|
+
|
22
|
+
你是一个任务规划评估专家,负责对任务规划的结果进行评估。
|
23
|
+
|
24
|
+
**任务要求**:
|
25
|
+
|
26
|
+
请你根据任务规划的结果,评估任务规划的质量和准确性,并给出相应的评分和反馈。
|
27
|
+
|
28
|
+
{% include "TASK_OUTPUT_FORMAT" %}
|
29
|
+
|
30
|
+
---
|
31
|
+
|
32
|
+
{% include "TASK_CONTEXTS" %}
|
33
|
+
|
34
|
+
---
|
35
|
+
|
36
|
+
{% include "CODE_CONTEXTS" %}
|
37
|
+
|
38
|
+
---
|
39
|
+
|
40
|
+
**当前子任务规划信息**:
|
41
|
+
|
42
|
+
### 当前子任务规划目标:
|
43
|
+
|
44
|
+
{{ task.subject }}
|
45
|
+
|
46
|
+
{% if task.coding_prompt %}
|
47
|
+
### 当前子任务代码需求:
|
48
|
+
|
49
|
+
{{ task.coding_prompt }}
|
50
|
+
|
51
|
+
### 当前子任务生成的代码:
|
52
|
+
|
53
|
+
```python
|
54
|
+
{{ task.source }}
|
55
|
+
```
|
56
|
+
|
57
|
+
### 当前代码执行的输出与结果:
|
58
|
+
|
59
|
+
{{ task.output }}
|
60
|
+
{% endif %}
|
61
|
+
|
62
|
+
### 当前子任务总结要求:
|
63
|
+
|
64
|
+
{{ task.summary_prompt }}
|
65
|
+
|
66
|
+
|
67
|
+
### 当前子任务输出的分析总结后的最终结果:
|
68
|
+
|
69
|
+
```markdown
|
70
|
+
{{ task.result }}
|
71
|
+
```
|
72
|
+
|
73
|
+
{% if task.important_infos %}
|
74
|
+
### 当前子任务输出的重要信息:
|
75
|
+
|
76
|
+
```json
|
77
|
+
{{ task.important_infos | json }}
|
78
|
+
```
|
79
|
+
{% endif %}
|
80
|
+
|
81
|
+
{% if task.request_below_supply_infos %}
|
82
|
+
### 当前子任务输出的请求用户补充确认的信息:
|
83
|
+
|
84
|
+
```json
|
85
|
+
{{ task.request_below_supply_infos | json }}
|
86
|
+
```
|
87
|
+
{% endif %}
|
88
|
+
|
89
|
+
---
|
90
|
+
|
91
|
+
请按要求给出当前子任务规划的评估结果:
|
92
|
+
"""
|
93
|
+
|
94
|
+
|
95
|
+
class FlowTaskExecEvalResult(BaseModel):
|
96
|
+
"""
|
97
|
+
任务规划评估结果
|
98
|
+
"""
|
99
|
+
|
100
|
+
is_correct: bool = Field(description="最终结果是否符合当前子任务的目标", examples=[True, False])
|
101
|
+
correct_score: float = Field(
|
102
|
+
description="最终结果符合当前子任务目标的分数,范围0-1,>=0.5表示符合目标,<0.5表示不符合目标",
|
103
|
+
examples=[0.95, 0.3],
|
104
|
+
)
|
105
|
+
planning_score: float = Field(
|
106
|
+
description="当前子任务的目标规划、代码生成、总结是否符合全局目标规划要求,范围0-1,>=0.5表示符合要求,<0.5表示不符合要求",
|
107
|
+
examples=[0.85, 0.25],
|
108
|
+
)
|
109
|
+
reasoning_score: float = Field(
|
110
|
+
description="当前子任务的推理过程是否合理,是否存在逻辑错误,是否存在与前置子任务相冲突的情况,"
|
111
|
+
"范围0-1,>=0.5表示合理、正确、无冲突,<0.5表示不合理",
|
112
|
+
examples=[0.9, 0.4],
|
113
|
+
)
|
114
|
+
coding_score: float = Field(
|
115
|
+
description="代码生成的质量评分,代码逻辑是否符合规划要求,是否存在逻辑错误,是否存在冗余、抽象不合理等情况,"
|
116
|
+
"范围0-1,>=0.5表示代码质量较高,<0.5表示代码质量较低",
|
117
|
+
examples=[0.75, 0.2],
|
118
|
+
)
|
119
|
+
important_info_score: float = Field(
|
120
|
+
description="重要信息分数,当前子任务的规划、代码生成、总结是否充分考虑了前置任务生成的重要信息,"
|
121
|
+
"以及当前子任务的重要信息是否完整、准确、无误导、无冲突,"
|
122
|
+
"范围0-1,>=0.5表示重要信息完整、准确,<0.5表示重要信息不完整或不准确",
|
123
|
+
examples=[0.9, 0.4],
|
124
|
+
)
|
125
|
+
user_supply_info_score: float = Field(
|
126
|
+
description="用户补充信息分数,当前子任务的规划、代码生成、总结是否充分考虑了用户补充的信息,"
|
127
|
+
"范围0-1,>=0.5表示充分考虑,<0.5表示未充分考虑",
|
128
|
+
examples=[0.8, 0.3],
|
129
|
+
)
|
130
|
+
feedback: Optional[str] = Field(default=None, description="评估反馈")
|
131
|
+
|
132
|
+
|
133
|
+
class FlowTaskExecEvaluator(BaseEvaluator):
|
134
|
+
"""
|
135
|
+
任务规划评估器
|
136
|
+
"""
|
137
|
+
|
138
|
+
PROMPT = FLOW_TASK_EXEC_EVAL_PROMPT
|
139
|
+
OUTPUT_JSON_SCHEMA = FlowTaskExecEvalResult
|
140
|
+
|
141
|
+
def on_reply(self, reply):
|
142
|
+
reply = super().on_reply(reply)
|
143
|
+
return FlowEvaluationRecord(
|
144
|
+
timestamp=time.time(),
|
145
|
+
evaluator="flow_task_executor",
|
146
|
+
correct_score=reply.correct_score,
|
147
|
+
planning_score=reply.planning_score,
|
148
|
+
reasoning_score=reply.reasoning_score,
|
149
|
+
coding_score=reply.coding_score,
|
150
|
+
important_score=reply.important_info_score,
|
151
|
+
user_supply_score=reply.user_supply_info_score,
|
152
|
+
)
|
@@ -7,14 +7,10 @@ https://opensource.org/licenses/MIT
|
|
7
7
|
|
8
8
|
from .base import BaseTaskFlow
|
9
9
|
from .master_planner import MasterPlannerFlow
|
10
|
-
from .task_executor_v1 import TaskExecutorFlowV1
|
11
|
-
from .task_executor_v2 import TaskExecutorFlowV2
|
12
10
|
from .task_executor_v3 import TaskExecutorFlowV3
|
13
11
|
|
14
12
|
__all__ = [
|
15
13
|
"BaseTaskFlow",
|
16
14
|
"MasterPlannerFlow",
|
17
|
-
"TaskExecutorFlowV1",
|
18
|
-
"TaskExecutorFlowV2",
|
19
15
|
"TaskExecutorFlowV3",
|
20
16
|
]
|
jupyter_agent/bot_flows/base.py
CHANGED
@@ -13,9 +13,12 @@ from enum import Enum
|
|
13
13
|
from typing import List, Dict, Optional, Type
|
14
14
|
from IPython.display import Markdown
|
15
15
|
from ..bot_agents.base import BaseAgent
|
16
|
+
from ..bot_evaluators.base import BaseEvaluator
|
17
|
+
from ..bot_evaluators.dummy_global import DummyGlobalEvaluator
|
18
|
+
from ..bot_evaluators.flow_task_executor import FlowTaskExecEvaluator
|
16
19
|
from ..bot_outputs import _D, _I, _W, _E, _F, _M, _B
|
17
20
|
from ..bot_outputs import set_stage, flush_output, output_evaluation
|
18
|
-
from ..
|
21
|
+
from ..bot_evaluation import FlowEvaluationRecord, StageEvaluationRecord, NotebookEvaluationRecord
|
19
22
|
|
20
23
|
TASK_AGENT_STATE_ERROR = "_AGENT_STATE_ERROR_32534526_"
|
21
24
|
TASK_STAGE_START = "start"
|
@@ -37,9 +40,10 @@ class StageNext[ST](BaseModel):
|
|
37
40
|
message: str = ""
|
38
41
|
|
39
42
|
|
40
|
-
class
|
43
|
+
class StageNode[ST, AS](BaseModel):
|
41
44
|
stage: ST | str
|
42
|
-
|
45
|
+
agents: Type[BaseAgent] | List[Type[BaseAgent]]
|
46
|
+
evaluators: Optional[Type[BaseEvaluator] | List[Type[BaseEvaluator]]] = None
|
43
47
|
states: Dict[AS | str, StageNext[ST] | List[StageNext[ST]] | Dict[TaskAction, StageNext[ST]] | ST | str] = {}
|
44
48
|
next_stage: Optional[StageNext[ST] | List[StageNext[ST]] | Dict[TaskAction, StageNext[ST]] | ST | str] = None
|
45
49
|
|
@@ -49,15 +53,18 @@ class BaseTaskFlow:
|
|
49
53
|
基础任务流程
|
50
54
|
"""
|
51
55
|
|
52
|
-
|
56
|
+
STAGE_NODES: List[StageNode] = []
|
53
57
|
START_STAGE = TASK_STAGE_START
|
54
58
|
STOP_STAGES = [TASK_STAGE_COMPLETED, TASK_STAGE_GLOBAL_FINISHED]
|
59
|
+
FLOW_EVALUATOR = FlowTaskExecEvaluator
|
60
|
+
GLOBAL_EVALUATOR = DummyGlobalEvaluator
|
55
61
|
|
56
|
-
def __init__(self, notebook_context, agent_factory):
|
62
|
+
def __init__(self, notebook_context, agent_factory, evaluator_factory=None):
|
57
63
|
self.notebook_context = notebook_context
|
58
64
|
self.agent_factory = agent_factory
|
59
|
-
self.
|
60
|
-
self.
|
65
|
+
self.evaluator_factory = evaluator_factory
|
66
|
+
self.stage_nodes = {}
|
67
|
+
self.prepare_stage_nodes()
|
61
68
|
|
62
69
|
@property
|
63
70
|
def task(self):
|
@@ -67,10 +74,10 @@ class BaseTaskFlow:
|
|
67
74
|
def cells(self):
|
68
75
|
return self.notebook_context.cells
|
69
76
|
|
70
|
-
def
|
71
|
-
for st in self.
|
77
|
+
def prepare_stage_nodes(self):
|
78
|
+
for st in self.STAGE_NODES:
|
72
79
|
assert not (st.next_stage and st.states), "next_stage and states are mutually exclusive"
|
73
|
-
self.
|
80
|
+
self.stage_nodes[st.stage] = st
|
74
81
|
if st.next_stage:
|
75
82
|
st.states[TaskAction.DEFAULT] = st.next_stage
|
76
83
|
st.next_stage = None
|
@@ -99,15 +106,29 @@ class BaseTaskFlow:
|
|
99
106
|
if TASK_AGENT_STATE_ERROR not in st.states:
|
100
107
|
st.states[TASK_AGENT_STATE_ERROR] = {"*": StageNext(stage=st.stage)}
|
101
108
|
|
102
|
-
def
|
103
|
-
for t in self.
|
109
|
+
def get_stage_agents(self, stage) -> List[BaseAgent]:
|
110
|
+
for t in self.STAGE_NODES:
|
104
111
|
if t.stage == stage:
|
105
|
-
|
112
|
+
if isinstance(t.agents, list):
|
113
|
+
return [self.agent_factory(a) for a in t.agents]
|
114
|
+
else:
|
115
|
+
return [self.agent_factory(t.agents)]
|
106
116
|
raise ValueError(f"No agent for stage `{stage}`")
|
107
117
|
|
118
|
+
def get_stage_evaluators(self, stage) -> List[BaseEvaluator]:
|
119
|
+
if self.evaluator_factory is None:
|
120
|
+
return []
|
121
|
+
for t in self.STAGE_NODES:
|
122
|
+
if t.stage == stage:
|
123
|
+
if isinstance(t.evaluators, list):
|
124
|
+
return [self.evaluator_factory(e) for e in t.evaluators]
|
125
|
+
else:
|
126
|
+
return [self.evaluator_factory(t.evaluators)]
|
127
|
+
return []
|
128
|
+
|
108
129
|
def _get_next_stage_trans(self, stage, state, action=TaskAction.CONTINUE):
|
109
130
|
|
110
|
-
st = self.
|
131
|
+
st = self.stage_nodes.get(stage)
|
111
132
|
if st:
|
112
133
|
state_ns = st.states.get(state) or st.states.get("*")
|
113
134
|
assert state_ns, f"No next stage for stage `{stage}` and state `{state}`"
|
@@ -157,16 +178,19 @@ class BaseTaskFlow:
|
|
157
178
|
stage_count = 0
|
158
179
|
# Initialize the task stage
|
159
180
|
stage = stage or self.START_STAGE
|
181
|
+
agent = None
|
160
182
|
while n_tries <= max_tries:
|
161
183
|
stage_st = time.time()
|
162
184
|
try:
|
163
185
|
stage_name = stage.value if isinstance(stage, Enum) else stage
|
164
186
|
stage_name = stage_name.replace(".", "-").capitalize()
|
165
187
|
set_stage(stage_name)
|
166
|
-
|
167
|
-
|
168
|
-
|
188
|
+
agents = self.get_stage_agents(stage)
|
189
|
+
for agent in agents:
|
190
|
+
_I(f"Executing stage `{stage}` with agent `{type(agent).__name__}` ...")
|
191
|
+
failed, state = agent()
|
169
192
|
except Exception as e:
|
193
|
+
_W(f"Error during task execution stage `{stage}`: `{type(e)}`: `{e}`")
|
170
194
|
_M(f"**Error** during task execution stage `{stage}`: `{type(e)}`: `{e}`")
|
171
195
|
_M(f"```python\n{traceback.format_exc()}\n```")
|
172
196
|
state = TASK_AGENT_STATE_ERROR
|
@@ -174,16 +198,39 @@ class BaseTaskFlow:
|
|
174
198
|
stage_count += 1
|
175
199
|
stage_duration = time.time() - stage_st
|
176
200
|
flow_duration += stage_duration
|
177
|
-
|
178
|
-
|
179
|
-
|
180
|
-
|
181
|
-
|
182
|
-
|
183
|
-
|
184
|
-
|
201
|
+
_I(f"Stage `{stage}` completed in {stage_duration:.2f} seconds with state `{state}` and failed `{failed}`")
|
202
|
+
if evaluators := self.get_stage_evaluators(stage):
|
203
|
+
for evaluator in evaluators:
|
204
|
+
# If the agent has evaluators, run them
|
205
|
+
try:
|
206
|
+
_I(f"Evaluating stage `{stage}` with evaluator `{type(evaluator).__name__}` ...")
|
207
|
+
evaluation_result = evaluator()
|
208
|
+
evaluation_result.timestamp = evaluation_result.timestamp or time.time()
|
209
|
+
evaluation_result.evaluator = evaluation_result.evaluator or type(evaluator).__name__
|
210
|
+
evaluation_result.cell_index = self.task.cell_idx
|
211
|
+
evaluation_result.flow = type(self).__name__
|
212
|
+
evaluation_result.stage = str(stage)
|
213
|
+
evaluation_result.agent = type(agent).__name__
|
214
|
+
evaluation_result.execution_duration = stage_duration
|
215
|
+
evaluation_result.is_success = not failed
|
216
|
+
output_evaluation(evaluation_result)
|
217
|
+
except Exception as e:
|
218
|
+
_W(f"Error during task evaluation stage `{stage}`: `{type(e)}`: `{e}`")
|
219
|
+
_M(f"**Error** during task evaluation stage `{stage}`: `{type(e)}`: `{e}`")
|
220
|
+
_M(f"```python\n{traceback.format_exc()}\n```")
|
221
|
+
else:
|
222
|
+
output_evaluation(
|
223
|
+
StageEvaluationRecord(
|
224
|
+
timestamp=time.time(),
|
225
|
+
evaluator="default",
|
226
|
+
cell_index=self.task.cell_idx,
|
227
|
+
flow=type(self).__name__,
|
228
|
+
stage=str(stage),
|
229
|
+
agent=type(agent).__name__,
|
230
|
+
execution_duration=stage_duration,
|
231
|
+
is_success=not failed,
|
232
|
+
)
|
185
233
|
)
|
186
|
-
)
|
187
234
|
|
188
235
|
if state != TASK_AGENT_STATE_ERROR:
|
189
236
|
# Agent did not fail, check if we have reached the final stage
|
@@ -191,7 +238,7 @@ class BaseTaskFlow:
|
|
191
238
|
self.task.agent_stage = next_stage
|
192
239
|
self.task.update_cell()
|
193
240
|
if next_stage in self.STOP_STAGES:
|
194
|
-
|
241
|
+
_I(f"Task execution **Stopped** at stage `{next_stage}`")
|
195
242
|
stage = next_stage
|
196
243
|
break
|
197
244
|
|
@@ -212,37 +259,69 @@ class BaseTaskFlow:
|
|
212
259
|
self.task.agent_stage = next_stage
|
213
260
|
self.task.update_cell()
|
214
261
|
if action == TaskAction.STOP:
|
215
|
-
|
262
|
+
_I(f"Task execution **Stopped**, and set next stage to `{next_stage}`")
|
216
263
|
stage = next_stage
|
217
264
|
break
|
218
265
|
else:
|
219
|
-
|
266
|
+
_I(f"Action: `{action}` transits stage to `{next_stage}`")
|
220
267
|
stage = next_stage
|
221
268
|
else:
|
222
269
|
# transit to the next stage without confirmation
|
223
270
|
next_stage = self.get_next_stage(stage, state, TaskAction.CONTINUE)
|
224
271
|
self.task.agent_stage = next_stage
|
225
272
|
self.task.update_cell()
|
226
|
-
|
273
|
+
_I(f"Transits stage to `{next_stage}`")
|
227
274
|
stage = next_stage
|
228
|
-
|
229
275
|
if not stage_continue:
|
230
276
|
break
|
231
277
|
# Finalize the task execution
|
232
278
|
stage_name = stage.value if isinstance(stage, Enum) else stage
|
233
279
|
if stage_name == TASK_STAGE_GLOBAL_FINISHED:
|
234
280
|
_M("Task execution **finished** globally.")
|
235
|
-
|
281
|
+
if self.evaluator_factory is not None and hasattr(self, "GLOBAL_EVALUATOR") and self.GLOBAL_EVALUATOR:
|
282
|
+
evaluator = self.evaluator_factory(self.GLOBAL_EVALUATOR)
|
283
|
+
_I(f"Evaluating notebook with evaluator `{type(evaluator).__name__}` ...")
|
284
|
+
evaluation_result = evaluator()
|
285
|
+
evaluation_result.timestamp = evaluation_result.timestamp or time.time()
|
286
|
+
evaluation_result.evaluator = evaluation_result.evaluator or type(evaluator).__name__
|
287
|
+
evaluation_result.cell_index = self.task.cell_idx
|
288
|
+
evaluation_result.is_success = True
|
289
|
+
output_evaluation(evaluation_result)
|
290
|
+
else:
|
291
|
+
output_evaluation(
|
292
|
+
NotebookEvaluationRecord(
|
293
|
+
timestamp=time.time(),
|
294
|
+
evaluator="default",
|
295
|
+
cell_index=self.task.cell_idx,
|
296
|
+
is_success=True,
|
297
|
+
)
|
298
|
+
)
|
236
299
|
elif stage_name == TASK_STAGE_COMPLETED:
|
237
|
-
|
238
|
-
|
239
|
-
|
240
|
-
|
241
|
-
|
242
|
-
|
243
|
-
|
244
|
-
|
300
|
+
_I(f"Task execution **completed** in {flow_duration:.2f} seconds with {stage_count} stages.")
|
301
|
+
if self.evaluator_factory is not None and hasattr(self, "FLOW_EVALUATOR") and self.FLOW_EVALUATOR:
|
302
|
+
evaluator = self.evaluator_factory(self.FLOW_EVALUATOR)
|
303
|
+
_I(f"Evaluating flow `{type(self).__name__}` with evaluator `{type(evaluator).__name__}` ...")
|
304
|
+
evaluation_result = evaluator()
|
305
|
+
evaluation_result.timestamp = evaluation_result.timestamp or time.time()
|
306
|
+
evaluation_result.evaluator = evaluation_result.evaluator or type(evaluator).__name__
|
307
|
+
evaluation_result.cell_index = self.task.cell_idx
|
308
|
+
evaluation_result.flow = type(self).__name__
|
309
|
+
evaluation_result.stage_count = stage_count
|
310
|
+
evaluation_result.execution_duration = flow_duration
|
311
|
+
evaluation_result.is_success = True
|
312
|
+
output_evaluation(evaluation_result)
|
313
|
+
else:
|
314
|
+
# If no evaluator, just output the evaluation record
|
315
|
+
output_evaluation(
|
316
|
+
FlowEvaluationRecord(
|
317
|
+
timestamp=time.time(),
|
318
|
+
evaluator="default",
|
319
|
+
cell_index=self.task.cell_idx,
|
320
|
+
flow=type(self).__name__,
|
321
|
+
stage_count=stage_count,
|
322
|
+
execution_duration=flow_duration,
|
323
|
+
is_success=True,
|
324
|
+
)
|
245
325
|
)
|
246
|
-
)
|
247
326
|
flush_output()
|
248
327
|
return stage
|
@@ -5,13 +5,24 @@ This software is released under the MIT License.
|
|
5
5
|
https://opensource.org/licenses/MIT
|
6
6
|
"""
|
7
7
|
|
8
|
-
from .base import BaseTaskFlow,
|
8
|
+
from .base import BaseTaskFlow, StageNode, TASK_STAGE_START, TASK_STAGE_COMPLETED
|
9
|
+
from ..bot_evaluators.flow_global_planning import FlowGlobalPlanningEvaluator
|
10
|
+
from ..bot_agents.master_planner import MasterPlannerAgent
|
11
|
+
from ..bot_agents.output_task_result import OutputTaskResult
|
12
|
+
from ..bot_agents.prepare_next_cell import PrepareNextCell
|
13
|
+
from ..bot_evaluators.dummy_task import DummyTaskEvaluator
|
9
14
|
|
10
15
|
|
11
16
|
class MasterPlannerFlow(BaseTaskFlow):
|
12
17
|
|
13
|
-
|
14
|
-
|
15
|
-
|
18
|
+
STAGE_NODES = [
|
19
|
+
StageNode(
|
20
|
+
stage=TASK_STAGE_START,
|
21
|
+
agents=MasterPlannerAgent,
|
22
|
+
evaluators=DummyTaskEvaluator,
|
23
|
+
next_stage=TASK_STAGE_COMPLETED,
|
24
|
+
),
|
25
|
+
StageNode(stage=TASK_STAGE_COMPLETED, agents=OutputTaskResult, next_stage=TASK_STAGE_COMPLETED),
|
16
26
|
]
|
17
27
|
STOP_STAGES = [TASK_STAGE_COMPLETED]
|
28
|
+
FLOW_EVALUATOR = FlowGlobalPlanningEvaluator
|
@@ -8,32 +8,34 @@ https://opensource.org/licenses/MIT
|
|
8
8
|
from enum import Enum
|
9
9
|
from .base import (
|
10
10
|
BaseTaskFlow,
|
11
|
-
|
11
|
+
StageNode,
|
12
12
|
StageNext,
|
13
13
|
TaskAction,
|
14
14
|
TASK_STAGE_COMPLETED,
|
15
15
|
TASK_STAGE_GLOBAL_FINISHED,
|
16
16
|
)
|
17
|
-
from ..bot_agents import
|
18
|
-
|
19
|
-
|
20
|
-
|
21
|
-
|
22
|
-
|
23
|
-
|
24
|
-
|
25
|
-
|
26
|
-
from ..bot_agents.task_planner_v3 import TaskPlannerState
|
17
|
+
from ..bot_agents.task_planner_v3 import TaskPlannerAgentV3, TaskPlannerState
|
18
|
+
from ..bot_agents.task_coder import TaskCodingAgent
|
19
|
+
from ..bot_agents.task_debuger import CodeDebugerAgent
|
20
|
+
from ..bot_agents.task_code_executor import CodeExecutor
|
21
|
+
from ..bot_agents.task_structrue_summarier import TaskStructureSummaryAgent, TaskStructureSummaryState
|
22
|
+
from ..bot_agents.task_structrue_reasoner import TaskStructureReasoningAgent, TaskStructureReasonState
|
23
|
+
from ..bot_agents.output_task_result import OutputTaskResult
|
24
|
+
from ..bot_agents.request_user_supply import RequestAboveUserSupplyAgent, RequestBelowUserSupplyAgent
|
25
|
+
from ..bot_agents.prepare_next_cell import PrepareNextCell
|
27
26
|
|
28
27
|
|
29
28
|
class TaskStage(str, Enum):
|
30
29
|
PLANNING = "planning"
|
31
30
|
PLANNING_PAUSED = "planning_paused"
|
31
|
+
REQUEST_INFO_ABOVE = "request_info_above"
|
32
|
+
REQUEST_INFO_BELOW = "request_info_below"
|
32
33
|
CODING = "coding"
|
33
34
|
EXECUTING = "executing"
|
34
35
|
DEBUGGING = "debugging"
|
35
36
|
REASONING = "reasoning"
|
36
37
|
SUMMARY = "summary"
|
38
|
+
PREPARE_NEXT = "prepare_next"
|
37
39
|
OUTPUT_RESULT = "output_result"
|
38
40
|
COMPLETED = TASK_STAGE_COMPLETED
|
39
41
|
GLOBAL_FINISHED = TASK_STAGE_GLOBAL_FINISHED
|
@@ -43,20 +45,25 @@ class TaskExecutorFlowV3(BaseTaskFlow):
|
|
43
45
|
|
44
46
|
START_STAGE = TaskStage.PLANNING
|
45
47
|
STOP_STAGES = [TaskStage.COMPLETED, TaskStage.PLANNING_PAUSED, TaskStage.GLOBAL_FINISHED]
|
46
|
-
|
47
|
-
|
48
|
+
STAGE_NODES = [
|
49
|
+
StageNode[TaskStage, TaskPlannerState](
|
48
50
|
stage=TaskStage.PLANNING,
|
49
|
-
|
51
|
+
agents=TaskPlannerAgentV3,
|
50
52
|
states={
|
51
53
|
TaskPlannerState.CODING_PLANNED: TaskStage.CODING,
|
52
54
|
TaskPlannerState.REASONING_PLANNED: TaskStage.REASONING,
|
53
|
-
TaskPlannerState.REQUEST_INFO: TaskStage.
|
55
|
+
TaskPlannerState.REQUEST_INFO: TaskStage.REQUEST_INFO_ABOVE,
|
54
56
|
TaskPlannerState.GLOBAL_FINISHED: TaskStage.GLOBAL_FINISHED,
|
55
57
|
},
|
56
58
|
),
|
57
|
-
|
59
|
+
StageNode[TaskStage, None](
|
60
|
+
stage=TaskStage.REQUEST_INFO_ABOVE,
|
61
|
+
agents=RequestAboveUserSupplyAgent,
|
62
|
+
next_stage=TaskStage.PLANNING_PAUSED,
|
63
|
+
),
|
64
|
+
StageNode[TaskStage, TaskPlannerState](
|
58
65
|
stage=TaskStage.PLANNING_PAUSED,
|
59
|
-
|
66
|
+
agents=TaskPlannerAgentV3,
|
60
67
|
states={
|
61
68
|
TaskPlannerState.CODING_PLANNED: TaskStage.CODING,
|
62
69
|
TaskPlannerState.REASONING_PLANNED: TaskStage.REASONING,
|
@@ -64,37 +71,49 @@ class TaskExecutorFlowV3(BaseTaskFlow):
|
|
64
71
|
TaskPlannerState.GLOBAL_FINISHED: TaskStage.COMPLETED,
|
65
72
|
},
|
66
73
|
),
|
67
|
-
|
68
|
-
|
69
|
-
),
|
70
|
-
StageTransition[TaskStage, bool](
|
74
|
+
StageNode[TaskStage, None](stage=TaskStage.CODING, agents=TaskCodingAgent, next_stage=TaskStage.EXECUTING),
|
75
|
+
StageNode[TaskStage, bool](
|
71
76
|
stage=TaskStage.EXECUTING,
|
72
|
-
|
77
|
+
agents=CodeExecutor,
|
73
78
|
states={True: TaskStage.SUMMARY, False: TaskStage.DEBUGGING},
|
74
79
|
),
|
75
|
-
|
76
|
-
|
77
|
-
|
78
|
-
|
79
|
-
|
80
|
+
StageNode[TaskStage, None](stage=TaskStage.DEBUGGING, agents=CodeDebugerAgent, next_stage=TaskStage.EXECUTING),
|
81
|
+
StageNode[TaskStage, TaskStructureReasonState](
|
82
|
+
stage=TaskStage.REASONING,
|
83
|
+
agents=TaskStructureReasoningAgent,
|
84
|
+
states={
|
85
|
+
TaskStructureReasonState.DONE: TaskStage.COMPLETED,
|
86
|
+
TaskStructureReasonState.REQUEST_INFO: TaskStage.REQUEST_INFO_BELOW,
|
87
|
+
},
|
80
88
|
),
|
81
|
-
|
89
|
+
StageNode[TaskStage, TaskStructureSummaryState](
|
82
90
|
stage=TaskStage.SUMMARY,
|
83
|
-
|
84
|
-
|
85
|
-
|
86
|
-
|
91
|
+
agents=TaskStructureSummaryAgent,
|
92
|
+
states={
|
93
|
+
TaskStructureSummaryState.DONE: {
|
94
|
+
TaskAction.DEFAULT: StageNext(stage=TaskStage.PREPARE_NEXT),
|
95
|
+
TaskAction.STOP: StageNext(stage=TaskStage.EXECUTING),
|
96
|
+
},
|
97
|
+
TaskStructureSummaryState.REQUEST_INFO: TaskStage.REQUEST_INFO_BELOW,
|
87
98
|
},
|
88
99
|
),
|
89
|
-
|
100
|
+
StageNode[TaskStage, None](
|
101
|
+
stage=TaskStage.PREPARE_NEXT, agents=PrepareNextCell, next_stage=TaskStage.COMPLETED
|
102
|
+
),
|
103
|
+
StageNode[TaskStage, None](
|
104
|
+
stage=TaskStage.REQUEST_INFO_BELOW,
|
105
|
+
agents=[PrepareNextCell, RequestBelowUserSupplyAgent],
|
106
|
+
next_stage=TaskStage.COMPLETED,
|
107
|
+
),
|
108
|
+
StageNode[TaskStage, bool](
|
90
109
|
stage=TaskStage.COMPLETED,
|
91
|
-
|
110
|
+
agents=CodeExecutor,
|
92
111
|
states={True: TaskStage.OUTPUT_RESULT, False: TaskStage.DEBUGGING},
|
93
112
|
),
|
94
|
-
|
95
|
-
stage=TaskStage.OUTPUT_RESULT,
|
113
|
+
StageNode[TaskStage, None](
|
114
|
+
stage=TaskStage.OUTPUT_RESULT, agents=OutputTaskResult, next_stage=TaskStage.COMPLETED
|
96
115
|
),
|
97
|
-
|
98
|
-
stage=TaskStage.GLOBAL_FINISHED,
|
116
|
+
StageNode[TaskStage, None](
|
117
|
+
stage=TaskStage.GLOBAL_FINISHED, agents=OutputTaskResult, next_stage=TaskStage.GLOBAL_FINISHED
|
99
118
|
),
|
100
119
|
]
|