coflows compatible+
Browse files- ControllerAtomicFlow.py +22 -8
- ControllerAtomicFlow.yaml +4 -0
- ControllerExecutorFlow.py +118 -35
- ControllerExecutorFlow.yaml +4 -6
- WikiSearchAtomicFlow.py +12 -8
- __init__.py +1 -1
- demo.yaml +4 -6
- run.py +70 -46
ControllerAtomicFlow.py
CHANGED
@@ -2,7 +2,7 @@ import json
|
|
2 |
from copy import deepcopy
|
3 |
from typing import Any, Dict, List
|
4 |
from flow_modules.aiflows.ChatFlowModule import ChatAtomicFlow
|
5 |
-
|
6 |
from dataclasses import dataclass
|
7 |
|
8 |
|
@@ -117,14 +117,28 @@ class ControllerAtomicFlow(ChatAtomicFlow):
|
|
117 |
# ~~~ Instantiate flow ~~~
|
118 |
return cls(**kwargs)
|
119 |
|
120 |
-
def run(self,
|
121 |
""" This method runs the flow. Note that the response of the LLM is in the JSON format, but it's not a hard constraint (it can hallucinate and return an invalid JSON)
|
122 |
|
123 |
-
:param
|
124 |
-
:type
|
125 |
-
:return: The output data of the flow (thought, reasoning, criticism, command, command_args)
|
126 |
-
:rtype: Dict[str, Any]
|
127 |
"""
|
128 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
129 |
response = json.loads(api_output)
|
130 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2 |
from copy import deepcopy
|
3 |
from typing import Any, Dict, List
|
4 |
from flow_modules.aiflows.ChatFlowModule import ChatAtomicFlow
|
5 |
+
from aiflows.messages import FlowMessage
|
6 |
from dataclasses import dataclass
|
7 |
|
8 |
|
|
|
117 |
# ~~~ Instantiate flow ~~~
|
118 |
return cls(**kwargs)
|
119 |
|
120 |
+
def run(self, input_message: FlowMessage):
|
121 |
""" This method runs the flow. Note that the response of the LLM is in the JSON format, but it's not a hard constraint (it can hallucinate and return an invalid JSON)
|
122 |
|
123 |
+
:param input_message: The input data of the flow.
|
124 |
+
:type input_message: FlowMessage
|
|
|
|
|
125 |
"""
|
126 |
+
|
127 |
+
input_data = input_message.data
|
128 |
+
|
129 |
+
if "goal" in input_data:
|
130 |
+
self.flow_state["goal"] = input_data["goal"]
|
131 |
+
|
132 |
+
else:
|
133 |
+
input_data["goal"] = self.flow_state["goal"]
|
134 |
+
|
135 |
+
api_output = self.query_llm(input_data)
|
136 |
+
|
137 |
response = json.loads(api_output)
|
138 |
+
|
139 |
+
reply = self._package_output_message(
|
140 |
+
input_message=input_message,
|
141 |
+
response=response
|
142 |
+
)
|
143 |
+
|
144 |
+
self.reply_to_message(reply = reply, to = input_message)
|
ControllerAtomicFlow.yaml
CHANGED
@@ -11,6 +11,7 @@ input_interface_non_initialized: # initial input keys
|
|
11 |
|
12 |
input_interface_initialized: # input_keys
|
13 |
- "observation"
|
|
|
14 |
|
15 |
#######################################################
|
16 |
# Output keys
|
@@ -67,10 +68,13 @@ system_message_prompt_template:
|
|
67 |
human_message_prompt_template:
|
68 |
_target_: aiflows.prompt_template.JinjaPrompt
|
69 |
template: |2-
|
|
|
|
|
70 |
Here is the response to your last action:
|
71 |
{{observation}}
|
72 |
input_variables:
|
73 |
- "observation"
|
|
|
74 |
|
75 |
init_human_message_prompt_template:
|
76 |
_target_: aiflows.prompt_template.JinjaPrompt
|
|
|
11 |
|
12 |
input_interface_initialized: # input_keys
|
13 |
- "observation"
|
14 |
+
- "goal"
|
15 |
|
16 |
#######################################################
|
17 |
# Output keys
|
|
|
68 |
human_message_prompt_template:
|
69 |
_target_: aiflows.prompt_template.JinjaPrompt
|
70 |
template: |2-
|
71 |
+
Here is the goal you need to achieve:
|
72 |
+
{{goal}}
|
73 |
Here is the response to your last action:
|
74 |
{{observation}}
|
75 |
input_variables:
|
76 |
- "observation"
|
77 |
+
- "goal"
|
78 |
|
79 |
init_human_message_prompt_template:
|
80 |
_target_: aiflows.prompt_template.JinjaPrompt
|
ControllerExecutorFlow.py
CHANGED
@@ -4,6 +4,7 @@ from aiflows.base_flows import CompositeFlow
|
|
4 |
from aiflows.utils import logging
|
5 |
|
6 |
from .ControllerAtomicFlow import ControllerAtomicFlow
|
|
|
7 |
from aiflows.interfaces import KeyInterface
|
8 |
logging.set_verbosity_debug()
|
9 |
log = logging.get_logger(__name__)
|
@@ -65,51 +66,133 @@ class ControllerExecutorFlow(CompositeFlow):
|
|
65 |
:param subflows: A list of subflows. Required when instantiating the subflow programmatically (it replaces subflows_config from flow_config).
|
66 |
"""
|
67 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
68 |
def _on_reach_max_round(self):
|
69 |
""" This method is called when the flow reaches the maximum amount of rounds. It updates the state of the flow and starts the process of terminating the flow."""
|
70 |
self._state_update_dict({
|
|
|
71 |
"answer": "The maximum amount of rounds was reached before the model found an answer.",
|
72 |
"status": "unfinished"
|
73 |
})
|
74 |
|
75 |
-
def
|
|
|
|
|
|
|
76 |
|
77 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
78 |
|
79 |
-
|
80 |
|
81 |
-
|
82 |
-
return {
|
83 |
-
"EARLY_EXIT": True,
|
84 |
-
"answer": controller_reply["command_args"]["answer"],
|
85 |
-
"status": "finished"
|
86 |
-
}
|
87 |
|
88 |
-
|
89 |
-
|
90 |
-
}
|
91 |
|
92 |
-
|
93 |
-
|
94 |
-
|
95 |
-
|
96 |
-
|
97 |
-
|
98 |
-
|
99 |
-
|
100 |
-
"executor_reply": input_data,
|
101 |
-
"controller_reply": None
|
102 |
-
}
|
103 |
-
|
104 |
-
for round in range(self.flow_config["max_rounds"]):
|
105 |
-
reply = self._single_round_controller_executor(reply)
|
106 |
|
107 |
-
|
108 |
-
|
109 |
-
|
110 |
-
|
111 |
-
|
112 |
-
|
113 |
-
|
114 |
-
|
115 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
4 |
from aiflows.utils import logging
|
5 |
|
6 |
from .ControllerAtomicFlow import ControllerAtomicFlow
|
7 |
+
from aiflows.messages import FlowMessage
|
8 |
from aiflows.interfaces import KeyInterface
|
9 |
logging.set_verbosity_debug()
|
10 |
log = logging.get_logger(__name__)
|
|
|
66 |
:param subflows: A list of subflows. Required when instantiating the subflow programmatically (it replaces subflows_config from flow_config).
|
67 |
"""
|
68 |
|
69 |
+
def __init__(self,**kwargs):
|
70 |
+
super().__init__(**kwargs)
|
71 |
+
|
72 |
+
self.input_interface_controller = KeyInterface(
|
73 |
+
keys_to_select = ["goal","observation"],
|
74 |
+
)
|
75 |
+
self.input_interface_first_round_controller = KeyInterface(
|
76 |
+
keys_to_select = ["goal"],
|
77 |
+
)
|
78 |
+
|
79 |
+
self.reply_interface = KeyInterface(
|
80 |
+
keys_to_select = ["answer","status", "EARLY_EXIT"],
|
81 |
+
)
|
82 |
+
|
83 |
+
self.next_flow_to_call = {
|
84 |
+
None: "Controller",
|
85 |
+
"Controller": "Executor",
|
86 |
+
"Executor": "Controller"
|
87 |
+
}
|
88 |
+
|
89 |
+
def generate_reply(self):
|
90 |
+
""" This method generates the reply of the flow. It's called when the flow is finished. """
|
91 |
+
|
92 |
+
reply = self._package_output_message(
|
93 |
+
input_message = self.flow_state["input_message"],
|
94 |
+
response = self.reply_interface(self.flow_state)
|
95 |
+
)
|
96 |
+
self.reply_to_message(reply,to=self.flow_state["input_message"])
|
97 |
+
|
98 |
+
def get_next_flow_to_call(self):
|
99 |
+
|
100 |
+
if self.flow_config["max_rounds"] is not None and self.flow_state["current_round"] >= self.flow_config["max_rounds"]:
|
101 |
+
return None
|
102 |
+
|
103 |
+
return self.next_flow_to_call[self.flow_state["last_called"]]
|
104 |
+
|
105 |
def _on_reach_max_round(self):
|
106 |
""" This method is called when the flow reaches the maximum amount of rounds. It updates the state of the flow and starts the process of terminating the flow."""
|
107 |
self._state_update_dict({
|
108 |
+
"EARLY_EXIT": False,
|
109 |
"answer": "The maximum amount of rounds was reached before the model found an answer.",
|
110 |
"status": "unfinished"
|
111 |
})
|
112 |
|
113 |
+
def set_up_flow_state(self):
|
114 |
+
super().set_up_flow_state()
|
115 |
+
self.flow_state["last_called"] = None
|
116 |
+
self.flow_state["current_round"] = 0
|
117 |
|
118 |
+
|
119 |
+
def call_controller(self):
|
120 |
+
#first_round
|
121 |
+
if self.flow_state["last_called"] is None:
|
122 |
+
input_interface = self.input_interface_first_round_controller
|
123 |
+
else:
|
124 |
+
input_interface = self.input_interface_controller
|
125 |
+
|
126 |
+
message = self._package_input_message(
|
127 |
+
data = input_interface(self.flow_state),
|
128 |
+
dst_flow = "Controller"
|
129 |
+
)
|
130 |
|
131 |
+
self.subflows["Controller"].send_message_async(message, pipe_to= self.flow_config["flow_ref"])
|
132 |
|
133 |
+
def call_executor(self):
|
|
|
|
|
|
|
|
|
|
|
134 |
|
135 |
+
#detect and early exit
|
136 |
+
if self.flow_state["command"] == "finish":
|
|
|
137 |
|
138 |
+
self._state_update_dict(
|
139 |
+
{
|
140 |
+
"EARLY_EXIT": True,
|
141 |
+
"answer": self.flow_state["command_args"]["answer"],
|
142 |
+
"status": "finished"
|
143 |
+
}
|
144 |
+
)
|
145 |
+
self.generate_reply()
|
|
|
|
|
|
|
|
|
|
|
|
|
146 |
|
147 |
+
#call executor
|
148 |
+
else:
|
149 |
+
executor_branch_to_call = self.flow_state["command"]
|
150 |
+
message = self._package_input_message(
|
151 |
+
data = self.flow_state["command_args"],
|
152 |
+
dst_flow = executor_branch_to_call
|
153 |
+
)
|
154 |
+
|
155 |
+
self.subflows[executor_branch_to_call].send_message_async(message, pipe_to= self.flow_config["flow_ref"])
|
156 |
+
|
157 |
+
|
158 |
+
def register_data_to_state(self, input_message):
|
159 |
+
last_called = self.flow_state["last_called"]
|
160 |
+
|
161 |
+
if last_called is None:
|
162 |
+
self.flow_state["input_message"] = input_message
|
163 |
+
self.flow_state["goal"] = input_message.data["goal"]
|
164 |
+
|
165 |
+
elif last_called == "Executor":
|
166 |
+
self.flow_state["observation"] = input_message.data
|
167 |
+
|
168 |
+
else:
|
169 |
+
self._state_update_dict(
|
170 |
+
{
|
171 |
+
"command": input_message.data["command"],
|
172 |
+
"command_args": input_message.data["command_args"]
|
173 |
+
}
|
174 |
+
)
|
175 |
+
|
176 |
+
def run(self,input_message: FlowMessage):
|
177 |
+
""" Runs the WikiSearch Atomic Flow. It's used to execute a Wikipedia search and get page summaries.
|
178 |
+
|
179 |
+
:param input_message: The input message
|
180 |
+
:type input_message: FlowMessage
|
181 |
+
"""
|
182 |
+
|
183 |
+
self.register_data_to_state(input_message)
|
184 |
+
|
185 |
+
flow_to_call = self.get_next_flow_to_call()
|
186 |
+
|
187 |
+
if flow_to_call == "Controller":
|
188 |
+
self.flow_state["observation"] = input_message.data
|
189 |
+
self.call_controller()
|
190 |
+
elif flow_to_call == "Executor":
|
191 |
+
self.call_executor()
|
192 |
+
self.flow_state["current_round"] += 1
|
193 |
+
else:
|
194 |
+
self._on_reach_max_round()
|
195 |
+
self.generate_reply()
|
196 |
+
|
197 |
+
self.flow_state["last_called"] = flow_to_call
|
198 |
+
|
ControllerExecutorFlow.yaml
CHANGED
@@ -15,9 +15,10 @@ subflows_config:
|
|
15 |
name: "ControllerAtomicFlow"
|
16 |
description: "A flow that calls other flows to solve a problem."
|
17 |
_target_: flow_modules.aiflows.ControllerAtomicFlow.instantiate_from_default_config
|
18 |
-
|
19 |
-
|
20 |
-
|
|
|
21 |
# E.g.,
|
22 |
# commands:
|
23 |
# wiki_search:
|
@@ -29,6 +30,3 @@ subflows_config:
|
|
29 |
# wiki_search:
|
30 |
# _target_: .WikiSearchAtomicFlow.instantiate_from_default_config
|
31 |
|
32 |
-
|
33 |
-
early_exit_key: "EARLY_EXIT"
|
34 |
-
|
|
|
15 |
name: "ControllerAtomicFlow"
|
16 |
description: "A flow that calls other flows to solve a problem."
|
17 |
_target_: flow_modules.aiflows.ControllerAtomicFlow.instantiate_from_default_config
|
18 |
+
commands:
|
19 |
+
finish:
|
20 |
+
description: "Signal that the objective has been satisfied, and returns the answer to the user."
|
21 |
+
input_args: ["answer"]
|
22 |
# E.g.,
|
23 |
# commands:
|
24 |
# wiki_search:
|
|
|
30 |
# wiki_search:
|
31 |
# _target_: .WikiSearchAtomicFlow.instantiate_from_default_config
|
32 |
|
|
|
|
|
|
WikiSearchAtomicFlow.py
CHANGED
@@ -3,7 +3,7 @@ from copy import deepcopy
|
|
3 |
from typing import List, Dict, Optional, Any
|
4 |
|
5 |
from aiflows.base_flows import AtomicFlow
|
6 |
-
|
7 |
from aiflows.utils import logging
|
8 |
from .wikipediaAPI import WikipediaAPIWrapper
|
9 |
|
@@ -44,15 +44,13 @@ class WikiSearchAtomicFlow(AtomicFlow):
|
|
44 |
super().__init__(**kwargs)
|
45 |
|
46 |
def run(self,
|
47 |
-
|
48 |
""" Runs the WikiSearch Atomic Flow. It's used to execute a Wikipedia search and get page summaries.
|
49 |
|
50 |
-
:param
|
51 |
-
:type
|
52 |
-
:return: The output data dictionary
|
53 |
-
:rtype: Dict[str, Any]
|
54 |
"""
|
55 |
-
|
56 |
# ~~~ Process input ~~~
|
57 |
term = input_data.get("search_term", None)
|
58 |
api_wrapper = WikipediaAPIWrapper(
|
@@ -70,4 +68,10 @@ class WikiSearchAtomicFlow(AtomicFlow):
|
|
70 |
|
71 |
# Log the update to the flow messages list
|
72 |
observation = search_response["wiki_content"] if search_response["wiki_content"] else search_response["relevant_pages"]
|
73 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
3 |
from typing import List, Dict, Optional, Any
|
4 |
|
5 |
from aiflows.base_flows import AtomicFlow
|
6 |
+
from aiflows.messages import FlowMessage
|
7 |
from aiflows.utils import logging
|
8 |
from .wikipediaAPI import WikipediaAPIWrapper
|
9 |
|
|
|
44 |
super().__init__(**kwargs)
|
45 |
|
46 |
def run(self,
|
47 |
+
input_message: FlowMessage):
|
48 |
""" Runs the WikiSearch Atomic Flow. It's used to execute a Wikipedia search and get page summaries.
|
49 |
|
50 |
+
:param input_message: The input message
|
51 |
+
:type input_message: FlowMessage
|
|
|
|
|
52 |
"""
|
53 |
+
input_data = input_message.data
|
54 |
# ~~~ Process input ~~~
|
55 |
term = input_data.get("search_term", None)
|
56 |
api_wrapper = WikipediaAPIWrapper(
|
|
|
68 |
|
69 |
# Log the update to the flow messages list
|
70 |
observation = search_response["wiki_content"] if search_response["wiki_content"] else search_response["relevant_pages"]
|
71 |
+
|
72 |
+
reply = self._package_output_message(
|
73 |
+
input_message = input_message,
|
74 |
+
response = {"wiki_content": observation},
|
75 |
+
)
|
76 |
+
|
77 |
+
self.reply_to_message(reply=reply, to=input_message)
|
__init__.py
CHANGED
@@ -1,6 +1,6 @@
|
|
1 |
# ~~~ Specify the dependencies ~~~
|
2 |
dependencies = [
|
3 |
-
{"url": "aiflows/ChatFlowModule", "revision": "
|
4 |
]
|
5 |
from aiflows import flow_verse
|
6 |
|
|
|
1 |
# ~~~ Specify the dependencies ~~~
|
2 |
dependencies = [
|
3 |
+
{"url": "aiflows/ChatFlowModule", "revision": "coflows"},
|
4 |
]
|
5 |
from aiflows import flow_verse
|
6 |
|
demo.yaml
CHANGED
@@ -9,12 +9,10 @@ subflows_config:
|
|
9 |
wiki_search:
|
10 |
description: "Performs a search on Wikipedia."
|
11 |
input_args: ["search_term"]
|
12 |
-
ddg_search:
|
13 |
-
|
14 |
-
|
15 |
-
|
16 |
-
description: "Signal that the objective has been satisfied, and returns the answer to the user."
|
17 |
-
input_args: ["answer"]
|
18 |
backend:
|
19 |
_target_: aiflows.backends.llm_lite.LiteLLMBackend
|
20 |
api_infos: ???
|
|
|
9 |
wiki_search:
|
10 |
description: "Performs a search on Wikipedia."
|
11 |
input_args: ["search_term"]
|
12 |
+
# ddg_search:
|
13 |
+
# description: "Query the search engine DuckDuckGo."
|
14 |
+
# input_args: ["query"]
|
15 |
+
|
|
|
|
|
16 |
backend:
|
17 |
_target_: aiflows.backends.llm_lite.LiteLLMBackend
|
18 |
api_infos: ???
|
run.py
CHANGED
@@ -1,3 +1,5 @@
|
|
|
|
|
|
1 |
import os
|
2 |
|
3 |
import hydra
|
@@ -9,87 +11,109 @@ from aiflows.utils.general_helpers import read_yaml_file, quick_load_api_keys
|
|
9 |
|
10 |
from aiflows import logging
|
11 |
from aiflows.flow_cache import CACHING_PARAMETERS, clear_cache
|
|
|
12 |
from aiflows.utils import serve_utils
|
13 |
from aiflows.workers import run_dispatch_worker_thread
|
14 |
from aiflows.messages import FlowMessage
|
15 |
from aiflows.interfaces import KeyInterface
|
|
|
|
|
16 |
|
17 |
CACHING_PARAMETERS.do_caching = False # Set to True in order to disable caching
|
18 |
# clear_cache() # Uncomment this line to clear the cache
|
19 |
|
20 |
logging.set_verbosity_debug()
|
21 |
|
22 |
-
|
23 |
-
# ~~~ Load Flow dependecies from FlowVerse ~~~
|
24 |
dependencies = [
|
25 |
-
{"url": "aiflows/ControllerExecutorFlowModule", "revision":
|
26 |
]
|
27 |
|
|
|
28 |
flow_verse.sync_dependencies(dependencies)
|
29 |
-
|
30 |
if __name__ == "__main__":
|
31 |
-
# ~~~ Set the API information ~~~
|
32 |
-
# OpenAI backend
|
33 |
-
api_information = [ApiInfo(backend_used="openai", api_key=os.getenv("OPENAI_API_KEY"))]
|
34 |
-
# Azure backend
|
35 |
-
# api_information = [ApiInfo(backend_used = "azure",
|
36 |
-
# api_base = os.getenv("AZURE_API_BASE"),
|
37 |
-
# api_key = os.getenv("AZURE_OPENAI_KEY"),
|
38 |
-
# api_version = os.getenv("AZURE_API_VERSION") )]
|
39 |
-
|
40 |
-
FLOW_MODULES_PATH = "./"
|
41 |
|
42 |
-
|
43 |
-
|
44 |
|
45 |
-
cl =
|
46 |
-
|
47 |
-
{"jwt": jwt, "addr": addr}
|
48 |
-
)
|
49 |
-
# path_to_output_file = "output.jsonl" # Uncomment this line to save the output to disk
|
50 |
|
|
|
51 |
root_dir = "."
|
52 |
cfg_path = os.path.join(root_dir, "demo.yaml")
|
53 |
cfg = read_yaml_file(cfg_path)
|
54 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
55 |
serve_utils.recursive_serve_flow(
|
56 |
cl = cl,
|
57 |
-
flow_type="
|
58 |
default_config=cfg,
|
59 |
default_state=None,
|
60 |
-
default_dispatch_point="coflows_dispatch"
|
61 |
)
|
62 |
|
63 |
-
#
|
64 |
-
|
65 |
-
|
66 |
-
quick_load_api_keys(cfg, api_information, key="api_infos")
|
67 |
|
68 |
-
#
|
69 |
-
# This can be a list of samples
|
70 |
-
# data = {"id": 0, "goal": "Answer the following question: What is the population of Canada?"} # Uses wikipedia
|
71 |
-
# data = {"id": 0, "goal": "Answer the following question: Who was the NBA champion in 2023?"}
|
72 |
-
data = {
|
73 |
-
"id": 0,
|
74 |
-
"goal": "Answer the following question: What is the profession and date of birth of Michael Jordan?",
|
75 |
-
}
|
76 |
-
# ~~~ Run inference ~~~
|
77 |
proxy_flow = serve_utils.recursive_mount(
|
78 |
cl=cl,
|
79 |
client_id="local",
|
80 |
-
flow_type="
|
81 |
-
config_overrides=
|
82 |
initial_state=None,
|
83 |
dispatch_point_override=None,
|
84 |
)
|
85 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
86 |
input_message = FlowMessage(
|
87 |
-
data=
|
88 |
-
src_flow="Coflows team",
|
89 |
-
dst_flow=proxy_flow,
|
90 |
-
is_input_msg=True
|
91 |
)
|
|
|
|
|
|
|
|
|
|
|
|
|
92 |
|
93 |
-
|
|
|
|
|
94 |
|
95 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
"""A simple script to run a Flow that can be used for development and debugging."""
|
2 |
+
|
3 |
import os
|
4 |
|
5 |
import hydra
|
|
|
11 |
|
12 |
from aiflows import logging
|
13 |
from aiflows.flow_cache import CACHING_PARAMETERS, clear_cache
|
14 |
+
|
15 |
from aiflows.utils import serve_utils
|
16 |
from aiflows.workers import run_dispatch_worker_thread
|
17 |
from aiflows.messages import FlowMessage
|
18 |
from aiflows.interfaces import KeyInterface
|
19 |
+
from aiflows.utils.colink_utils import start_colink_server
|
20 |
+
from aiflows.workers import run_dispatch_worker_thread
|
21 |
|
22 |
CACHING_PARAMETERS.do_caching = False # Set to True in order to disable caching
|
23 |
# clear_cache() # Uncomment this line to clear the cache
|
24 |
|
25 |
logging.set_verbosity_debug()
|
26 |
|
27 |
+
|
|
|
28 |
dependencies = [
|
29 |
+
{"url": "aiflows/ControllerExecutorFlowModule", "revision": os.getcwd()}
|
30 |
]
|
31 |
|
32 |
+
from aiflows import flow_verse
|
33 |
flow_verse.sync_dependencies(dependencies)
|
|
|
34 |
if __name__ == "__main__":
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
35 |
|
36 |
+
#1. ~~~~~ Set up a colink server ~~~~
|
37 |
+
FLOW_MODULES_PATH = "./"
|
38 |
|
39 |
+
cl = start_colink_server()
|
40 |
+
|
|
|
|
|
|
|
41 |
|
42 |
+
#2. ~~~~~Load flow config~~~~~~
|
43 |
root_dir = "."
|
44 |
cfg_path = os.path.join(root_dir, "demo.yaml")
|
45 |
cfg = read_yaml_file(cfg_path)
|
46 |
+
|
47 |
+
#2.1 ~~~ Set the API information ~~~
|
48 |
+
# OpenAI backend
|
49 |
+
api_information = [ApiInfo(backend_used="openai",
|
50 |
+
api_key = os.getenv("OPENAI_API_KEY"))]
|
51 |
+
# # Azure backend
|
52 |
+
# api_information = ApiInfo(backend_used = "azure",
|
53 |
+
# api_base = os.getenv("AZURE_API_BASE"),
|
54 |
+
# api_key = os.getenv("AZURE_OPENAI_KEY"),
|
55 |
+
# api_version = os.getenv("AZURE_API_VERSION") )
|
56 |
+
|
57 |
+
|
58 |
+
quick_load_api_keys(cfg, api_information, key="api_infos")
|
59 |
+
|
60 |
+
|
61 |
+
#3. ~~~~ Serve The Flow ~~~~
|
62 |
serve_utils.recursive_serve_flow(
|
63 |
cl = cl,
|
64 |
+
flow_type="ControllerExecutorFlowModule",
|
65 |
default_config=cfg,
|
66 |
default_state=None,
|
67 |
+
default_dispatch_point="coflows_dispatch"
|
68 |
)
|
69 |
|
70 |
+
#4. ~~~~~Start A Worker Thread~~~~~
|
71 |
+
run_dispatch_worker_thread(cl, dispatch_point="coflows_dispatch", flow_modules_base_path=FLOW_MODULES_PATH)
|
|
|
|
|
72 |
|
73 |
+
#5. ~~~~~Mount the flow and get its proxy~~~~~~
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
74 |
proxy_flow = serve_utils.recursive_mount(
|
75 |
cl=cl,
|
76 |
client_id="local",
|
77 |
+
flow_type="ControllerExecutorFlowModule",
|
78 |
+
config_overrides=None,
|
79 |
initial_state=None,
|
80 |
dispatch_point_override=None,
|
81 |
)
|
82 |
+
|
83 |
+
#6. ~~~ Get the data ~~~
|
84 |
+
data = {
|
85 |
+
"id": 0,
|
86 |
+
"goal": "Answer the following question: What is the profession and date of birth of Michael Jordan?",
|
87 |
+
}
|
88 |
+
|
89 |
+
|
90 |
+
#option1: use the FlowMessage class
|
91 |
input_message = FlowMessage(
|
92 |
+
data=data,
|
|
|
|
|
|
|
93 |
)
|
94 |
+
|
95 |
+
#option2: use the proxy_flow
|
96 |
+
#input_message = proxy_flow._package_input_message(data = data)
|
97 |
+
|
98 |
+
#7. ~~~ Run inference ~~~
|
99 |
+
future = proxy_flow.send_message_blocking(input_message)
|
100 |
|
101 |
+
#uncomment this line if you would like to get the full message back
|
102 |
+
#reply_message = future.get_message()
|
103 |
+
reply_data = future.get_data()
|
104 |
|
105 |
+
# ~~~ Print the output ~~~
|
106 |
+
print("~~~~~~Reply~~~~~~")
|
107 |
+
print(reply_data)
|
108 |
+
|
109 |
+
|
110 |
+
#8. ~~~~ (Optional) apply output interface on reply ~~~~
|
111 |
+
# output_interface = KeyInterface(
|
112 |
+
# keys_to_rename={"api_output": "answer"},
|
113 |
+
# )
|
114 |
+
# print("Output: ", output_interface(reply_data))
|
115 |
+
|
116 |
+
|
117 |
+
#9. ~~~~~Optional: Unserve Flow~~~~~~
|
118 |
+
# serve_utils.delete_served_flow(cl, "FlowModule")
|
119 |
+
|