Spaces:
Sleeping
Sleeping
handle audio, remove extra sub-agent
Browse files- basic_agent.py +19 -25
- basic_agent_test.py +9 -3
basic_agent.py
CHANGED
|
@@ -26,12 +26,13 @@ Remember - The FINAL ANSWER must be a string, a number or a comma separated list
|
|
| 26 |
|
| 27 |
@smolagents.tool
|
| 28 |
def convert_from_url(url: str) -> str:
|
| 29 |
-
"""Convert
|
| 30 |
|
| 31 |
-
This method takes a
|
| 32 |
-
a markdown
|
| 33 |
-
|
| 34 |
-
|
|
|
|
| 35 |
|
| 36 |
Args:
|
| 37 |
url: the url to access the file. If the file is on the
|
|
@@ -42,7 +43,7 @@ def convert_from_url(url: str) -> str:
|
|
| 42 |
string that is in the markdown format.
|
| 43 |
"""
|
| 44 |
# Don't need third-party plugins.
|
| 45 |
-
md = markitdown.MarkItDown(enable_plugins=
|
| 46 |
result = md.convert(url)
|
| 47 |
return result.text_content
|
| 48 |
|
|
@@ -89,22 +90,6 @@ class BasicAgent:
|
|
| 89 |
|
| 90 |
print(f"NEW5: BasicAgent {self.manager_model_id=} {self.code_model_id=}")
|
| 91 |
|
| 92 |
-
file_convert_tool = [
|
| 93 |
-
convert_from_url,
|
| 94 |
-
smolagents.FinalAnswerTool(),
|
| 95 |
-
]
|
| 96 |
-
self.convert_files_to_markdown = smolagents.CodeAgent(
|
| 97 |
-
name="convert_files_to_markdown",
|
| 98 |
-
description=
|
| 99 |
-
"Given a url, it converts the file from various formats (such as pdf, excel, powerpoint, audio, csv, etc) into the markdown format",
|
| 100 |
-
model=self.code_model,
|
| 101 |
-
tools=file_convert_tool,
|
| 102 |
-
max_steps=6,
|
| 103 |
-
verbosity_level=2,
|
| 104 |
-
planning_interval=None,
|
| 105 |
-
additional_authorized_imports=["markitdown"],
|
| 106 |
-
)
|
| 107 |
-
|
| 108 |
math_tools = [
|
| 109 |
smolagents.FinalAnswerTool(),
|
| 110 |
]
|
|
@@ -153,23 +138,32 @@ class BasicAgent:
|
|
| 153 |
additional_authorized_imports=["wikipedia-api"],
|
| 154 |
)
|
| 155 |
|
|
|
|
|
|
|
|
|
|
|
|
|
| 156 |
self.manager_agent = smolagents.CodeAgent(
|
| 157 |
name="manager_agent",
|
| 158 |
description="Manger of other agents",
|
| 159 |
-
tools=
|
| 160 |
model=self.manager_model,
|
| 161 |
add_base_tools=True,
|
| 162 |
max_steps=10,
|
| 163 |
verbosity_level=2,
|
| 164 |
planning_interval=None,
|
| 165 |
additional_authorized_imports=[
|
| 166 |
-
"duckduckgo_search",
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 167 |
],
|
| 168 |
managed_agents=[
|
| 169 |
self.web_search_agent,
|
| 170 |
self.wikipedia_search_agent,
|
| 171 |
self.math_agent,
|
| 172 |
-
self.convert_files_to_markdown,
|
| 173 |
],
|
| 174 |
final_answer_checks=[self.check_results],
|
| 175 |
)
|
|
|
|
| 26 |
|
| 27 |
@smolagents.tool
|
| 28 |
def convert_from_url(url: str) -> str:
|
| 29 |
+
"""Convert data at a url into a markdown string.
|
| 30 |
|
| 31 |
+
This method takes a url as input and converts the contents
|
| 32 |
+
into a markdown. It can convert file formats such as
|
| 33 |
+
pdf, excel, xlsx, csv, powerpoint, ppt, etc.
|
| 34 |
+
|
| 35 |
+
It can also transcribe audio files such as mp3.
|
| 36 |
|
| 37 |
Args:
|
| 38 |
url: the url to access the file. If the file is on the
|
|
|
|
| 43 |
string that is in the markdown format.
|
| 44 |
"""
|
| 45 |
# Don't need third-party plugins.
|
| 46 |
+
md = markitdown.MarkItDown(enable_plugins=False)
|
| 47 |
result = md.convert(url)
|
| 48 |
return result.text_content
|
| 49 |
|
|
|
|
| 90 |
|
| 91 |
print(f"NEW5: BasicAgent {self.manager_model_id=} {self.code_model_id=}")
|
| 92 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 93 |
math_tools = [
|
| 94 |
smolagents.FinalAnswerTool(),
|
| 95 |
]
|
|
|
|
| 138 |
additional_authorized_imports=["wikipedia-api"],
|
| 139 |
)
|
| 140 |
|
| 141 |
+
manager_tools = [
|
| 142 |
+
convert_from_url,
|
| 143 |
+
smolagents.FinalAnswerTool(),
|
| 144 |
+
]
|
| 145 |
self.manager_agent = smolagents.CodeAgent(
|
| 146 |
name="manager_agent",
|
| 147 |
description="Manger of other agents",
|
| 148 |
+
tools=manager_tools,
|
| 149 |
model=self.manager_model,
|
| 150 |
add_base_tools=True,
|
| 151 |
max_steps=10,
|
| 152 |
verbosity_level=2,
|
| 153 |
planning_interval=None,
|
| 154 |
additional_authorized_imports=[
|
| 155 |
+
"duckduckgo_search",
|
| 156 |
+
"wikipedia-api",
|
| 157 |
+
"markitdown",
|
| 158 |
+
"requests",
|
| 159 |
+
"pandas",
|
| 160 |
+
"io",
|
| 161 |
],
|
| 162 |
managed_agents=[
|
| 163 |
self.web_search_agent,
|
| 164 |
self.wikipedia_search_agent,
|
| 165 |
self.math_agent,
|
| 166 |
+
# self.convert_files_to_markdown,
|
| 167 |
],
|
| 168 |
final_answer_checks=[self.check_results],
|
| 169 |
)
|
basic_agent_test.py
CHANGED
|
@@ -9,6 +9,8 @@ LOG = logging.getLogger(__name__)
|
|
| 9 |
|
| 10 |
QUESTIONS_PATH = "../questions.json"
|
| 11 |
|
|
|
|
|
|
|
| 12 |
# Local:
|
| 13 |
# https://huggingface.co/docs/smolagents/tutorials/inspect_runs
|
| 14 |
# https://cobusgreyling.medium.com/introduce-inspectability-to-huggingface-smolagents-571bd3f8da4c
|
|
@@ -53,7 +55,11 @@ print(f"{len(questions)=}")
|
|
| 53 |
# 'task_id': '7bd855d8-463d-4ed5-93ca-5fe35145f733'},
|
| 54 |
#
|
| 55 |
|
| 56 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
| 57 |
question = questions[want]
|
| 58 |
|
| 59 |
q_task_id = question["task_id"]
|
|
@@ -61,12 +67,12 @@ q_text = question["question"]
|
|
| 61 |
q_filename = question.get("file_name", "")
|
| 62 |
|
| 63 |
if q_filename:
|
| 64 |
-
file_url = f"{
|
| 65 |
print("Question {task_id=} has attachment: {file_url=}")
|
| 66 |
quest_with_url = f"{q_text}\nFile url: {file_url}"
|
| 67 |
else:
|
| 68 |
quest_with_url = q_text
|
| 69 |
|
| 70 |
-
answer = ba(
|
| 71 |
|
| 72 |
LOG.warning(f"{answer=}")
|
|
|
|
| 9 |
|
| 10 |
QUESTIONS_PATH = "../questions.json"
|
| 11 |
|
| 12 |
+
DEFAULT_API_URL = "https://agents-course-unit4-scoring.hf.space"
|
| 13 |
+
|
| 14 |
# Local:
|
| 15 |
# https://huggingface.co/docs/smolagents/tutorials/inspect_runs
|
| 16 |
# https://cobusgreyling.medium.com/introduce-inspectability-to-huggingface-smolagents-571bd3f8da4c
|
|
|
|
| 55 |
# 'task_id': '7bd855d8-463d-4ed5-93ca-5fe35145f733'},
|
| 56 |
#
|
| 57 |
|
| 58 |
+
# 18 has an xlsx file
|
| 59 |
+
# 13 has an mp3 file
|
| 60 |
+
# 11 has a py file
|
| 61 |
+
# 9 has an mp3 file
|
| 62 |
+
want = 9
|
| 63 |
question = questions[want]
|
| 64 |
|
| 65 |
q_task_id = question["task_id"]
|
|
|
|
| 67 |
q_filename = question.get("file_name", "")
|
| 68 |
|
| 69 |
if q_filename:
|
| 70 |
+
file_url = f"{DEFAULT_API_URL}/files/{q_task_id}"
|
| 71 |
print("Question {task_id=} has attachment: {file_url=}")
|
| 72 |
quest_with_url = f"{q_text}\nFile url: {file_url}"
|
| 73 |
else:
|
| 74 |
quest_with_url = q_text
|
| 75 |
|
| 76 |
+
answer = ba(quest_with_url)
|
| 77 |
|
| 78 |
LOG.warning(f"{answer=}")
|