Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

V3 improvments set2 #99

Merged
merged 3 commits into from
May 29, 2024
Merged
Show file tree
Hide file tree
Changes from all commits
Commits
File filter

Filter by extension

Filter by extension

Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
12 changes: 9 additions & 3 deletions vision_agent/agent/agent_coder.py
Original file line number Diff line number Diff line change
Expand Up @@ -67,11 +67,17 @@ def parse_file_name(s: str) -> str:
return "".join([p for p in s.split(" ") if p.endswith(".png")])


def write_program(question: str, feedback: str, model: LLM) -> str:
def write_program(
question: str, feedback: str, model: LLM, media: Optional[Union[str, Path]] = None
) -> str:
prompt = PROGRAM.format(
docstring=TOOL_DOCSTRING, question=question, feedback=feedback
)
completion = model(prompt)
if isinstance(model, OpenAILMM):
completion = model(prompt, images=[media] if media else None)
else:
completion = model(prompt)

return preprocess_data(completion)


Expand Down Expand Up @@ -168,7 +174,7 @@ def chat(
code = ""
feedback = ""
for _ in range(self.max_turns):
code = write_program(question, feedback, self.coder_agent)
code = write_program(question, feedback, self.coder_agent, media=media)
if self.verbose:
_CONSOLE.print(
Syntax(code, "python", theme="gruvbox-dark", line_numbers=True)
Expand Down
17 changes: 13 additions & 4 deletions vision_agent/agent/vision_agent.py
Original file line number Diff line number Diff line change
Expand Up @@ -21,6 +21,7 @@
USER_REQ,
)
from vision_agent.llm import LLM, OpenAILLM
from vision_agent.lmm import LMM, OpenAILMM
from vision_agent.tools import TOOL_DESCRIPTIONS, TOOLS_DF, UTILITIES_DOCSTRING
from vision_agent.utils import Execute
from vision_agent.utils.sim import Sim
Expand Down Expand Up @@ -77,7 +78,8 @@ def write_plan(
chat: List[Dict[str, str]],
tool_desc: str,
working_memory: str,
model: LLM,
model: Union[LLM, LMM],
media: Optional[List[Union[str, Path]]] = None,
) -> List[Dict[str, str]]:
chat = copy.deepcopy(chat)
if chat[-1]["role"] != "user":
Expand All @@ -87,7 +89,10 @@ def write_plan(
context = USER_REQ.format(user_request=user_request)
prompt = PLAN.format(context=context, tool_desc=tool_desc, feedback=working_memory)
chat[-1]["content"] = prompt
return extract_json(model.chat(chat))["plan"] # type: ignore
if isinstance(model, OpenAILMM):
return extract_json(model.chat(chat, images=media))["plan"] # type: ignore
else:
return extract_json(model.chat(chat))["plan"] # type: ignore


def reflect(
Expand Down Expand Up @@ -324,7 +329,7 @@ def __call__(
input = [{"role": "user", "content": input}]
results = self.chat_with_workflow(input, media)
results.pop("working_memory")
return results["code"] # type: ignore
return results # type: ignore

def chat_with_workflow(
self,
Expand Down Expand Up @@ -363,7 +368,11 @@ def chat_with_workflow(

while not success and retries < self.max_retries:
plan_i = write_plan(
chat, TOOL_DESCRIPTIONS, format_memory(working_memory), self.planner
chat,
TOOL_DESCRIPTIONS,
format_memory(working_memory),
self.planner,
media=[media] if media else None,
)
plan_i_str = "\n-".join([e["instructions"] for e in plan_i])
if self.verbosity >= 1:
Expand Down
12 changes: 7 additions & 5 deletions vision_agent/agent/vision_agent_prompts.py
Original file line number Diff line number Diff line change
Expand Up @@ -169,11 +169,13 @@ def find_text(image_path: str, text: str) -> str:
1. Verify the fundamental functionality under normal conditions.
2. Ensure each test case is well-documented with comments explaining the scenario it covers.
3. Your test case MUST run only on the given image which is {media}
4. DO NOT use any non-existent or dummy image or video files that are not provided by the user's instructions.
5. DO NOT mock any functions, you must test their functionality as is.
6. DO NOT assert the output value, run the code and verify it runs without any errors and assert only the output format or data structure.
7. DO NOT import the testing function as it will available in the testing environment.
8. Print the output of the function that is being tested.
4. Your test case MUST run only with the given values which is available in the question - {question}
5. DO NOT use any non-existent or dummy image or video files that are not provided by the user's instructions.
6. DO NOT mock any functions, you must test their functionality as is.
7. DO NOT assert the output value, run the code and assert only the output format or data structure.
8. DO NOT use try except block to handle the error, let the error be raised if the code is incorrect.
9. DO NOT import the testing function as it will available in the testing environment.
10. Print the output of the function that is being tested.
"""


Expand Down
Loading