vision-agent 0.2.129__py3-none-any.whl → 0.2.131__py3-none-any.whl

vision_agent/agent/vision_agent.py

@@ -14,7 +14,7 @@ from vision_agent.agent.vision_agent_prompts import (
 )
 from vision_agent.lmm import LMM, Message, OpenAILMM
 from vision_agent.tools import META_TOOL_DOCSTRING
-from vision_agent.tools.meta_tools import Artifacts
+from vision_agent.tools.meta_tools import Artifacts, use_extra_vision_agent_args
 from vision_agent.utils import CodeInterpreterFactory
 from vision_agent.utils.execute import CodeInterpreter, Execution
 
@@ -87,11 +87,18 @@ def run_code_action(
     return result, obs
 
 
-def parse_execution(response: str) -> Optional[str]:
+def parse_execution(
+    response: str,
+    test_multi_plan: bool = True,
+    customed_tool_names: Optional[List[str]] = None,
+) -> Optional[str]:
     code = None
     if "<execute_python>" in response:
         code = response[response.find("<execute_python>") + len("<execute_python>") :]
         code = code[: code.find("</execute_python>")]
+
+    if code is not None:
+        code = use_extra_vision_agent_args(code, test_multi_plan, customed_tool_names)
     return code
 
 
@@ -174,6 +181,8 @@ class VisionAgent(Agent):
         self,
         chat: List[Message],
         artifacts: Optional[Artifacts] = None,
+        test_multi_plan: bool = True,
+        customized_tool_names: Optional[List[str]] = None,
     ) -> Tuple[List[Message], Artifacts]:
         """Chat with VisionAgent, it will use code to execute actions to accomplish
         its tasks.
@@ -184,6 +193,12 @@ class VisionAgent(Agent):
                 or if it contains media files, it should be in the format of:
                 [{"role": "user", "content": "describe your task here...", "media": ["image1.jpg", "image2.jpg"]}]
             artifacts (Optional[Artifacts]): The artifacts to use in the task.
+            test_multi_plan (bool): If True, it will test tools for multiple plans and
+                pick the best one based off of the tool results. If False, it will go
+                with the first plan.
+            customized_tool_names (List[str]): A list of customized tools for agent to
+                pick and use. If not provided, default to full tool set from
+                vision_agent.tools.
 
         Returns:
             List[Message]: The conversation response.
@@ -262,7 +277,9 @@ class VisionAgent(Agent):
                 if response["let_user_respond"]:
                     break
 
-                code_action = parse_execution(response["response"])
+                code_action = parse_execution(
+                    response["response"], test_multi_plan, customized_tool_names
+                )
 
                 if code_action is not None:
                     result, obs = run_code_action(

vision_agent/agent/vision_agent_prompts.py

@@ -1,5 +1,5 @@
 VA_CODE = """
-**Role**: You are a helpful conversational agent that assists users with their requests by writing code to solve it.
+**Role**: You are a helpful agent that assists users with writing code.
 
 **Taks**: As a conversational agent, you are required to understand the user's request and provide a helpful response. Use a Chain-of-Thought approach to break down the problem, create a plan, and then provide a response. Ensure that your response is clear, concise, and helpful. You can use an interactive Python (Jupyter Notebook) environment, executing code with <execution_python>. You are given access to an `artifacts` object which contains files shared between you and the user. `artifacts` will be automatically saved everytime you execute python code.
 
@@ -56,7 +56,9 @@ OBSERVATION:
 
 
 AGENT: {"thoughts": "Two dogs are detected, I will show this to the user and ask them if the result looks good.", "response": "I have written the code to detect dogs and shown the output, do the results look good to you?", "let_user_respond": true}
+"""
 
+EXAMPLES_CODE1_EXTRA = """
 USER: The the image only has one dog, can you fix this?
 
 [Artifacts loaded]
@@ -105,25 +107,24 @@ AGENT: {"thoughts": "I will use the generate_vision_code to count the workers wi
 
 OBSERVATION:
 [Artifact code.py]
-0|from vision_agent.tools import load_image, owl_v2, closest_box_distance, overlay_bounding_boxes, save_image
+0|from vision_agent.tools import load_image, florence2_phrase_grounding, closest_box_distance, overlay_bounding_boxes, save_image
 1|def count_workers_with_helmets(image_path: str, output_path: str):
 2|    image = load_image(image_path)
-3|    detections = owl_v2("worker, helmet", image)
+3|    detections = florence2_phrase_grounding("worker, helmet", image)
 4|    workers = [d for d in detections if d['label'] == 'worker']
 5|    helmets = [d for d in detections if d['label'] == 'helmet']
 6|    count = 0
 6|    for worker in workers:
-7|        person_box = worker['bbox']
-8|        person_has_helmet = False
-9|        for helmet in helmets:
-10|            if closest_box_distance(worker['box'], helmet['box']) < 0.01:
-11|                person_has_helmet = True
-12|                break
-13|        if person_has_helmet:
-14|            count += 1
-15|    overlay_bounding_boxes(image, detections)
-16|    save_image(output_path, image)
-15|    return count
+7|        person_has_helmet = False
+8|        for helmet in helmets:
+9|            if closest_box_distance(worker['box'], helmet['box']) < 0.01:
+10|                person_has_helmet = True
+11|                break
+11|        if person_has_helmet:
+13|            count += 1
+14|    overlay_bounding_boxes(image, detections)
+15|    save_image(output_path, image)
+16|    return count
 [End of artifact]
 
 AGENT: {"thoughts": "I have generated the code to count the workers with helmets in the image, I must now run the code and print the output and write the visualization to the artifacts so I can see the result and the user can see the visaulization.", "response": "<execute_python>from code import count_workers_with_helmets\n    print(count_workers_with_helmets('/path/to/images/workers.png', 'workers_viz.png'))\n    write_media_artifact(artifacts, 'workers_viz.png')</execute_python>", "let_user_respond": false}
@@ -133,4 +134,38 @@ OBSERVATION:
 2
 
 AGENT: {"thoughts": "Two workers with helmets are detected, I will show this to the user and ask them if the result looks good.", "response": "I have written the code to count the workers wearing helmets in code.py and saved the visualization under 'workers_viz.png'.", "let_user_respond": true}
+
+USER: The detections are slightly off. Can you fine tune florence2 using these labels? "[{'image_path': 'image1.png': 'labels': ['worker', 'helmet'], 'bboxes': [[235, 118, 294, 241], [232, 118, 294, 128]]}, {'image_path': 'image2.png': 'labels': ['worker', 'helmet'], 'bboxes': [[235, 118, 294, 241], [232, 118, 294, 128]]}]"
+
+AGENT: {"thoughts": "Because the user has supplied me with labels I can call florence2_fine_tuning on their behalf to fine tune the model", "response": "I will fine tune florence2 with the labels you provided <execute_python>florence2_fine_tuning([{'image_path': 'image1.png': 'labels': ['worker', 'helmet'], 'bboxes': [[235, 118, 294, 241], [232, 118, 294, 128]]}, {'image_path': 'image2.png': 'labels': ['worker', 'helmet'], 'bboxes': [[235, 118, 294, 241], [232, 118, 294, 128]]}], "phrase_grounding")</execute_python>", "let_user_respond": false}
+
+OBSERVATION:
+[Florence2 fine tuning id: 23b3b022-5ebf-4798-9373-20ef36429abf]
+
+
+AGENT: {"thoughts": "The model has finished fine tuning, I will now replace the original florence2_phrase_grounding call with the fine tuning id.", "response": "I will not update the code to use the fine tuned model. <execute_python>use_florence2_fine_tuning(artifacts, "code.py", "phrase_grounding", "23b3b022-5ebf-4798-9373-20ef36429abf")</execute_python>", "let_user_respond": false}
+
+OBSERVATION:
+[Artifact code.py edits]
+---
++++
+@@ -1,7 +1,7 @@
+ from vision_agent.tools import load_image, florence2_phrase_grounding, closest_box_distance, overlay_bounding_boxes, save_image
+ def count_workers_with_helmets(image_path: str, output_path: str):
+     image = load_image(image_path)
+-    detections = florence2_phrase_grounding("worker, helmet", image)
++    detections = florence2_phrase_grounding("worker, helmet", image, "23b3b022-5ebf-4798-9373-20ef36429abf")
+     workers = [d for d in detections if d['label'] == 'worker']
+     helmets = [d for d in detections if d['label'] == 'helmet']
+     count = 0
+[End of edits]
+
+
+AGENT: {"thoughts": "I have updated the code to use the fine tuned model, I must now run the code and print the output and write the visualization to the artifacts so I can see the result and the user can see the visaulization.", "response": "<execute_python>from code import count_workers_with_helmets\n    print(count_workers_with_helmets('/path/to/images/workers.png', 'workers_viz.png'))\n    write_media_artifact(artifacts, 'workers_viz.png')</execute_python>", "let_user_respond": false}
+
+OBSERVATION:
+----- stdout -----
+3
+
+AGENT: {"thoughts": "Three workers with helmets are detected, I will show this to the user and ask them if the result looks good.", "response": "I have updated the code using the fine tuned florence2_phrase_grounding model to count the workers wearing helmets in code.py and saved the visualization under 'workers_viz.png'.", "let_user_respond": true}
 """

vision_agent/tools/meta_tools.py

@@ -297,7 +297,12 @@ def edit_code_artifact(
 
 
 def generate_vision_code(
-    artifacts: Artifacts, name: str, chat: str, media: List[str]
+    artifacts: Artifacts,
+    name: str,
+    chat: str,
+    media: List[str],
+    test_multi_plan: bool = True,
+    customized_tool_names: Optional[List[str]] = None,
 ) -> str:
     """Generates python code to solve vision based tasks.
 
@@ -306,6 +311,8 @@ def generate_vision_code(
         name (str): The name of the artifact to save the code to.
         chat (str): The chat message from the user.
         media (List[str]): The media files to use.
+        test_multi_plan (bool): Do not change this parameter.
+        customized_tool_names (Optional[List[str]]): Do not change this parameter.
 
     Returns:
         str: The generated code.
@@ -330,7 +337,11 @@ def generate_vision_code(
         agent = va.agent.VisionAgentCoder()
 
     fixed_chat: List[Message] = [{"role": "user", "content": chat, "media": media}]
-    response = agent.chat_with_workflow(fixed_chat, test_multi_plan=True)
+    response = agent.chat_with_workflow(
+        fixed_chat,
+        test_multi_plan=test_multi_plan,
+        customized_tool_names=customized_tool_names,
+    )
     redisplay_results(response["test_result"])
     code = response["code"]
     artifacts[name] = code
@@ -342,7 +353,11 @@ def generate_vision_code(
 
 
 def edit_vision_code(
-    artifacts: Artifacts, name: str, chat_history: List[str], media: List[str]
+    artifacts: Artifacts,
+    name: str,
+    chat_history: List[str],
+    media: List[str],
+    customized_tool_names: Optional[List[str]] = None,
 ) -> str:
     """Edits python code to solve a vision based task.
 
@@ -350,6 +365,7 @@ def edit_vision_code(
         artifacts (Artifacts): The artifacts object to save the code to.
         name (str): The file path to the code.
         chat_history (List[str]): The chat history to used to generate the code.
+        customized_tool_names (Optional[List[str]]): Do not change this parameter.
 
     Returns:
         str: The edited code.
@@ -386,7 +402,11 @@ def edit_vision_code(
             fixed_chat_history.append({"role": "assistant", "content": code})
             fixed_chat_history.append({"role": "user", "content": chat})
 
-    response = agent.chat_with_workflow(fixed_chat_history, test_multi_plan=False)
+    response = agent.chat_with_workflow(
+        fixed_chat_history,
+        test_multi_plan=False,
+        customized_tool_names=customized_tool_names,
+    )
     redisplay_results(response["test_result"])
     code = response["code"]
     artifacts[name] = code
@@ -425,18 +445,19 @@ def get_tool_descriptions() -> str:
 
 
 def florence2_fine_tuning(bboxes: List[Dict[str, Any]], task: str) -> str:
-    """'florence2_fine_tuning' is a tool that fine-tune florence2 to be able to detect
+    """DO NOT use this function unless the user has supplied you with bboxes.
+    'florence2_fine_tuning' is a tool that fine-tune florence2 to be able to detect
     objects in an image based on a given dataset. It returns the fine tuning job id.
 
     Parameters:
-        bboxes (List[BboxInput]): A list of BboxInput containing the
-            image path, labels and bounding boxes.
+        bboxes (List[BboxInput]): A list of BboxInput containing the image path, labels
+            and bounding boxes. The coordinates are unnormalized.
         task (str): The florencev2 fine-tuning task. The options are
             'phrase_grounding'.
 
     Returns:
-        UUID: The fine tuning job id, this id will used to retrieve the fine
-            tuned model.
+        str: The fine tuning job id, this id will used to retrieve the fine tuned
+            model.
 
     Example
     -------
@@ -473,6 +494,54 @@ def get_diff(before: str, after: str) -> str:
     )
 
 
+def get_diff_with_prompts(name: str, before: str, after: str) -> str:
+    diff = get_diff(before, after)
+    return f"[Artifact {name} edits]\n{diff}\n[End of edits]"
+
+
+def use_extra_vision_agent_args(
+    code: str,
+    test_multi_plan: bool = True,
+    customized_tool_names: Optional[List[str]] = None,
+) -> str:
+    """This is for forcing arguments passed by the user to VisionAgent into the
+    VisionAgentCoder call.
+
+    Parameters:
+        code (str): The code to edit.
+        test_multi_plan (bool): Do not change this parameter.
+        customized_tool_names (Optional[List[str]]): Do not change this parameter.
+
+    Returns:
+        str: The edited code.
+    """
+    generate_pattern = r"generate_vision_code\(\s*([^\)]+)\)"
+
+    def generate_replacer(match: re.Match) -> str:
+        arg = match.group(1)
+        out_str = f"generate_vision_code({arg}, test_multi_plan={test_multi_plan}"
+        if customized_tool_names is not None:
+            out_str += f", customized_tool_names={customized_tool_names})"
+        else:
+            out_str += ")"
+        return out_str
+
+    edit_pattern = r"edit_vision_code\(\s*([^\)]+)\)"
+
+    def edit_replacer(match: re.Match) -> str:
+        arg = match.group(1)
+        out_str = f"edit_vision_code({arg}"
+        if customized_tool_names is not None:
+            out_str += f", customized_tool_names={customized_tool_names})"
+        else:
+            out_str += ")"
+        return out_str
+
+    new_code = re.sub(generate_pattern, generate_replacer, code)
+    new_code = re.sub(edit_pattern, edit_replacer, new_code)
+    return new_code
+
+
 def use_florence2_fine_tuning(
     artifacts: Artifacts, name: str, task: str, fine_tune_id: str
 ) -> str:
@@ -521,7 +590,7 @@ def use_florence2_fine_tuning(
 
     artifacts[name] = new_code
 
-    diff = get_diff(code, new_code)
+    diff = get_diff_with_prompts(name, code, new_code)
     print(diff)
     return diff
 

vision_agent/tools/tools.py

@@ -1945,15 +1945,4 @@ TOOLS_DF = get_tools_df(TOOLS)  # type: ignore
 TOOL_DESCRIPTIONS = get_tool_descriptions(TOOLS)  # type: ignore
 TOOL_DOCSTRING = get_tool_documentation(TOOLS)  # type: ignore
 TOOLS_INFO = get_tools_info(FUNCTION_TOOLS)  # type: ignore
-UTILITIES_DOCSTRING = get_tool_documentation(
-    [
-        save_json,
-        load_image,
-        save_image,
-        save_video,
-        overlay_bounding_boxes,
-        overlay_segmentation_masks,
-        overlay_heat_map,
-        overlay_counting_results,
-    ]
-)
+UTILITIES_DOCSTRING = get_tool_documentation(UTIL_TOOLS)  # type: ignore

vision_agent/utils/execute.py

@@ -243,6 +243,9 @@ class Logs(BaseModel):
             f"----- stdout -----\n{stdout_str}\n----- stderr -----\n{stderr_str}"
         )
 
+    def to_json(self) -> dict[str, list[str]]:
+        return {"stdout": self.stdout, "stderr": self.stderr}
+
 
 class Error(BaseModel):
     """Represents an error that occurred during the execution of a cell. The error

{vision_agent-0.2.129.dist-info → vision_agent-0.2.131.dist-info}/METADATA

@@ -1,6 +1,6 @@
 Metadata-Version: 2.1
 Name: vision-agent
-Version: 0.2.129
+Version: 0.2.131
 Summary: Toolset for Vision Agent
 Author: Landing AI
 Author-email: dev@landing.ai

{vision_agent-0.2.129.dist-info → vision_agent-0.2.131.dist-info}/RECORD

@@ -2,10 +2,10 @@ vision_agent/__init__.py,sha256=EAb4-f9iyuEYkBrX4ag1syM8Syx8118_t0R6_C34M9w,57
 vision_agent/agent/__init__.py,sha256=FRwiux1FGvGccetyUCtY46KP01fQteqorm-JtFepovI,176
 vision_agent/agent/agent.py,sha256=2cjIOxEuSJrqbfPXYoV0qER5ihXsPFCoEFJa4jpqan0,597
 vision_agent/agent/agent_utils.py,sha256=22LiPhkJlS5mVeo2dIi259pc2NgA7PGHRpcbnrtKo78,1930
-vision_agent/agent/vision_agent.py,sha256=WM1_o0VAQokAKlDr-0lpFxCRwUm_eFfFNWP-wSNjo7s,11180
+vision_agent/agent/vision_agent.py,sha256=7Xa_TjjbUXhlPmKpmXGCAIdT0-PJzRL2rFaACszTXX0,12001
 vision_agent/agent/vision_agent_coder.py,sha256=_2QQd_nTGojkk2ZOiMevVCY6-eUA9q1QdCWH7-Noq4w,34237
 vision_agent/agent/vision_agent_coder_prompts.py,sha256=nj4iRRSAWYHjKqyUSp12aTCV1D5iUVCHeezVXoozS4M,12687
-vision_agent/agent/vision_agent_prompts.py,sha256=K1nLo3XKQ-IqCom1TRwh3cMoGZNxNwEgZqf3uJ6eL18,7221
+vision_agent/agent/vision_agent_prompts.py,sha256=-fXiIIb48duXVljWYcJ0Y4ZzfNnRFi3C5cKdF4SdDo8,10075
 vision_agent/clients/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
 vision_agent/clients/http.py,sha256=k883i6M_4nl7zwwHSI-yP5sAgQZIDPM1nrKD6YFJ3Xs,2009
 vision_agent/clients/landing_public_api.py,sha256=lU2ev6E8NICmR8DMUljuGcVFy5VNJQ4WQkWC8WnnJEc,1503
@@ -15,19 +15,19 @@ vision_agent/lmm/__init__.py,sha256=YuUZRsMHdn8cMOv6iBU8yUqlIOLrbZQqZl9KPnofsHQ,
 vision_agent/lmm/lmm.py,sha256=092oefI65_QSRvQm2znXkjTdzlZTh-Ni_38610kfbJg,16836
 vision_agent/lmm/types.py,sha256=ZEXR_ptBL0ZwDMTDYkgxUCmSZFmBYPQd2jreNzr_8UY,221
 vision_agent/tools/__init__.py,sha256=nx60_hujcnLz3d2wQlCbcerUmT6R2vxRy66IsQjdB3M,2364
-vision_agent/tools/meta_tools.py,sha256=KeGiw2OtY8ARpGbtWjoNAoO1dwevt7LbCupaJX61MkE,18929
+vision_agent/tools/meta_tools.py,sha256=qbf_dzVmhf4zhv-xY1zaqRFshDlvj_7ilFQtSr70hdQ,21213
 vision_agent/tools/prompts.py,sha256=V1z4YJLXZuUl_iZ5rY0M5hHc_2tmMEUKr0WocXKGt4E,1430
 vision_agent/tools/tool_utils.py,sha256=PjdataKjPpiFSq1QBAAWHJUGPPn4p4dr07TPSlhXvFk,7758
-vision_agent/tools/tools.py,sha256=p6QUo7V03UZOKBAGfabVWdPm9vUT9tyP_utCv0yKfcY,68659
+vision_agent/tools/tools.py,sha256=ywfolLhf8OWnavXTTEYscUvOUM0ECNy-ff3WLMdUhhQ,68465
 vision_agent/tools/tools_types.py,sha256=rLpCUODPY0yI65SLOTJOxfHFfqWM3WjOq-AYX25Chjk,2356
 vision_agent/utils/__init__.py,sha256=7fMgbZiEwbNS0fBOS_hJI5PuEYBblw36zLi_UjUzvj4,244
 vision_agent/utils/exceptions.py,sha256=booSPSuoULF7OXRr_YbC4dtKt6gM_HyiFQHBuaW86C4,2052
-vision_agent/utils/execute.py,sha256=gc4R_0BKUrZyhiKvIxOpYuzQPYVWQEqxr3ANy1lJAw4,27037
+vision_agent/utils/execute.py,sha256=uIAupmUtesUvhCQJ8oeGxDeX7SYDz0TLxFhPB4HD2NA,27147
 vision_agent/utils/image_utils.py,sha256=zTTOJFOieMzwIquTFnW7T6ssx9o6XfoZ0Unqyk7GJrg,10746
 vision_agent/utils/sim.py,sha256=ebE9Cs00pVEDI1HMjAzUBk88tQQmc2U-yAzIDinnekU,5572
 vision_agent/utils/type_defs.py,sha256=BE12s3JNQy36QvauXHjwyeffVh5enfcvd4vTzSwvEZI,1384
 vision_agent/utils/video.py,sha256=GmJqu_3WhBMEwP4HToMMp8EwgftliHSpv5nd-QEDOcs,4528
-vision_agent-0.2.129.dist-info/LICENSE,sha256=xx0jnfkXJvxRnG63LTGOxlggYnIysveWIZ6H3PNdCrQ,11357
-vision_agent-0.2.129.dist-info/METADATA,sha256=uMW3dpm48GLsgwAA_grlCghpKPA32caHy2SrCWhYtdI,12295
-vision_agent-0.2.129.dist-info/WHEEL,sha256=7Z8_27uaHI_UZAc4Uox4PpBhQ9Y5_modZXWMxtUi4NU,88
-vision_agent-0.2.129.dist-info/RECORD,,
+vision_agent-0.2.131.dist-info/LICENSE,sha256=xx0jnfkXJvxRnG63LTGOxlggYnIysveWIZ6H3PNdCrQ,11357
+vision_agent-0.2.131.dist-info/METADATA,sha256=LCIVXm-Le9Uw6Vp-XMvmmkhMRPRJJlBZmJPF28Bn6Hs,12295
+vision_agent-0.2.131.dist-info/WHEEL,sha256=7Z8_27uaHI_UZAc4Uox4PpBhQ9Y5_modZXWMxtUi4NU,88
+vision_agent-0.2.131.dist-info/RECORD,,