Upload 2 files (#1)

- Upload 2 files (dc3177f056c92ec22fa26a0fc46247ee7fb4b443)

tools.py

@@ -0,0 +1,126 @@
+import os
+from smolagents import Tool, tool
+from huggingface_hub import HfApi
+from dotenv import load_dotenv
+from huggingface_hub import InferenceClient
+from utils import upload_file
+load_dotenv()
+HF_TOKEN = os.environ.get("HF_TOKEN")
+
+api = HfApi()
+client = InferenceClient(
+    provider="hf-inference",
+    api_key=HF_TOKEN,
+)
+
+# --- Constants ---
+local_data_path = "../data"
+
+if not os.path.exists(local_data_path):
+    os.makedirs(local_data_path)
+
+
+
+@tool
+def image_question_answering(image_path: str, prompt: str) -> str:
+    """
+    This function takes a image path and a prompt, and returns the answer to the question.
+    Args:
+        image_path: The path to the image file
+        prompt: The prompt to the question
+    Returns:
+        The answer to the question
+    """
+    file_extension = image_path.split(".")[-1]
+    if file_extension in [".mp4", ".avi", ".mov", ".wmv", ".mkv", ".webm"]:
+        return "Media type not supported. Please upload an image."
+   
+    if image_path.startswith("http"):
+        media_url = image_path
+    else:
+        media_url = upload_file(image_path)
+
+    
+    messages = [
+                {
+                    "role": "user",
+                    "content": [
+                        {
+                            "type": "text",
+                            "text": prompt,
+                        },
+                        {
+                            "type": "image_url",
+                            "image_url": {"url": media_url},
+                        }
+                    ],
+                }
+            ]
+
+    completion = client.chat.completions.create(
+            model="meta-llama/Llama-3.2-11B-Vision-Instruct",
+            messages=messages,
+    )
+    return completion.choices[0].message
+
+
+@tool
+def transcribe_audio(file_local_path: str) -> str:
+    """
+    Transcribe the audio file and return the transcript
+    Args:
+        file_local_path: The local path to the audio file
+    Returns:
+        The transcript of the audio file
+    """
+    file_url = upload_file(file_local_path)
+
+    asr_tool = Tool.from_space(
+        "hf-audio/whisper-large-v3",
+        api_name="/predict_1", # from file
+        name="transcribe_audio",
+        description="Use this tool to transcribe the audio"
+    )
+
+    transcript = asr_tool(file_url)
+    return transcript
+
+
+class GetFileTool(Tool):
+    name = "get_file"
+    description = "Download a file from the given file name"
+    inputs = {
+        "file_name": {
+            "type": "string",
+            "description": "Download the file from the given file name and outputs the local path"
+        }
+
+    }
+    output_type = "string"
+
+    def forward(self, file_name: str) -> str:
+        import requests
+
+        if file_name == "":
+            return "No file name provided"
+
+        task_id = file_name.split(".")[0]
+        url = f"https://agents-course-unit4-scoring.hf.space/files/{task_id}"
+        headers = {
+            "accept": "application/json"
+        }
+        req = requests.get(url, headers=headers)
+
+        if req.status_code != 200:
+            return "File not found, please check the file name and try again."
+        
+        local_file_path = local_data_path + "/" + file_name
+
+        with open(local_file_path, "wb") as f:
+            f.write(req.content)
+        
+        print(f"File saved to {local_file_path}. You can read this file to process its contents.")
+        return local_file_path
+
+
+

utils.py

@@ -0,0 +1,55 @@
+from smolagents import OpenAIServerModel
+from dotenv import load_dotenv
+from huggingface_hub import HfApi
+import os
+
+load_dotenv()
+OPENAI_API_KEY = os.environ.get("OPENAI_API_KEY")
+repo_id = "zaradana/temp_files"
+
+api = HfApi()
+
+def check_asnwer_format(final_answer, agent_memory):
+    multimodal_model = OpenAIServerModel("gpt-4o", max_tokens=8096, api_key=OPENAI_API_KEY)
+    prompt = (
+        f"Here is a user-given task and the agent steps: {agent_memory.get_succinct_steps()}. "
+        f"Here is the final answer: {final_answer}. "
+        "Please check that the answer is in the requested format. "
+        "First list reasons why yes/no, then write your final decision: PASS in caps lock if it is satisfactory, FAIL if it is not."
+    )
+    messages = [
+        {
+            "role": "user",
+            "content": [
+                {
+                    "type": "text",
+                    "text": prompt,
+                }
+            ],
+        }
+    ]
+    output = multimodal_model(messages).content
+    print("Feedback: ", output)
+    if "FAIL" in output:
+        raise Exception(output)
+    return True
+
+
+def upload_file(file_local_path: str) -> str:
+    """
+    Upload a file to the Hugging Face Hub and return the URL
+    Args:
+        file_local_path: The local path to the file
+    Returns:
+        The URL of the uploaded file
+    """
+    file_name = file_local_path.split("/")[-1]
+
+    api.upload_file(
+        path_or_fileobj=file_local_path,
+        path_in_repo=file_name,
+        repo_id=repo_id,
+        repo_type="dataset"
+    )
+    file_url = f"https://huggingface.co/datasets/{repo_id}/resolve/main/{file_name}"
+    return file_url