Update app.py

app.py

@@ -9,15 +9,20 @@ def image_to_data_url(image_path):
         return None
     with Image.open(image_path) as img:
         buffered = io.BytesIO()
-        img.save(buffered, format=img.format)
+        img_format = img.format if img.format else "JPEG"
+        img.save(buffered, format=img_format)
         img_str = base64.b64encode(buffered.getvalue()).decode()
-        return f"data:image/{img.format.lower()};base64,{img_str}"
+        return f"data:image/{img_format.lower()};base64,{img_str}"
 
 def process_input(image, image_url, prompt, model, hf_token):
     if not hf_token.startswith("hf_"):
         raise gr.Error("Invalid Hugging Face token. It should start with 'hf_'")
     
-    client = InferenceClient(provider="cohere", api_key=hf_token)
+    client = InferenceClient(
+        model=model,
+        token=hf_token,
+        provider="cohere"
+    )
 
     image_data = None
     if image is not None:
@@ -38,7 +43,6 @@ def process_input(image, image_url, prompt, model, hf_token):
     
     try:
         stream = client.chat.completions.create(
-            model=model,
             messages=messages,
             max_tokens=512,
             stream=True,
@@ -46,9 +50,14 @@ def process_input(image, image_url, prompt, model, hf_token):
         
         full_response = ""
         for chunk in stream:
-            content = chunk.choices[0].delta.content or ""
-            full_response += content
-            yield full_response
+            if hasattr(chunk.choices[0], 'delta') and hasattr(chunk.choices[0].delta, 'content'):
+                content = chunk.choices[0].delta.content or ""
+                full_response += content
+                yield full_response
+            elif hasattr(chunk, 'content'):
+                content = chunk.content or ""
+                full_response += content
+                yield full_response
     except Exception as e:
         raise gr.Error(f"API Error: {str(e)}")
 
@@ -63,7 +72,6 @@ with gr.Blocks() as demo:
     
     *Explore state-of-the-art vision-language models by Cohere through this interface.  
     Supports image inputs via upload or URL, with streaming responses.*
-
     Read more about Aya Vision [here](https://cohere.com/research/aya)
     
     **Get your HF token:** [Hugging Face Settings](https://huggingface.co/settings/tokens)
@@ -140,4 +148,4 @@ with gr.Blocks() as demo:
     )
 
 if __name__ == "__main__":
-    demo.queue().launch()
+    demo.launch()