Break: Argument name and versioning

.github/workflows/package.json

@@ -22,6 +22,10 @@
               "tag": "Add",
               "release": "minor"
             },
+            {
+              "tag": "Break",
+              "release": "major"
+            },
             {
               "tag": "Improve",
               "release": "patch"
@@ -46,6 +50,10 @@
               "tag": "Add",
               "release": "minor"
             },
+            {
+              "tag": "Break",
+              "release": "major"
+            },
             {
               "tag": "Improve",
               "release": "patch"

README.md

@@ -141,8 +141,8 @@ decoded_text = processor.batch_decode(output[:, prompt_len:])[0]
 
 The generative models can be used for chat-like experiences, where the user can provide both text and images as input.
 To use that feature, you can start with the following CLI command:
-
-```bash
+image
+```bashimage
 uform-chat --model unum-cloud/uform-gen-chat --image_path=zebra.jpg
 uform-chat --model unum-cloud/uform-gen-chat --image_path=zebra.jpg --device="cuda:0" --fp16
 ```

src/uform/chat.py

@@ -14,9 +14,9 @@ def parse_args():
     parser = ArgumentParser(description="Chat with UForm generative model")
 
     parser.add_argument("--model", type=str, default="unum-cloud/uform-gen-chat")
-    parser.add_argument("--image_path", type=str, help="", required=True)
+    parser.add_argument("--image", type=str, help="", required=True)
     parser.add_argument("--device", type=str, required=True)
-    parser.add_argument("--fp16", action="store_true")  
+    parser.add_argument("--fp16", action="store_true")
 
     return parser.parse_args()
 
@@ -30,18 +30,18 @@ def run_chat(opts, model, processor):
 
     messages = [{"role": "system", "content": "You are a helpful assistant."}]
     is_first_message = True
-    if opts.image_path.startswith("http"):
+    if opts.image.startswith("http"):
         image = (
             processor.image_processor(
-                Image.open(requests.get(opts.image_path, stream=True).raw)
+                Image.open(requests.get(opts.image, stream=True).raw)
             )
             .unsqueeze(0)
             .to(torch.bfloat16 if opts.fp16 else torch.float32)
             .to(opts.device)
         )
     else:
         image = (
-            processor.image_processor(Image.open(opts.image_path))
+            processor.image_processor(Image.open(opts.image))
             .unsqueeze(0)
             .to(torch.bfloat16 if opts.fp16 else torch.float32)
             .to(opts.device)