Update minigpt-4 readme

open-mmlab · Oct 12, 2023 · 11abc5b · 11abc5b
1 parent 59b105f
commit 11abc5b
Show file tree

Hide file tree

Showing 4 changed files with 19 additions and 7 deletions.
diff --git a/configs/minigpt4/README.md b/configs/minigpt4/README.md
@@ -34,9 +34,10 @@ For Vicuna model, please refer to [MiniGPT-4 page](https://github.com/Vision-CAI
 
 ### Pretrained models
 
-| Model                           | Params (M) | Flops (G) |                  Config                  |                                                    Download                                                    |
-| :------------------------------ | :--------: | :-------: | :--------------------------------------: | :------------------------------------------------------------------------------------------------------------: |
-| `minigpt-4_vicuna-7b_caption`\* |  8121.32   |    N/A    | [config](minigpt-4_vicuna-7b_caption.py) | [model](https://download.openmmlab.com/mmpretrain/v1.0/minigpt4/minigpt-4_linear-projection_20230615-714b5f52.pth) |
+| Model                           | Params (M) | Flops (G) |                   Config                   |                                                   Download                                                   |
+| :------------------------------ | :--------: | :-------: | :----------------------------------------: | :----------------------------------------------------------------------------------------------------------: |
+| `minigpt-4_baichuan-7b_caption` |  8094.77   |    N/A    | [config](minigpt-4_baichuan-7b_caption.py) | [model](https://download.openmmlab.com/mmclassification/v1/minigpt4/minigpt-4_linear_baichuan7b_20231011-5dca7ed6.pth) |
+| `minigpt-4_vicuna-7b_caption`\* |  8121.32   |    N/A    |  [config](minigpt-4_vicuna-7b_caption.py)  | [model](https://download.openmmlab.com/mmclassification/v1/minigpt4/minigpt-4_linear_vicuna7b_20230615-714b5f52.pth) |
 
 *Models with * are converted from the [official repo](https://github.com/Vision-CAIR/MiniGPT-4/tree/main). The config files of these models are only for inference. We haven't reproduce the training results.*
 

diff --git a/configs/minigpt4/metafile.yml b/configs/minigpt4/metafile.yml
@@ -19,8 +19,19 @@ Models:
       - Task: Image Caption
         Dataset: COCO
         Metrics: null
-    Weights: https://download.openmmlab.com/mmpretrain/v1.0/minigpt4/minigpt-4_linear-projection_20230615-714b5f52.pth
+    Weights: https://download.openmmlab.com/mmclassification/v1/minigpt4/minigpt-4_linear_vicuna7b_20230615-714b5f52.pth
     Config: configs/minigpt4/minigpt-4_vicuna-7b_caption.py
     Converted From:
       Weights: https://github.com/Vision-CAIR/MiniGPT-4/tree/main
       Code: https://github.com/Vision-CAIR/MiniGPT-4/tree/main
+  - Name: minigpt-4_baichuan-7b_caption
+    Metadata:
+      FLOPs: null
+      Parameters: 8094769024
+    In Collection: MiniGPT4
+    Results:
+      - Task: Image Caption
+        Dataset: COCO
+        Metrics: null
+    Weights: https://download.openmmlab.com/mmclassification/v1/minigpt4/minigpt-4_linear_baichuan7b_20231011-5dca7ed6.pth
+    Config: configs/minigpt4/minigpt-4_baichuan-7b_caption.py
diff --git a/configs/minigpt4/minigpt-4_baichuan-7b_caption.py b/configs/minigpt4/minigpt-4_baichuan-7b_caption.py
@@ -101,11 +101,11 @@
     ),
     lang_encoder=dict(
         type='AutoModelForCausalLM',
-        name_or_path='YOUR_PATH_TO_BAICHUAN',
+        name_or_path='baichuan-inc/baichuan-7B',
         trust_remote_code=True),
     tokenizer=dict(
         type='AutoTokenizer',
-        name_or_path='YOUR_PATH_TO_BAICHUAN',
+        name_or_path='baichuan-inc/baichuan-7B',
         trust_remote_code=True),
     task='caption',
     prompt_template=dict([('en', '###Ask: {} ###Answer: '),

diff --git a/projects/gradio_demo/minigpt4_demo.py b/projects/gradio_demo/minigpt4_demo.py
@@ -37,7 +37,7 @@ def get_free_device():
 device = get_free_device()
 inferencer = ImageCaptionInferencer(model=args.cfg, pretrained=args.ckpt)
 model = inferencer.model
-chat = Chat(inferencer, device=device, is_half=True)
+chat = Chat(inferencer, device=device, is_half=(device.type != 'cpu'))
 
 
 def reset(chat_state, img_list):