fix for llama

.gitignore

@@ -0,0 +1 @@
+heron

app.py

@@ -15,16 +15,18 @@ from transformers import (
 )
 
 
-os.system(
-    "git clone https://github.com/turingmotors/heron.git"
-    "&& export CUDA_HOME=/usr/local/cuda; pip install -e heron"
-)
+if os.path.exists("heron") == False:
+    os.system(
+        "git clone https://github.com/turingmotors/heron.git"
+        "&& export CUDA_HOME=/usr/local/cuda; pip install -e heron"
+    )
 
 sys.path.insert(0, "./heron")
 from heron.models.git_llm.git_japanese_stablelm_alpha import (
     GitJapaneseStableLMAlphaConfig,
     GitJapaneseStableLMAlphaForCausalLM,
 )
+from heron.models.git_llm.git_llama import GitLlamaConfig, GitLlamaForCausalLM
 
 logger = logging.getLogger(__name__)
 
@@ -59,7 +61,7 @@ class KeywordsStoppingCriteria(StoppingCriteria):
 def preprocess(history, image):
     text = ""
     for one_history in history:
-        text += f"###human: {one_history[0]}\n###gpt: "
+        text += f"##human: {one_history[0]}\n##gpt: "
     # do preprocessing
     inputs = processor(
         text,
@@ -82,7 +84,7 @@ def add_text(textbox, history):
 title_markdown = """
 # Heronチャットデモ
 
-- モデル: [turing-motors/heron-chat-git-ja-stablelm-base-7b-v0](https://huggingface.co/turing-motors/heron-chat-git-ja-stablelm-base-7b-v0)
+- モデル: [TBD](TBD)
 - 学習コード: [Heron](https://github.com/turingmotors/heron)
 """
 
@@ -117,7 +119,8 @@ def stream_bot(imagebox, history):
     history[-1][1] = ""
     for new_text in streamer:
         history[-1][1] += new_text
-        history[-1][1] = history[-1][1].replace(EOS_WORDS, "")
+        while history[-1][1].endswith("#"):
+            history[-1][1] = history[-1][1][:-1]
         time.sleep(0.05)
         yield history
 
@@ -143,25 +146,45 @@ def build_demo():
 
                 gr.Examples(
                     examples=[
+                        # [
+                        #     "./images/bus_kyoto.png",
+                        #     "この道路を運転する時には何に気をつけるべきですか？",
+                        # ],
+                        # [
+                        #     "./images/bear.png",
+                        #     "この画像には何が写っていますか？",
+                        # ],
+                        # [
+                        #     "./images/water_bus.png",
+                        #     "画像には何が写っていますか？",
+                        # ],
+                        # [
+                        #     "./images/extreme_ironing.jpg",
+                        #     "この画像の面白い点は何ですか？",
+                        # ],
+                        # [
+                        #     "./images/heron.png",
+                        #     "この画像はどういう点が面白いですか？",
+                        # ],
                         [
                             "./images/bus_kyoto.png",
-                            "この道路を運転する時には何に気をつけるべきですか？",
+                            "What should you be careful of when driving on this road?",
                         ],
                         [
                             "./images/bear.png",
-                            "この画像には何が写っていますか？",
+                            "What is shown in this image?",
                         ],
                         [
                             "./images/water_bus.png",
-                            "画像には何が写っていますか？",
+                            "What is depicted in the picture?",
                         ],
                         [
                             "./images/extreme_ironing.jpg",
-                            "この画像の面白い点は何ですか？",
+                            "What is the unusual aspect of this image?",
                         ],
                         [
                             "./images/heron.png",
-                            "この画像はどういう点が面白いですか？",
+                            "What is intriguing about this picture?",
                         ],
                     ],
                     inputs=[imagebox, textbox],
@@ -206,34 +229,29 @@ def build_demo():
 
 
 if __name__ == "__main__":
-    EOS_WORDS = "###"
+    EOS_WORDS = "##"
 
     device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
     max_length = 512
 
     vision_model_name = "openai/clip-vit-large-patch14-336"
-    MODEL_NAME = "turing-motors/heron-chat-git-ja-stablelm-base-7b-v0"
-    PROCESSOR_PATH = "turing-motors/heron-chat-git-ja-stablelm-base-7b-v0"
+    MODEL_NAME = "turing-motors/inoichi-exp176-llama2"
+    PROCESSOR_PATH = "turing-motors/inoichi-exp175-llama2"
 
     # prepare a pretrained model
-    git_config = GitJapaneseStableLMAlphaConfig.from_pretrained(MODEL_NAME)
+    git_config = GitLlamaConfig.from_pretrained(MODEL_NAME)
     git_config.set_vision_configs(
-        num_image_with_embedding=1,
-        vision_model_name=vision_model_name,
+        num_image_with_embedding=1, vision_model_name=vision_model_name
     )
-    model = GitJapaneseStableLMAlphaForCausalLM.from_pretrained(
+    model = GitLlamaForCausalLM.from_pretrained(
         MODEL_NAME, config=git_config, torch_dtype=torch.float16
     )
+
     model.eval()
     model.to(device)
 
     # prepare a processor
-    processor = AutoProcessor.from_pretrained("microsoft/git-base")
-    processor.image_processor = CLIPImageProcessor.from_pretrained(vision_model_name)
-    processor.tokenizer = LlamaTokenizer.from_pretrained(
-        "novelai/nerdstash-tokenizer-v1",
-        additional_special_tokens=["▁▁"],
-    )
+    processor = AutoProcessor.from_pretrained(PROCESSOR_PATH)
 
     demo = build_demo()
-    demo.queue(concurrency_count=1, max_size=5, api_open=False).launch()
+    demo.queue(concurrency_count=1, max_size=5, api_open=False).launch(share=True)