update

.gitignore

@@ -0,0 +1,159 @@
+# Byte-compiled / optimized / DLL files
+__pycache__/
+*.py[cod]
+*$py.class
+
+# C extensions
+*.so
+
+# Distribution / packaging
+.Python
+build/
+develop-eggs/
+dist/
+downloads/
+eggs/
+.eggs/
+lib/
+lib64/
+parts/
+sdist/
+var/
+wheels/
+share/python-wheels/
+*.egg-info/
+.installed.cfg
+*.egg
+MANIFEST
+
+# PyInstaller
+#  Usually these files are written by a python script from a template
+#  before PyInstaller builds the exe, so as to inject date/other infos into it.
+*.manifest
+*.spec
+
+# Installer logs
+pip-log.txt
+pip-delete-this-directory.txt
+
+# Unit test / coverage reports
+htmlcov/
+.tox/
+.nox/
+.coverage
+.coverage.*
+.cache
+nosetests.xml
+coverage.xml
+*.cover
+*.py,cover
+.hypothesis/
+.pytest_cache/
+cover/
+
+# Translations
+*.mo
+*.pot
+
+# Django stuff:
+*.log
+local_settings.py
+db.sqlite3
+db.sqlite3-journal
+
+# Flask stuff:
+instance/
+.webassets-cache
+
+# Scrapy stuff:
+.scrapy
+
+# Sphinx documentation
+docs/_build/
+
+# PyBuilder
+.pybuilder/
+target/
+
+# Jupyter Notebook
+.ipynb_checkpoints
+
+# IPython
+profile_default/
+ipython_config.py
+
+# pyenv
+#   For a library or package, you might want to ignore these files since the code is
+#   intended to run in multiple environments; otherwise, check them in:
+# .python-version
+
+# pipenv
+#   According to pypa/pipenv#598, it is recommended to include Pipfile.lock in version control.
+#   However, in case of collaboration, if having platform-specific dependencies or dependencies
+#   having no cross-platform support, pipenv may install dependencies that don't work, or not
+#   install all needed dependencies.
+#Pipfile.lock
+
+# poetry
+#   Similar to Pipfile.lock, it is generally recommended to include poetry.lock in version control.
+#   This is especially recommended for binary packages to ensure reproducibility, and is more
+#   commonly ignored for libraries.
+#   https://python-poetry.org/docs/basic-usage/#commit-your-poetrylock-file-to-version-control
+#poetry.lock
+
+# pdm
+#   Similar to Pipfile.lock, it is generally recommended to include pdm.lock in version control.
+#pdm.lock
+#   pdm stores project-wide configurations in .pdm.toml, but it is recommended to not include it
+#   in version control.
+#   https://pdm.fming.dev/#use-with-ide
+.pdm.toml
+
+# PEP 582; used by e.g. github.com/David-OConnor/pyflow and github.com/pdm-project/pdm
+__pypackages__/
+
+# Celery stuff
+celerybeat-schedule
+celerybeat.pid
+
+# SageMath parsed files
+*.sage.py
+
+# Environments
+.env
+.venv
+env/
+venv/
+ENV/
+env.bak/
+venv.bak/
+
+# Spyder project settings
+.spyderproject
+.spyproject
+
+# Rope project settings
+.ropeproject
+
+# mkdocs documentation
+/site
+
+# mypy
+.mypy_cache/
+.dmypy.json
+dmypy.json
+
+# Pyre type checker
+.pyre/
+
+# pytype static type analyzer
+.pytype/
+
+# Cython debug symbols
+cython_debug/
+
+# PyCharm
+#  JetBrains specific template is maintained in a separate JetBrains.gitignore that can
+#  be found at https://github.com/github/gitignore/blob/main/Global/JetBrains.gitignore
+#  and can be added to the global gitignore or merged into this file.  For a more nuclear
+#  option (not recommended) you can uncomment the following to ignore the entire idea folder.

app.py

@@ -1,24 +1,120 @@
 import gradio as gr
 import spaces
+import time
 from PIL import Image
 from models.mllava import MLlavaProcessor, LlavaForConditionalGeneration, chat_mllava, MLlavaForConditionalGeneration
 from typing import List
-processor = MLlavaProcessor()
-model = LlavaForConditionalGeneration.from_pretrained("MFuyu/mllava_v2_4096")
+processor = MLlavaProcessor.from_pretrained("MFuyu/mllava_llava_debug_nlvr2_v5_4096")
+model = LlavaForConditionalGeneration.from_pretrained("MFuyu/mllava_llava_debug_nlvr2_v5_4096")
 
 @spaces.GPU
-def generate(text:str, images:List[Image.Image], history: List[dict]):
+def generate(text:str, images:List[Image.Image], history: List[dict], **kwargs):
+    global processor, model
     model = model.to("cuda")
-    
-    for text, history in chat_mllava(text, images, model, processor, history=history, stream=True):
-        yield text, history
+    if not images:
+        images = None
+    for text, history in chat_mllava(text, images, model, processor, history=history, stream=True, **kwargs):
+        yield text
 
-def build_demo():
+    return text
+
+def enable_next_image(uploaded_images, image):
+    uploaded_images.append(image)
+    return uploaded_images, gr.MultimodalTextbox(value=None, interactive=False)
+
+def add_message(history, message):
+    if message["files"]:
+        for file in message["files"]:
+            history.append([(file,), None])
+    if message["text"]:
+        history.append([message["text"], None])
+    return history, gr.MultimodalTextbox(value=None)
+
+def print_like_dislike(x: gr.LikeData):
+    print(x.index, x.value, x.liked)
+
+
+def get_chat_history(history):
+    chat_history = []
+    for i, message in enumerate(history):
+        if isinstance(message[0], str):
+            chat_history.append({"role": "user", "text": message[0]})
+            if i != len(history) - 1:
+                assert message[1], "The bot message is not provided, internal error"
+                chat_history.append({"role": "assistant", "text": message[1]})
+            else:
+                assert not message[1], "the bot message internal error, get: {}".format(message[1])
+                chat_history.append({"role": "assistant", "text": ""})
+    return chat_history
+
+def get_chat_images(history):
+    images = []
+    for message in history:
+        if isinstance(message[0], tuple):
+            images.extend(message[0])
+    return images
+            
+def bot(history):
+    print(history)
+    cur_messages = {"text": "", "images": []}
+    for message in history[::-1]:
+        if message[1]:
+            break
+        if isinstance(message[0], str):
+            cur_messages["text"] = message[0] + " " + cur_messages["text"]
+        elif isinstance(message[0], tuple):
+            cur_messages["images"].extend(message[0])
+    cur_messages["text"] = cur_messages["text"].strip()
+    cur_messages["images"] = cur_messages["images"][::-1]
+    if not cur_messages["text"]:
+        raise gr.Error("Please enter a message")
+    if cur_messages['text'].count("<image>") < len(cur_messages['images']):
+        gr.Warning("The number of images uploaded is more than the number of <image> placeholders in the text. Will automatically prepend <image> to the text.")
+        cur_messages['text'] = "<image> "* (len(cur_messages['images']) - cur_messages['text'].count("<image>")) + cur_messages['text']
+        history[-1][0] = cur_messages["text"]
+    if cur_messages['text'].count("<image>") > len(cur_messages['images']):
+        gr.Warning("The number of images uploaded is less than the number of <image> placeholders in the text. Will automatically remove extra <image> placeholders from the text.")
+        cur_messages['text'] = cur_messages['text'][::-1].replace("<image>"[::-1], "", cur_messages['text'].count("<image>") - len(cur_messages['images']))[::-1]
+        history[-1][0] = cur_messages["text"]
     
+    chat_history = get_chat_history(history)
+    chat_images = get_chat_images(history)
+    generation_kwargs = {
+        "max_new_tokens": 4096,
+        "temperature": 0.7,
+        "top_p": 1.0,
+        "do_sample": True,
+    }
+    print(None, chat_images, chat_history, generation_kwargs)
+    response = generate(None, chat_images, chat_history, **generation_kwargs)
+            
+    for _output in response:
+        history[-1][1] = _output
+        time.sleep(0.05)
+        yield history
+        
+def build_demo():
+    with gr.Blocks() as demo:
+        chatbot = gr.Chatbot(line_breaks=True)
+        chat_input = gr.MultimodalTextbox(interactive=True, file_types=["image"], placeholder="Enter message or upload images. Please use <image> to indicate the position of uploaded images", show_label=True)
+        
+        chat_msg = chat_input.submit(add_message, [chatbot, chat_input], [chatbot, chat_input])
+        bot_msg = chat_msg.success(bot, chatbot, chatbot, api_name="bot_response")
+        
+        chatbot.like(print_like_dislike, None, None)
+        
+        with gr.Row():
+            send_button = gr.Button("Send")
+            clear_button = gr.ClearButton([chatbot, chat_input])
+        
+        send_button.click(
+            add_message, [chatbot, chat_input], [chatbot, chat_input]
+        ).then(
+            bot, chatbot, chatbot, api_name="bot_response"
+        )
+    return demo    
     
 
 if __name__ == "__main__":
-    processor = MLlavaProcessor()
-    model = LlavaForConditionalGeneration.from_pretrained("MFuyu/mllava_v2_4096")
     demo = build_demo()
     demo.launch()

models/mllava/utils.py

@@ -3,11 +3,11 @@ import torch
 from .modeling_llava import LlavaForConditionalGeneration
 from .processing_llava import MLlavaProcessor
 from ..conversation import conv_mllava_v1_mmtag as default_conv
-from typing import List, Tuple
+from typing import List, Tuple, Union, Tuple
 
 def chat_mllava(
     text:str, 
-    images: List[PIL.Image.Image], 
+    images: List[Union[PIL.Image.Image, str]], 
     model:LlavaForConditionalGeneration, 
     processor:MLlavaProcessor, 
     max_input_length:int=None, 
@@ -38,13 +38,26 @@ def chat_mllava(
             conv.append_message(message["role"], message["text"])
     else:
         history = []
-    conv.append_message(conv.roles[0], text)
-    conv.append_message(conv.roles[1], "")
+    
+    if text is not None:
+        conv.append_message(conv.roles[0], text)
+        conv.append_message(conv.roles[1], "")
+        history.append({"role": conv.roles[0], "text": text})
+        history.append({"role": conv.roles[1], "text": ""})
+    else:
+        assert history, "The history should not be empty if the text is None"
+        assert history[-1]['role'] == conv.roles[1], "The last message in the history should be the assistant, an empty message"
+        assert history[-2]['text'], "The last user message in the history should not be empty"
+        assert history[-1]['text'] == "", "The last assistant message in the history should be empty"
     
     prompt = conv.get_prompt()
+    if images:
+        for i in range(len(images)):
+            if isinstance(images[i], str):
+                images[i] = PIL.Image.open(images[i])
     
     inputs = processor(images=images, text=prompt, return_tensors="pt", truncation=True, max_length=max_input_length)
-    inputs = {k: v.to(model.device) for k, v in inputs.items()}
+    inputs = {k: v.to(model.device) if v is not None else v for k, v in inputs.items()}
     
     if stream:
         from transformers import TextIteratorStreamer
@@ -54,8 +67,6 @@ def chat_mllava(
         inputs.update(kwargs)
         thread = Thread(target=model.generate, kwargs=inputs)
         thread.start()
-        history.append({"role": conv.roles[0], "text": text})
-        history.append({"role": conv.roles[1], "text": ""})
         for _output in streamer:
             history[-1]["text"] += _output
             yield history[-1]["text"], history
@@ -67,7 +78,6 @@ def chat_mllava(
         generated_ids = output_ids[inputs["input_ids"].shape[-1]:]
         generated_text = processor.decode(generated_ids, skip_special_tokens=True)
 
-        history.append({"role": conv.roles[0], "text": text})
-        history.append({"role": conv.roles[1], "text": generated_text})
+        history[-1]["text"] = history[-1]["text"].strip()
         
         return generated_text, history

requirements.txt

@@ -1 +1,6 @@
-git
+torch
+transformers
+Pillow
+gradio
+spaces
+multiprocess