Update

.pre-commit-config.yaml

@@ -1,6 +1,6 @@
 repos:
   - repo: https://github.com/pre-commit/pre-commit-hooks
-    rev: v5.0.0
+    rev: v6.0.0
     hooks:
       - id: check-executables-have-shebangs
       - id: check-json
@@ -14,13 +14,13 @@ repos:
       - id: requirements-txt-fixer
       - id: trailing-whitespace
   - repo: https://github.com/astral-sh/ruff-pre-commit
-    rev: v0.9.9
+    rev: v0.15.4
     hooks:
-      - id: ruff
+      - id: ruff-check
         args: ["--fix"]
       - id: ruff-format
   - repo: https://github.com/pre-commit/mirrors-mypy
-    rev: v1.15.0
+    rev: v1.19.1
     hooks:
       - id: mypy
         args: ["--ignore-missing-imports"]

.python-version

@@ -1 +1 @@
-3.10
+3.12

README.md

@@ -4,7 +4,8 @@ emoji: 🔥
 colorFrom: blue
 colorTo: yellow
 sdk: gradio
-sdk_version: 5.40.0
+sdk_version: 6.8.0
+python_version: "3.12"
 app_file: app.py
 pinned: false
 ---

app.py

@@ -12,12 +12,12 @@ import spaces
 import torch
 from loguru import logger
 from PIL import Image
-from transformers import AutoProcessor, Gemma3ForConditionalGeneration, TextIteratorStreamer
+from transformers import AutoProcessor, BatchFeature, Gemma3ForConditionalGeneration, TextIteratorStreamer
 
-model_id = os.getenv("MODEL_ID", "google/gemma-3-12b-it")
-processor = AutoProcessor.from_pretrained(model_id, padding_side="left")
+MODEL_ID = os.getenv("MODEL_ID", "google/gemma-3-12b-it")
+processor = AutoProcessor.from_pretrained(MODEL_ID, padding_side="left")
 model = Gemma3ForConditionalGeneration.from_pretrained(
-    model_id, device_map="auto", torch_dtype=torch.bfloat16, attn_implementation="eager"
+    MODEL_ID, device_map="auto", torch_dtype=torch.bfloat16, attn_implementation="eager"
 )
 
 MAX_NUM_IMAGES = int(os.getenv("MAX_NUM_IMAGES", "5"))
@@ -164,6 +164,39 @@ def process_history(history: list[dict]) -> list[dict]:
 
 
 @spaces.GPU(duration=120)
+def _run_on_gpu(inputs: BatchFeature, max_new_tokens: int) -> Iterator[str]:
+    inputs = inputs.to(device=model.device, dtype=torch.bfloat16)
+
+    streamer = TextIteratorStreamer(processor, timeout=30.0, skip_prompt=True, skip_special_tokens=True)
+    generate_kwargs = {
+        **inputs,
+        "streamer": streamer,
+        "max_new_tokens": max_new_tokens,
+        "disable_compile": True,
+    }
+
+    exception_holder: list[Exception] = []
+
+    def _generate() -> None:
+        try:
+            model.generate(**generate_kwargs)
+        except Exception as e:  # noqa: BLE001
+            exception_holder.append(e)
+
+    thread = Thread(target=_generate)
+    thread.start()
+
+    chunks: list[str] = []
+    for text in streamer:
+        chunks.append(text)
+        yield "".join(chunks)
+
+    thread.join()
+    if exception_holder:
+        msg = f"Generation failed: {exception_holder[0]}"
+        raise gr.Error(msg)
+
+
 def run(message: dict, history: list[dict], system_prompt: str = "", max_new_tokens: int = 512) -> Iterator[str]:
     if not validate_media_constraints(message, history):
         yield ""
@@ -181,22 +214,9 @@ def run(message: dict, history: list[dict], system_prompt: str = "", max_new_tok
         tokenize=True,
         return_dict=True,
         return_tensors="pt",
-    ).to(device=model.device, dtype=torch.bfloat16)
-
-    streamer = TextIteratorStreamer(processor, timeout=30.0, skip_prompt=True, skip_special_tokens=True)
-    generate_kwargs = dict(
-        inputs,
-        streamer=streamer,
-        max_new_tokens=max_new_tokens,
-        disable_compile=True,
     )
-    t = Thread(target=model.generate, kwargs=generate_kwargs)
-    t.start()
 
-    output = ""
-    for delta in streamer:
-        output += delta
-        yield output
+    yield from _run_on_gpu(inputs=inputs, max_new_tokens=max_new_tokens)
 
 
 examples = [
@@ -330,8 +350,7 @@ You can upload images, interleaved images and videos. Note that video input only
 
 demo = gr.ChatInterface(
     fn=run,
-    type="messages",
-    chatbot=gr.Chatbot(type="messages", scale=1, allow_tags=["image"]),
+    chatbot=gr.Chatbot(scale=1, allow_tags=["image"]),
     textbox=gr.MultimodalTextbox(file_types=["image", ".mp4"], file_count="multiple", autofocus=True),
     multimodal=True,
     additional_inputs=[
@@ -344,9 +363,9 @@ demo = gr.ChatInterface(
     examples=examples,
     run_examples_on_click=False,
     cache_examples=False,
-    css_paths="style.css",
     delete_cache=(1800, 1800),
+    fill_height=True,
 )
 
 if __name__ == "__main__":
-    demo.launch()
+    demo.launch(css_paths="style.css")

pyproject.toml

@@ -3,18 +3,17 @@ name = "gemma-3-12b-it"
 version = "0.1.0"
 description = ""
 readme = "README.md"
-requires-python = ">=3.10"
+requires-python = ">=3.12"
 dependencies = [
-    "accelerate>=1.9.0",
-    "gradio>=5.40.0",
-    "hf-transfer>=0.1.9",
+    "accelerate>=1.12.0",
+    "gradio>=6.8.0",
     "loguru>=0.7.3",
     "opencv-python-headless>=4.11.0.86",
     "protobuf>=6.30.0",
     "sentencepiece>=0.2.0",
-    "spaces>=0.39.0",
-    "torch==2.4.0",
-    "transformers==4.51.3",
+    "spaces>=0.47.0",
+    "torch==2.9.1",
+    "transformers>=5.2.0",
 ]
 
 [tool.ruff]
@@ -39,7 +38,7 @@ ignore = [
     "EM101",   # raw-string-in-exception
     "FBT001",  # boolean-type-hint-positional-argument
     "FBT002",  # boolean-default-value-positional-argument
-    "PD901",   # pandas-df-variable-name
+    "ISC001",  # single-line-implicit-string-concatenation
     "PGH003",  # blanket-type-ignore
     "PLR0913", # too-many-arguments
     "PLR0915", # too-many-statements
@@ -57,3 +56,13 @@ convention = "google"
 
 [tool.ruff.format]
 docstring-code-format = true
+
+[dependency-groups]
+dev = [
+    "mypy>=1.19.1",
+    "pre-commit>=4.5.1",
+    "ruff>=0.15.4",
+]
+hf-spaces = [
+    "datasets",
+]

requirements.txt

@@ -1,51 +1,77 @@
 # This file was autogenerated by uv via the following command:
-#    uv pip compile pyproject.toml -o requirements.txt
-accelerate==1.9.0
-    # via gemma-3-12b-it (pyproject.toml)
+#    uv export --no-hashes --no-dev --group hf-spaces --no-emit-package typer-slim -o requirements.txt
+accelerate==1.12.0
+    # via gemma-3-12b-it
 aiofiles==24.1.0
     # via gradio
+aiohappyeyeballs==2.6.1
+    # via aiohttp
+aiohttp==3.13.3
+    # via fsspec
+aiosignal==1.4.0
+    # via aiohttp
+annotated-doc==0.0.4
+    # via
+    #   fastapi
+    #   typer
 annotated-types==0.7.0
     # via pydantic
-anyio==4.10.0
+anyio==4.12.1
     # via
     #   gradio
     #   httpx
     #   starlette
-brotli==1.1.0
+attrs==25.4.0
+    # via aiohttp
+audioop-lts==0.2.2 ; python_full_version >= '3.13'
     # via gradio
-certifi==2025.8.3
+brotli==1.2.0
+    # via gradio
+certifi==2026.2.25
     # via
     #   httpcore
     #   httpx
     #   requests
-charset-normalizer==3.4.2
+charset-normalizer==3.4.4
     # via requests
-click==8.2.1
+click==8.3.1
     # via
     #   typer
     #   uvicorn
-exceptiongroup==1.3.0
-    # via anyio
-fastapi==0.116.1
+colorama==0.4.6 ; sys_platform == 'win32'
+    # via
+    #   click
+    #   loguru
+    #   tqdm
+datasets==4.6.1
+dill==0.4.0
+    # via
+    #   datasets
+    #   multiprocess
+fastapi==0.135.1
     # via gradio
-ffmpy==0.6.1
+ffmpy==1.0.0
     # via gradio
-filelock==3.18.0
+filelock==3.25.0
     # via
+    #   datasets
     #   huggingface-hub
     #   torch
-    #   transformers
-    #   triton
-fsspec==2025.7.0
+frozenlist==1.8.0
     # via
+    #   aiohttp
+    #   aiosignal
+fsspec==2026.2.0
+    # via
+    #   datasets
     #   gradio-client
     #   huggingface-hub
     #   torch
-gradio==5.40.0
+gradio==6.8.0
     # via
-    #   gemma-3-12b-it (pyproject.toml)
+    #   gemma-3-12b-it
     #   spaces
-gradio-client==1.11.0
+gradio-client==2.2.0
     # via gradio
 groovy==0.1.2
     # via gradio
@@ -53,39 +79,41 @@ h11==0.16.0
     # via
     #   httpcore
     #   uvicorn
-hf-transfer==0.1.9
-    # via gemma-3-12b-it (pyproject.toml)
-hf-xet==1.1.5
+hf-xet==1.3.2 ; platform_machine == 'AMD64' or platform_machine == 'aarch64' or platform_machine == 'amd64' or platform_machine == 'arm64' or platform_machine == 'x86_64'
     # via huggingface-hub
 httpcore==1.0.9
     # via httpx
 httpx==0.28.1
     # via
+    #   datasets
     #   gradio
     #   gradio-client
+    #   huggingface-hub
     #   safehttpx
     #   spaces
-huggingface-hub==0.34.3
+huggingface-hub==1.5.0
     # via
     #   accelerate
+    #   datasets
     #   gradio
     #   gradio-client
     #   tokenizers
     #   transformers
-idna==3.10
+idna==3.11
     # via
     #   anyio
     #   httpx
     #   requests
+    #   yarl
 jinja2==3.1.6
     # via
     #   gradio
     #   torch
 loguru==0.7.3
-    # via gemma-3-12b-it (pyproject.toml)
-markdown-it-py==3.0.0
+    # via gemma-3-12b-it
+markdown-it-py==4.0.0
     # via rich
-markupsafe==3.0.2
+markupsafe==3.0.3
     # via
     #   gradio
     #   jinja2
@@ -93,74 +121,98 @@ mdurl==0.1.2
     # via markdown-it-py
 mpmath==1.3.0
     # via sympy
-networkx==3.4.2
+multidict==6.7.1
+    # via
+    #   aiohttp
+    #   yarl
+multiprocess==0.70.18
+    # via datasets
+networkx==3.6.1
     # via torch
-numpy==2.2.6
+numpy==2.4.2
     # via
     #   accelerate
+    #   datasets
     #   gradio
     #   opencv-python-headless
     #   pandas
     #   transformers
-nvidia-cublas-cu12==12.1.3.1
+nvidia-cublas-cu12==12.8.4.1 ; platform_machine == 'x86_64' and sys_platform == 'linux'
     # via
     #   nvidia-cudnn-cu12
     #   nvidia-cusolver-cu12
     #   torch
-nvidia-cuda-cupti-cu12==12.1.105
+nvidia-cuda-cupti-cu12==12.8.90 ; platform_machine == 'x86_64' and sys_platform == 'linux'
+    # via torch
+nvidia-cuda-nvrtc-cu12==12.8.93 ; platform_machine == 'x86_64' and sys_platform == 'linux'
     # via torch
-nvidia-cuda-nvrtc-cu12==12.1.105
+nvidia-cuda-runtime-cu12==12.8.90 ; platform_machine == 'x86_64' and sys_platform == 'linux'
     # via torch
-nvidia-cuda-runtime-cu12==12.1.105
+nvidia-cudnn-cu12==9.10.2.21 ; platform_machine == 'x86_64' and sys_platform == 'linux'
     # via torch
-nvidia-cudnn-cu12==9.1.0.70
+nvidia-cufft-cu12==11.3.3.83 ; platform_machine == 'x86_64' and sys_platform == 'linux'
     # via torch
-nvidia-cufft-cu12==11.0.2.54
+nvidia-cufile-cu12==1.13.1.3 ; platform_machine == 'x86_64' and sys_platform == 'linux'
     # via torch
-nvidia-curand-cu12==10.3.2.106
+nvidia-curand-cu12==10.3.9.90 ; platform_machine == 'x86_64' and sys_platform == 'linux'
     # via torch
-nvidia-cusolver-cu12==11.4.5.107
+nvidia-cusolver-cu12==11.7.3.90 ; platform_machine == 'x86_64' and sys_platform == 'linux'
     # via torch
-nvidia-cusparse-cu12==12.1.0.106
+nvidia-cusparse-cu12==12.5.8.93 ; platform_machine == 'x86_64' and sys_platform == 'linux'
     # via
     #   nvidia-cusolver-cu12
     #   torch
-nvidia-nccl-cu12==2.20.5
+nvidia-cusparselt-cu12==0.7.1 ; platform_machine == 'x86_64' and sys_platform == 'linux'
     # via torch
-nvidia-nvjitlink-cu12==12.9.86
+nvidia-nccl-cu12==2.27.5 ; platform_machine == 'x86_64' and sys_platform == 'linux'
+    # via torch
+nvidia-nvjitlink-cu12==12.8.93 ; platform_machine == 'x86_64' and sys_platform == 'linux'
     # via
+    #   nvidia-cufft-cu12
     #   nvidia-cusolver-cu12
     #   nvidia-cusparse-cu12
-nvidia-nvtx-cu12==12.1.105
+    #   torch
+nvidia-nvshmem-cu12==3.3.20 ; platform_machine == 'x86_64' and sys_platform == 'linux'
     # via torch
-opencv-python-headless==4.12.0.88
-    # via gemma-3-12b-it (pyproject.toml)
-orjson==3.11.1
+nvidia-nvtx-cu12==12.8.90 ; platform_machine == 'x86_64' and sys_platform == 'linux'
+    # via torch
+opencv-python-headless==4.13.0.92
+    # via gemma-3-12b-it
+orjson==3.11.7
     # via gradio
-packaging==25.0
+packaging==26.0
     # via
     #   accelerate
+    #   datasets
     #   gradio
     #   gradio-client
     #   huggingface-hub
     #   spaces
     #   transformers
-pandas==2.3.1
-    # via gradio
-pillow==11.3.0
+pandas==3.0.1
+    # via
+    #   datasets
+    #   gradio
+pillow==12.1.1
     # via gradio
-protobuf==6.31.1
-    # via gemma-3-12b-it (pyproject.toml)
+propcache==0.4.1
+    # via
+    #   aiohttp
+    #   yarl
+protobuf==7.34.0
+    # via gemma-3-12b-it
 psutil==5.9.8
     # via
     #   accelerate
     #   spaces
-pydantic==2.11.7
+pyarrow==23.0.1
+    # via datasets
+pydantic==2.12.5
     # via
     #   fastapi
     #   gradio
     #   spaces
-pydantic-core==2.33.2
+pydantic-core==2.41.5
     # via pydantic
 pydub==0.25.1
     # via gradio
@@ -168,73 +220,75 @@ pygments==2.19.2
     # via rich
 python-dateutil==2.9.0.post0
     # via pandas
-python-multipart==0.0.20
+python-multipart==0.0.22
     # via gradio
 pytz==2025.2
-    # via pandas
-pyyaml==6.0.2
+    # via gradio
+pyyaml==6.0.3
     # via
     #   accelerate
+    #   datasets
     #   gradio
     #   huggingface-hub
     #   transformers
-regex==2025.7.34
+regex==2026.2.28
     # via transformers
-requests==2.32.4
+requests==2.32.5
     # via
-    #   huggingface-hub
+    #   datasets
     #   spaces
-    #   transformers
-rich==14.1.0
+rich==14.3.3
     # via typer
-ruff==0.12.7
+safehttpx==0.1.7
     # via gradio
-safehttpx==0.1.6
-    # via gradio
-safetensors==0.5.3
+safetensors==0.7.0
     # via
     #   accelerate
     #   transformers
 semantic-version==2.10.0
     # via gradio
-sentencepiece==0.2.0
-    # via gemma-3-12b-it (pyproject.toml)
+sentencepiece==0.2.1
+    # via gemma-3-12b-it
+setuptools==82.0.0
+    # via torch
 shellingham==1.5.4
     # via typer
 six==1.17.0
     # via python-dateutil
-sniffio==1.3.1
-    # via anyio
-spaces==0.39.0
-    # via gemma-3-12b-it (pyproject.toml)
-starlette==0.47.2
+spaces==0.47.0
+    # via gemma-3-12b-it
+starlette==0.52.1
     # via
     #   fastapi
     #   gradio
 sympy==1.14.0
     # via torch
-tokenizers==0.21.4
+tokenizers==0.22.2
     # via transformers
 tomlkit==0.13.3
     # via gradio
-torch==2.4.0
+torch==2.9.1
     # via
-    #   gemma-3-12b-it (pyproject.toml)
     #   accelerate
-tqdm==4.67.1
+    #   gemma-3-12b-it
+tqdm==4.67.3
     # via
+    #   datasets
     #   huggingface-hub
     #   transformers
-transformers==4.51.3
-    # via gemma-3-12b-it (pyproject.toml)
-triton==3.0.0
+transformers==5.2.0
+    # via gemma-3-12b-it
+triton==3.5.1 ; platform_machine == 'x86_64' and sys_platform == 'linux'
     # via torch
-typer==0.16.0
-    # via gradio
-typing-extensions==4.14.1
+typer==0.24.1
     # via
+    #   gradio
+    #   huggingface-hub
+    #   typer-slim
+typing-extensions==4.15.0
+    # via
+    #   aiosignal
     #   anyio
-    #   exceptiongroup
     #   fastapi
     #   gradio
     #   gradio-client
@@ -244,16 +298,20 @@ typing-extensions==4.14.1
     #   spaces
     #   starlette
     #   torch
-    #   typer
     #   typing-inspection
-    #   uvicorn
-typing-inspection==0.4.1
-    # via pydantic
-tzdata==2025.2
+typing-inspection==0.4.2
+    # via
+    #   fastapi
+    #   pydantic
+tzdata==2025.3 ; sys_platform == 'emscripten' or sys_platform == 'win32'
     # via pandas
-urllib3==2.5.0
+urllib3==2.6.3
     # via requests
-uvicorn==0.35.0
+uvicorn==0.41.0
     # via gradio
-websockets==15.0.1
-    # via gradio-client
+win32-setctime==1.2.0 ; sys_platform == 'win32'
+    # via loguru
+xxhash==3.6.0
+    # via datasets
+yarl==1.23.0
+    # via aiohttp