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: v0.12.0
+    rev: v0.15.4
     hooks:
       - id: ruff-check
         args: ["--fix"]
       - id: ruff-format
   - repo: https://github.com/pre-commit/mirrors-mypy
-    rev: v1.16.1
+    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: red
 colorTo: purple
 sdk: gradio
-sdk_version: 5.40.0
+sdk_version: 6.8.0
+python_version: "3.12"
 app_file: app.py
 pinned: false
 ---

app.py

@@ -9,13 +9,13 @@ import gradio as gr
 import spaces
 import torch
 from gradio.utils import get_upload_folder
-from transformers import AutoModelForImageTextToText, AutoProcessor
+from transformers import AutoModelForImageTextToText, AutoProcessor, BatchFeature
 from transformers.generation.streamers import TextIteratorStreamer
 
-model_id = "google/gemma-3n-E4B-it"
+MODEL_ID = "google/gemma-3n-E4B-it"
 
-processor = AutoProcessor.from_pretrained(model_id)
-model = AutoModelForImageTextToText.from_pretrained(model_id, device_map="auto", torch_dtype=torch.bfloat16)
+processor = AutoProcessor.from_pretrained(MODEL_ID)
+model = AutoModelForImageTextToText.from_pretrained(MODEL_ID, device_map="auto", torch_dtype=torch.bfloat16)
 
 IMAGE_FILE_TYPES = (".jpg", ".jpeg", ".png", ".webp")
 VIDEO_FILE_TYPES = (".mp4", ".mov", ".webm")
@@ -50,15 +50,12 @@ def count_files_in_new_message(paths: list[str]) -> tuple[int, int]:
     return video_count, non_video_count
 
 
-def validate_media_constraints(message: dict) -> bool:
+def validate_media_constraints(message: dict) -> None:
     video_count, non_video_count = count_files_in_new_message(message["files"])
     if video_count > 1:
-        gr.Warning("Only one video is supported.")
-        return False
+        raise gr.Error("Only one video is supported.")
     if video_count == 1 and non_video_count > 0:
-        gr.Warning("Mixing images and videos is not allowed.")
-        return False
-    return True
+        raise gr.Error("Mixing images and videos is not allowed.")
 
 
 def extract_frames_to_tempdir(
@@ -154,10 +151,42 @@ def process_history(history: list[dict]) -> list[dict]:
 
 @spaces.GPU(duration=120)
 @torch.inference_mode()
+def _generate_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,
+        "do_sample": False,
+        "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 generate(message: dict, history: list[dict], system_prompt: str = "", max_new_tokens: int = 512) -> Iterator[str]:
-    if not validate_media_constraints(message):
-        yield ""
-        return
+    validate_media_constraints(message)
 
     messages = []
     if system_prompt:
@@ -174,29 +203,10 @@ def generate(message: dict, history: list[dict], system_prompt: str = "", max_ne
     )
     n_tokens = inputs["input_ids"].shape[1]
     if n_tokens > MAX_INPUT_TOKENS:
-        gr.Warning(
-            f"Input too long. Max {MAX_INPUT_TOKENS} tokens. Got {n_tokens} tokens. This limit is set to avoid CUDA out-of-memory errors in this Space."
-        )
-        yield ""
-        return
-
-    inputs = inputs.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,
-        do_sample=False,
-        disable_compile=True,
-    )
-    t = Thread(target=model.generate, kwargs=generate_kwargs)
-    t.start()
+        msg = f"Input too long ({n_tokens} tokens). Maximum is {MAX_INPUT_TOKENS} tokens."
+        raise gr.Error(msg)
 
-    output = ""
-    for delta in streamer:
-        output += delta
-        yield output
+    yield from _generate_on_gpu(inputs=inputs, max_new_tokens=max_new_tokens)
 
 
 examples = [
@@ -234,7 +244,6 @@ examples = [
 
 demo = gr.ChatInterface(
     fn=generate,
-    type="messages",
     textbox=gr.MultimodalTextbox(
         file_types=list(IMAGE_FILE_TYPES + VIDEO_FILE_TYPES + AUDIO_FILE_TYPES),
         file_count="multiple",
@@ -250,9 +259,8 @@ demo = gr.ChatInterface(
     examples=examples,
     run_examples_on_click=False,
     cache_examples=False,
-    css_paths="style.css",
     delete_cache=(1800, 1800),
 )
 
 if __name__ == "__main__":
-    demo.launch()
+    demo.launch(css_paths="style.css")

pyproject.toml

@@ -3,18 +3,17 @@ name = "gemma-3n-e4b-it"
 version = "0.1.0"
 description = ""
 readme = "README.md"
-requires-python = ">=3.10"
+requires-python = ">=3.12"
 dependencies = [
-    "accelerate>=1.9.0",
+    "accelerate>=1.12.0",
     "av>=14.4.0",
-    "gradio>=5.40.0",
-    "hf-transfer>=0.1.9",
+    "gradio>=6.8.0",
     "librosa>=0.11.0",
-    "spaces>=0.39.0",
+    "spaces>=0.47.0",
     "timm>=1.0.16",
-    "torch==2.5.1",
+    "torch==2.9.1",
     "torchvision>=0.20.1",
-    "transformers>=4.54.1",
+    "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 +56,12 @@ convention = "google"
 
 [tool.ruff.format]
 docstring-code-format = true
+
+[dependency-groups]
+dev = [
+    "pre-commit>=4.5.1",
+    "ruff>=0.15.4",
+]
+hf-spaces = [
+    "datasets",
+]

requirements.txt

@@ -1,59 +1,87 @@
 # This file was autogenerated by uv via the following command:
-#    uv pip compile pyproject.toml -o requirements.txt
-accelerate==1.9.0
-    # via gemma-3n-e4b-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-3n-e4b-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
-audioread==3.0.1
+attrs==25.4.0
+    # via aiohttp
+audioop-lts==0.2.2 ; python_full_version >= '3.13'
+    # via
+    #   gradio
+    #   standard-aifc
+    #   standard-sunau
+audioread==3.1.0
     # via librosa
-av==15.0.0
-    # via gemma-3n-e4b-it (pyproject.toml)
-brotli==1.1.0
+av==16.1.0
+    # via gemma-3n-e4b-it
+brotli==1.2.0
     # via gradio
-certifi==2025.8.3
+certifi==2026.2.25
     # via
     #   httpcore
     #   httpx
     #   requests
-cffi==1.17.1
+cffi==2.0.0
     # via soundfile
-charset-normalizer==3.4.2
+charset-normalizer==3.4.4
     # via requests
-click==8.2.1
+click==8.3.1
     # via
     #   typer
     #   uvicorn
+colorama==0.4.6 ; sys_platform == 'win32'
+    # via
+    #   click
+    #   tqdm
+datasets==4.6.1
 decorator==5.2.1
     # via librosa
-exceptiongroup==1.3.0
-    # via anyio
-fastapi==0.116.1
+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-3n-e4b-it (pyproject.toml)
+    #   gemma-3n-e4b-it
     #   spaces
-gradio-client==1.11.0
+gradio-client==2.2.0
     # via gradio
 groovy==0.1.2
     # via gradio
@@ -61,48 +89,50 @@ h11==0.16.0
     # via
     #   httpcore
     #   uvicorn
-hf-transfer==0.1.9
-    # via gemma-3n-e4b-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
     #   timm
     #   tokenizers
     #   transformers
-idna==3.10
+idna==3.11
     # via
     #   anyio
     #   httpx
     #   requests
+    #   yarl
 jinja2==3.1.6
     # via
     #   gradio
     #   torch
-joblib==1.5.1
+joblib==1.5.3
     # via
     #   librosa
     #   scikit-learn
 lazy-loader==0.4
     # via librosa
 librosa==0.11.0
-    # via gemma-3n-e4b-it (pyproject.toml)
-llvmlite==0.44.0
+    # via gemma-3n-e4b-it
+llvmlite==0.46.0
     # via numba
-markdown-it-py==3.0.0
+markdown-it-py==4.0.0
     # via rich
-markupsafe==3.0.2
+markupsafe==3.0.3
     # via
     #   gradio
     #   jinja2
@@ -110,15 +140,22 @@ mdurl==0.1.2
     # via markdown-it-py
 mpmath==1.3.0
     # via sympy
-msgpack==1.1.1
+msgpack==1.1.2
     # via librosa
-networkx==3.4.2
+multidict==6.7.1
+    # via
+    #   aiohttp
+    #   yarl
+multiprocess==0.70.18
+    # via datasets
+networkx==3.6.1
     # via torch
-numba==0.61.2
+numba==0.64.0
     # via librosa
-numpy==2.2.6
+numpy==2.4.2
     # via
     #   accelerate
+    #   datasets
     #   gradio
     #   librosa
     #   numba
@@ -129,43 +166,51 @@ numpy==2.2.6
     #   soxr
     #   torchvision
     #   transformers
-nvidia-cublas-cu12==12.4.5.8
+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==12.4.127
+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==12.4.127
+nvidia-cuda-runtime-cu12==12.8.90 ; platform_machine == 'x86_64' and sys_platform == 'linux'
     # via torch
-nvidia-cuda-runtime-cu12==12.4.127
+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==11.2.1.3
+nvidia-cufile-cu12==1.13.1.3 ; platform_machine == 'x86_64' and sys_platform == 'linux'
     # via torch
-nvidia-curand-cu12==10.3.5.147
+nvidia-curand-cu12==10.3.9.90 ; platform_machine == 'x86_64' and sys_platform == 'linux'
     # via torch
-nvidia-cusolver-cu12==11.6.1.9
+nvidia-cusolver-cu12==11.7.3.90 ; platform_machine == 'x86_64' and sys_platform == 'linux'
     # via torch
-nvidia-cusparse-cu12==12.3.1.170
+nvidia-cusparse-cu12==12.5.8.93 ; platform_machine == 'x86_64' and sys_platform == 'linux'
     # via
     #   nvidia-cusolver-cu12
     #   torch
-nvidia-nccl-cu12==2.21.5
+nvidia-cusparselt-cu12==0.7.1 ; platform_machine == 'x86_64' and sys_platform == 'linux'
+    # via torch
+nvidia-nccl-cu12==2.27.5 ; platform_machine == 'x86_64' and sys_platform == 'linux'
     # via torch
-nvidia-nvjitlink-cu12==12.4.127
+nvidia-nvjitlink-cu12==12.8.93 ; platform_machine == 'x86_64' and sys_platform == 'linux'
     # via
+    #   nvidia-cufft-cu12
     #   nvidia-cusolver-cu12
     #   nvidia-cusparse-cu12
     #   torch
-nvidia-nvtx-cu12==12.4.127
+nvidia-nvshmem-cu12==3.3.20 ; platform_machine == 'x86_64' and sys_platform == 'linux'
     # via torch
-orjson==3.11.1
+nvidia-nvtx-cu12==12.8.90 ; platform_machine == 'x86_64' and sys_platform == 'linux'
+    # via torch
+orjson==3.11.7
     # via gradio
-packaging==25.0
+packaging==26.0
     # via
     #   accelerate
+    #   datasets
     #   gradio
     #   gradio-client
     #   huggingface-hub
@@ -173,28 +218,36 @@ packaging==25.0
     #   pooch
     #   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
     #   torchvision
-platformdirs==4.3.8
+platformdirs==4.9.2
     # via pooch
-pooch==1.8.2
+pooch==1.9.0
     # via librosa
+propcache==0.4.1
+    # via
+    #   aiohttp
+    #   yarl
 psutil==5.9.8
     # via
     #   accelerate
     #   spaces
-pycparser==2.22
+pyarrow==23.0.1
+    # via datasets
+pycparser==3.0 ; implementation_name != 'PyPy'
     # via cffi
-pydantic==2.11.7
+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
@@ -202,94 +255,106 @@ 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
     #   timm
     #   transformers
-regex==2025.7.34
+regex==2026.2.28
     # via transformers
-requests==2.32.4
+requests==2.32.5
     # via
-    #   huggingface-hub
+    #   datasets
     #   pooch
     #   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
     #   timm
     #   transformers
-scikit-learn==1.7.1
+scikit-learn==1.8.0
     # via librosa
-scipy==1.15.3
+scipy==1.17.1
     # via
     #   librosa
     #   scikit-learn
 semantic-version==2.10.0
     # via gradio
+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
 soundfile==0.13.1
     # via librosa
-soxr==0.5.0.post1
+soxr==1.0.0
     # via librosa
-spaces==0.39.0
-    # via gemma-3n-e4b-it (pyproject.toml)
-starlette==0.47.2
+spaces==0.47.0
+    # via gemma-3n-e4b-it
+standard-aifc==3.13.0 ; python_full_version >= '3.13'
+    # via
+    #   audioread
+    #   librosa
+standard-chunk==3.13.0 ; python_full_version >= '3.13'
+    # via standard-aifc
+standard-sunau==3.13.0 ; python_full_version >= '3.13'
+    # via
+    #   audioread
+    #   librosa
+starlette==0.52.1
     # via
     #   fastapi
     #   gradio
-sympy==1.13.1
+sympy==1.14.0
     # via torch
 threadpoolctl==3.6.0
     # via scikit-learn
-timm==1.0.19
-    # via gemma-3n-e4b-it (pyproject.toml)
-tokenizers==0.21.4
+timm==1.0.25
+    # via gemma-3n-e4b-it
+tokenizers==0.22.2
     # via transformers
 tomlkit==0.13.3
     # via gradio
-torch==2.5.1
+torch==2.9.1
     # via
-    #   gemma-3n-e4b-it (pyproject.toml)
     #   accelerate
+    #   gemma-3n-e4b-it
     #   timm
     #   torchvision
-torchvision==0.20.1
+torchvision==0.24.1
     # via
-    #   gemma-3n-e4b-it (pyproject.toml)
+    #   gemma-3n-e4b-it
     #   timm
-tqdm==4.67.1
+tqdm==4.67.3
     # via
+    #   datasets
     #   huggingface-hub
     #   transformers
-transformers==4.54.1
-    # via gemma-3n-e4b-it (pyproject.toml)
-triton==3.1.0
+transformers==5.2.0
+    # via gemma-3n-e4b-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
@@ -300,16 +365,18 @@ 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
+xxhash==3.6.0
+    # via datasets
+yarl==1.23.0
+    # via aiohttp