try in docker

.dockerignore

@@ -0,0 +1,2 @@
+__pycache__
+bloomz.cpp

.gitignore

@@ -0,0 +1,142 @@
+# Byte-compiled / optimized / DLL files
+__pycache__/
+*.py[cod]
+*$py.class
+
+# C extensions
+*.so
+
+# Distribution / packaging
+.Python
+build/
+develop-eggs/
+dist/
+downloads/
+eggs/
+.eggs/
+lib64/
+parts/
+sdist/
+var/
+wheels/
+pip-wheel-metadata/
+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/
+
+# 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
+target/
+
+# Jupyter Notebook
+.ipynb_checkpoints
+
+# IPython
+profile_default/
+ipython_config.py
+
+# pyenv
+.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
+
+# PEP 582; used by e.g. github.com/David-OConnor/pyflow
+__pypackages__/
+
+# Celery stuff
+celerybeat-schedule
+celerybeat.pid
+
+# SageMath parsed files
+*.sage.py
+
+# Environments
+.env
+.venv
+.venv*
+env/
+venv/
+ENV/
+env.bak/
+venv.bak/
+.venv*
+
+# Spyder project settings
+.spyderproject
+.spyproject
+
+# Rope project settings
+.ropeproject
+
+# mkdocs documentation
+/site
+
+# mypy
+.mypy_cache/
+.dmypy.json
+dmypy.json
+
+# Pyre type checker
+.pyre/
+.vscode/
+.idea/
+
+.DS_Store
+
+# Ruff
+.ruff_cache
+
+# Spell checker config
+cspell.json
+
+bloomz.cpp

Dockerfile

@@ -0,0 +1,44 @@
+# See https://hub.docker.com/r/nikolaik/python-nodejs
+#     https://github.com/nikolaik/docker-python-nodejs
+# Default user is 'pn' with uid 1000, gid 1000
+FROM python:3.10
+
+RUN apt-get update && apt-get install -y \
+	git \
+	cmake \
+	&& rm -rf /var/lib/apt/lists/*
+
+# User
+RUN useradd -m -u 1000 user
+USER user
+ENV HOME=/home/user \
+	PATH=/home/user/.local/bin:$PATH
+WORKDIR /home/user/app
+
+# Install Python deps
+RUN pip install --no-cache-dir pip && \
+    pip install --no-cache-dir \
+        torch \
+        numpy \
+        transformers
+
+# Prepare bloomz.cpp
+RUN git clone https://github.com/NouamaneTazi/bloomz.cpp@4fc96cbf2e2c257eaca1cd7b7ed8e31741e672ee
+
+WORKDIR $HOME/app/bloomz.cpp
+
+RUN make
+
+# Add files
+COPY . /home/user/app
+
+ENV PYTHONPATH=$HOME/app \
+	PYTHONUNBUFFERED=1 \
+	GRADIO_ALLOW_FLAGGING=never \
+	GRADIO_NUM_PORTS=1 \
+	GRADIO_SERVER_NAME=0.0.0.0 \
+	GRADIO_THEME=huggingface \
+	SYSTEM=spaces
+
+# Run script
+CMD ["python", "app.py"]

README.md

@@ -3,9 +3,8 @@ title: Bloom.cpp Converter
 emoji: 👁
 colorFrom: green
 colorTo: pink
-sdk: gradio
-sdk_version: 3.21.0
-app_file: app.py
+sdk: docker
+app_port: 7860
 pinned: false
 ---
 

app.py

@@ -0,0 +1,77 @@
+from pathlib import Path
+
+import gradio as gr
+from huggingface_hub import HfApi, Repository
+from huggingface_hub.utils import RepositoryNotFoundError
+
+from convert import convert
+
+REPO_PATH = Path("bloomz.cpp")
+
+repo = Repository(local_dir="bloomz.cpp", clone_from="https://github.com/NouamaneTazi/bloomz.cpp")
+
+
+def run(token: str, model_id: str, precision: str, quantization: bool) -> str:
+    if token == "" or model_id == "":
+        return """
+        ### Invalid input 🐞
+        
+        Please fill a token and model_id.
+        """
+
+    api = HfApi(token=token)
+    try:
+        # TODO: make a PR to bloomz.cpp to be able to pass a token
+        api.model_info(repo_id=model_id, token=False)  # only public repos are accessible
+    except RepositoryNotFoundError:
+        return f"""
+        ### Error 😢😢😢
+        
+        Repository {model_id} not found. Only public models are convertible at the moment.
+        """
+
+    try:
+        model_path = convert(model_id=model_id, precision=precision, quantization=quantization)
+        print("[commit_info]", model_path)
+
+        return f"""
+        ### Success 🔥
+        """
+        return f"""
+        ### Success 🔥
+        Yay! This model was successfully converted and a PR was open using your token, here:
+        # [{commit_info.pr_url}]({commit_info.pr_url})
+        """
+    except Exception as e:
+        return f"""
+        ### Error 😢😢😢
+        
+        {e}
+        """
+
+
+DESCRIPTION = """
+The steps are the following:
+- Paste a read-access token from hf.co/settings/tokens. Read access is enough given that we will open a PR against the source repo.
+- Input a model id from the Hub
+- Click "Submit"
+- That's it! You'll get feedback if it works or not, and if it worked, you'll get the URL of the opened PR 🔥
+⚠️ For now only `pytorch_model.bin` files are supported but we'll extend in the future.
+"""
+
+demo = gr.Interface(
+    title="Convert any BLOOM-like model to be compatible with bloomz.cpp",
+    description=DESCRIPTION,
+    allow_flagging="never",
+    article="Check out the [bloomz.cpp](https://github.com/NouamaneTazi/bloomz.cpp) repo on GitHub",
+    inputs=[
+        gr.Text(max_lines=1, label="your hf_token"),
+        gr.Text(max_lines=1, label="model_id (e.g.: bigscience/bloomz-7b1)"),
+        gr.Radio(choices=["FP16", "FP32"], label="Precision", value="FP16"),
+        gr.Checkbox(value=False, label="4-bits quantization"),
+    ],
+    outputs=[gr.Markdown(label="output")],
+    fn=run,
+).queue()
+
+demo.launch()

convert.py

@@ -0,0 +1,36 @@
+from pathlib import Path
+from subprocess import run
+
+
+def convert(model_id:str, precision:str, quantization:bool)->Path:
+    # Conversion
+    cmd = [
+        "python",
+        "./bloomz.cpp/convert-hf-to-ggml.py",
+        model_id, # bigscience/bloomz-560m
+        "./bloomz.cpp/models",
+    ]
+    if precision == "FP32":
+        cmd.append("--use-fp32")
+    run(cmd, check=True)
+
+    # Model file should exist
+    f_suffix = "f32" if precision == "FP32" else "f16"
+    _, model_name = model_id.split("/")
+    model_path = Path(f"./bloomz.cpp/models/ggml-model-{model_name}-{f_suffix}.bin")
+    assert model_path.is_file()
+
+    # Quantization
+    if quantization:
+        cmd = [
+            "./bloomz.cpp/quantize",
+            f"./bloomz.cpp/models/ggml-model-{model_name}-{f_suffix}.bin",
+            f"./bloomz.cpp/models/ggml-model-{model_name}-{f_suffix}-q4_0.bin",
+            "2",
+        ]
+        run(cmd, check=True)
+        model_path = Path(f"./bloomz.cpp/models/ggml-model-{model_name}-{f_suffix}.bin")
+        assert model_path.is_file()
+
+    # Return
+    return model_path