init

.dockerignore

@@ -0,0 +1,3 @@
+/downloads
+/llama.cpp
+*.gguf

.gitignore

@@ -0,0 +1,167 @@
+# 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/latest/usage/project/#working-with-version-control
+.pdm.toml
+.pdm-python
+.pdm-build/
+
+# 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.
+#.idea/
+
+/downloads
+!/downloads/.keep
+/llama.cpp
+*.gguf

Dockerfile

@@ -0,0 +1,44 @@
+FROM python:3.10.13-slim-bullseye
+
+ENV DEBIAN_FRONTEND=noninteractive
+RUN apt-get update && \
+    apt-get upgrade -y && \
+    apt-get install -y --no-install-recommends \
+    git \
+    git-lfs \
+    wget \
+    curl \
+    # python build dependencies \
+    build-essential
+
+RUN useradd -m -u 1000 user
+USER user
+ENV HOME=/home/user \
+    PATH=/home/user/.local/bin:${PATH}
+WORKDIR ${HOME}/app
+
+RUN pip install --no-cache-dir -U pip setuptools wheel && \
+    pip install "huggingface-hub" "hf-transfer" "gradio[oauth]>=4.28.0" "gradio_huggingfacehub_search==0.0.7" "APScheduler"
+
+COPY --chown=1000 . ${HOME}/app
+# TODO: revert once the PR is merged
+# RUN git clone https://github.com/ggerganov/llama.cpp --depth 1
+RUN git clone https://github.com/ngxson/llama.cpp -b xsn/lora_convert_base_is_optional --depth 1
+RUN pip install -r llama.cpp/requirements.txt
+
+ENV PYTHONPATH=${HOME}/app \
+    PYTHONUNBUFFERED=1 \
+    HF_HUB_ENABLE_HF_TRANSFER=1 \
+    GRADIO_ALLOW_FLAGGING=never \
+    GRADIO_NUM_PORTS=1 \
+    GRADIO_SERVER_NAME=0.0.0.0 \
+    GRADIO_THEME=huggingface \
+    TQDM_POSITION=-1 \
+    TQDM_MININTERVAL=1 \
+    SYSTEM=spaces \
+    LD_LIBRARY_PATH=/usr/local/cuda/lib64:${LD_LIBRARY_PATH} \
+    PATH=/usr/local/nvidia/bin:${PATH}
+
+
+ENTRYPOINT /bin/bash start.sh
+

app.py

@@ -0,0 +1,177 @@
+import os
+import subprocess
+import signal
+os.environ["GRADIO_ANALYTICS_ENABLED"] = "False"
+import gradio as gr
+import tempfile
+
+from huggingface_hub import HfApi, ModelCard, whoami
+from gradio_huggingfacehub_search import HuggingfaceHubSearch
+from pathlib import Path
+from textwrap import dedent
+from apscheduler.schedulers.background import BackgroundScheduler
+
+
+HF_TOKEN = os.environ.get("HF_TOKEN")
+CONVERSION_SCRIPT = "convert_lora_to_gguf.py"
+
+def process_model(peft_model_id: str, q_method: str, private_repo, oauth_token: gr.OAuthToken | None):
+    if oauth_token.token is None:
+        raise ValueError("You must be logged in to use GGUF-my-lora")
+    model_name = peft_model_id.split('/')[-1]
+    gguf_output_name = f"{model_name}-{q_method.lower()}.gguf"
+
+    try:
+        api = HfApi(token=oauth_token.token)
+
+        dl_pattern = ["*.md", "*.json", "*.model"]
+
+        pattern = (
+            "*.safetensors"
+            if any(
+                file.path.endswith(".safetensors")
+                for file in api.list_repo_tree(
+                    repo_id=peft_model_id,
+                    recursive=True,
+                )
+            )
+            else "*.bin"
+        )
+
+        dl_pattern += [pattern]
+
+        if not os.path.exists("downloads"):
+            os.makedirs("downloads")
+
+        with tempfile.TemporaryDirectory(dir="downloads") as tmpdir:
+            # Keep the model name as the dirname so the model name metadata is populated correctly
+            local_dir = Path(tmpdir)/model_name
+            print(local_dir)
+            api.snapshot_download(repo_id=peft_model_id, local_dir=local_dir, local_dir_use_symlinks=False, allow_patterns=dl_pattern)
+            print("Model downloaded successfully!")
+            print(f"Current working directory: {os.getcwd()}")
+            print(f"Model directory contents: {os.listdir(local_dir)}")
+
+            adapter_config_dir = local_dir/"adapter_config.json"
+            if not os.path.exists(adapter_config_dir):
+                raise Exception("adapter_config.json not found. Please ensure the selected repo is a PEFT LoRA model.")
+
+            fp16_conversion = f"python llama.cpp/{CONVERSION_SCRIPT} {local_dir} --outtype {q_method.lower()} --outfile {gguf_output_name}"
+            result = subprocess.run(fp16_conversion, shell=True, capture_output=True)
+            print(result)
+            if result.returncode != 0:
+                raise Exception(f"Error converting to GGUF {q_method}: {result.stderr}")
+            print("Model converted to GGUF successfully!")
+            print(f"Converted model path: {gguf_output_name}")
+
+        # Create empty repo
+        username = whoami(oauth_token.token)["name"]
+        new_repo_url = api.create_repo(repo_id=f"{username}/{model_name}-{q_method}-GGUF", exist_ok=True, private=private_repo)
+        new_repo_id = new_repo_url.repo_id
+        print("Repo created successfully!", new_repo_url)
+
+        # Upload the GGUF model
+        api.upload_file(
+            path_or_fileobj=gguf_output_name,
+            path_in_repo=gguf_output_name,
+            repo_id=new_repo_id,
+        )
+        print("Uploaded", gguf_output_name)
+
+        try:
+            card = ModelCard.load(peft_model_id, token=oauth_token.token)
+        except:
+            card = ModelCard("")
+        if card.data.tags is None:
+            card.data.tags = []
+        card.data.tags.append("llama-cpp")
+        card.data.tags.append("gguf-my-lora")
+        card.data.base_model = peft_model_id
+        card.text = dedent(
+            f"""
+            # {new_repo_id}
+            This LoRA adapter was converted to GGUF format from [`{peft_model_id}`](https://huggingface.co/{peft_model_id}) via the ggml.ai's [GGUF-my-lora](https://huggingface.co/spaces/ggml-org/gguf-my-lora) space.
+            Refer to the [original adapter repository](https://huggingface.co/{peft_model_id}) for more details.
+            
+            ## Use with llama.cpp
+            
+            ```bash
+            # with cli
+            llama-cli -m base_model.gguf --lora {gguf_output_name} (...other args)
+
+            # with server
+            llama-server -m base_model.gguf --lora {gguf_output_name} (...other args)
+            ```
+
+            To know more about LoRA usage with llama.cpp server, refer to the [llama.cpp server documentation](https://github.com/ggerganov/llama.cpp/blob/master/examples/server/README.md).
+            """
+        )
+        card.save(f"README.md")
+
+        api.upload_file(
+            path_or_fileobj=f"README.md",
+            path_in_repo=f"README.md",
+            repo_id=new_repo_id,
+        )
+
+        return (
+            f'<h1>✅ DONE</h1><br/><br/>Find your repo here: <a href="{new_repo_url}" target="_blank" style="text-decoration:underline">{new_repo_id}</a>'
+        )
+    except Exception as e:
+        return (f"<h1>❌ ERROR</h1><br/><br/>{e}")
+
+
+css="""/* Custom CSS to allow scrolling */
+.gradio-container {overflow-y: auto;}
+"""
+# Create Gradio interface
+with gr.Blocks(css=css) as demo: 
+    gr.Markdown("You must be logged in to use GGUF-my-lora.")
+    gr.LoginButton(min_width=250)
+
+    peft_model_id = HuggingfaceHubSearch(
+        label="PEFT LoRA repository",
+        placeholder="Search for repository on Huggingface",
+        search_type="model",
+    )
+
+    q_method = gr.Dropdown(
+        ["F32", "F16", "Q8_0"],
+        label="Quantization Method",
+        info="(Note: Quantization less than Q8 produces very poor results)",
+        value="F16",
+        filterable=False,
+        visible=True
+    )
+
+    private_repo = gr.Checkbox(
+        value=False,
+        label="Private Repo",
+        info="Create a private repo under your username."
+    )
+
+    iface = gr.Interface(
+        fn=process_model,
+        inputs=[
+            peft_model_id,
+            q_method,
+            private_repo,
+        ],
+        outputs=[
+            gr.Markdown(label="output"),
+        ],
+        title="Convert PEFT LoRA adapters to GGUF, blazingly fast ⚡!",
+        description="The space takes a PEFT LoRA (stored on a HF repo) as an input, converts it to GGUF and creates a Public repo under your HF user namespace.",
+        api_name=False
+    )
+
+
+def restart_space():
+    HfApi().restart_space(repo_id="ggml-org/gguf-my-lora", token=HF_TOKEN, factory_reboot=True)
+
+scheduler = BackgroundScheduler()
+scheduler.add_job(restart_space, "interval", seconds=21600)
+scheduler.start()
+
+# Launch the interface
+demo.queue(default_concurrency_limit=1, max_size=5).launch(debug=True, show_api=False)

docker-compose.yml

@@ -0,0 +1,16 @@
+# Docker compose file to LOCAL development
+
+services:
+  gguf-my-lora:
+    build:
+      context: .
+      dockerfile: Dockerfile
+    image: gguf-my-lora
+    container_name: gguf-my-lora
+    ports:
+      - "7860:7860"
+    volumes:
+      - .:/home/user/app
+    environment:
+      - RUN_LOCALLY=1
+      - HF_TOKEN=${HF_TOKEN}

start.sh

@@ -0,0 +1,10 @@
+#!/bin/bash
+
+if [ ! -d "llama.cpp" ]; then
+  # only run in dev env
+  # TODO: revert once the PR is merged
+  # git clone https://github.com/ggerganov/llama.cpp --depth 1
+  git clone https://github.com/ngxson/llama.cpp -b xsn/lora_convert_base_is_optional --depth 1
+fi
+
+python app.py