import os import shutil import subprocess import gradio as gr from huggingface_hub import create_repo, HfApi from huggingface_hub import snapshot_download from huggingface_hub import whoami from huggingface_hub import ModelCard from gradio_huggingfacehub_search import HuggingfaceHubSearch from apscheduler.schedulers.background import BackgroundScheduler from textwrap import dedent LLAMA_LIKE_ARCHS = ["MistralForCausalLM",] HF_TOKEN = os.environ.get("HF_TOKEN") def script_to_use(model_id, api): info = api.model_info(model_id) if info.config is None: return None arch = info.config.get("architectures", None) if arch is None: return None arch = arch[0] return "convert.py" if arch in LLAMA_LIKE_ARCHS else "convert-hf-to-gguf.py" def process_model(model_id, q_method, private_repo, oauth_token: gr.OAuthToken | None): if oauth_token.token is None: raise ValueError("You must be logged in to use GGUF-my-repo") model_name = model_id.split('/')[-1] fp16 = f"{model_name}/{model_name.lower()}.fp16.bin" 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=model_id, recursive=True, ) ) else "*.bin" ) dl_pattern += pattern api.snapshot_download(repo_id=model_id, local_dir=model_name, local_dir_use_symlinks=False, allow_patterns=dl_pattern) print("Model downloaded successully!") conversion_script = script_to_use(model_id, api) fp16_conversion = f"python llama.cpp/{conversion_script} {model_name} --outtype f16 --outfile {fp16}" result = subprocess.run(fp16_conversion, shell=True, capture_output=True) print(result) if result.returncode != 0: raise Exception(f"Error converting to fp16: {result.stderr}") print("Model converted to fp16 successully!") qtype = f"{model_name}/{model_name.lower()}.{q_method.upper()}.gguf" quantise_ggml = f"./llama.cpp/quantize {fp16} {qtype} {q_method}" result = subprocess.run(quantise_ggml, shell=True, capture_output=True) if result.returncode != 0: raise Exception(f"Error quantizing: {result.stderr}") print("Quantised successfully!") # Create empty repo new_repo_url = api.create_repo(repo_id=f"{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) try: card = ModelCard.load(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-repo") card.text = dedent( f""" # {new_repo_id} This model was converted to GGUF format from [`{model_id}`](https://huggingface.co/{model_id}) using llama.cpp via the ggml.ai's [GGUF-my-repo](https://huggingface.co/spaces/ggml-org/gguf-my-repo) space. Refer to the [original model card](https://huggingface.co/{model_id}) for more details on the model. ## Use with llama.cpp Install llama.cpp through brew. ```bash brew install ggerganov/ggerganov/llama.cpp ``` Invoke the llama.cpp server or the CLI. CLI: ```bash llama-cli --hf-repo {new_repo_id} --model {qtype.split("/")[-1]} -p "The meaning to life and the universe is" ``` Server: ```bash llama-server --hf-repo {new_repo_id} --model {qtype.split("/")[-1]} -c 2048 ``` Note: You can also use this checkpoint directly through the [usage steps](https://github.com/ggerganov/llama.cpp?tab=readme-ov-file#usage) listed in the Llama.cpp repo as well. ``` git clone https://github.com/ggerganov/llama.cpp && \ cd llama.cpp && \ make && \ ./main -m {qtype.split("/")[-1]} -n 128 ``` """ ) card.save(os.path.join(model_name, "README-new.md")) api.upload_file( path_or_fileobj=qtype, path_in_repo=qtype.split("/")[-1], repo_id=new_repo_id, ) api.upload_file( path_or_fileobj=f"{model_name}/README-new.md", path_in_repo="README.md", repo_id=new_repo_id, ) print("Uploaded successfully!") return ( f'Find your repo here', "llama.png", ) except Exception as e: return (f"Error: {e}", "error.png") finally: shutil.rmtree(model_name, ignore_errors=True) print("Folder cleaned up successfully!") # Create Gradio interface iface = gr.Interface( fn=process_model, inputs=[ HuggingfaceHubSearch( label="Hub Model ID", placeholder="Search for model id on Huggingface", search_type="model", ), gr.Dropdown( ["Q2_K", "Q3_K_S", "Q3_K_M", "Q3_K_L", "Q4_0", "Q4_K_S", "Q4_K_M", "Q5_0", "Q5_K_S", "Q5_K_M", "Q6_K", "Q8_0"], label="Quantization Method", info="GGML quantisation type", value="Q4_K_M", filterable=False ), gr.Checkbox( value=False, label="Private Repo", info="Create a private repo under your username." ), ], outputs=[ gr.Markdown(label="output"), gr.Image(show_label=False), ], title="Create your own GGUF Quants, blazingly fast ⚡!", description="The space takes an HF repo as an input, quantises it and creates a Public repo containing the selected quant under your HF user namespace.", ) with gr.Blocks() as demo: gr.Markdown("You must be logged in to use GGUF-my-repo.") gr.LoginButton(min_width=250) iface.render() def restart_space(): HfApi().restart_space(repo_id="ggml-org/gguf-my-repo", 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)