scripts/tests/modal_volume.py

import modal
app = modal.App("example-app-with-volumes")


#Download checkpoints
model_cache = modal.Volume.from_name("checkpoints-cache", create_if_missing=True)
download_image = (
    modal.Image.debian_slim(python_version="3.11")
    .pip_install("huggingface_hub[hf_transfer]==0.26.2")
    .env({"HF_HUB_ENABLE_HF_TRANSFER": "1"})
)

@app.function(image=download_image, volumes={cache_dir: model_cache}, timeout=30 * MINUTES)
def download_model(repo_id, allow_patterns, revision: Optional[str] = None):
    from huggingface_hub import snapshot_download

    print(f"🦙 downloading model from {repo_id} if not present")
    snapshot_download(
        repo_id=repo_id,
        revision=revision,
        local_dir=cache_dir,
        allow_patterns=allow_patterns,
    )
    model_cache.commit()  # ensure other Modal Functions can see our writes before we quit
    print("🦙 model loaded")


@app.local_entrypoint()
def main(video: Optional[str] = None, audio: str = "DeepSeek-R1"): 
    """Run llama.cpp inference on Modal for phi-4 or deepseek r1."""
    import shlex
    org_name = "unsloth"
    # two sample models: the diminutive phi-4 and the chonky deepseek r1
    if model.lower() == "phi-4":
        model_name = "phi-4-GGUF"
        quant = "Q2_K"
        model_entrypoint_file = f"phi-4-{quant}.gguf"
        model_pattern = f"*{quant}*"
        revision = None
        parsed_args = DEFAULT_PHI_ARGS if args is None else shlex.split(args)
    elif model.lower() == "deepseek-r1":
        model_name = "DeepSeek-R1-GGUF"
        quant = "UD-IQ1_S"
        model_entrypoint_file = (
            f"{model}-{quant}/DeepSeek-R1-{quant}-00001-of-00003.gguf"
        )
        model_pattern = f"*{quant}*"
        revision = "02656f62d2aa9da4d3f0cdb34c341d30dd87c3b6"
        parsed_args = DEFAULT_DEEPSEEK_R1_ARGS if args is None else shlex.split(args)
    else:
        raise ValueError(f"Unknown model {model}")

    repo_id = f"{org_name}/{model_name}"
    download_model.remote(repo_id, [model_pattern], revision)

    # call out to a `.remote` Function on Modal for inference
    result = llama_cpp_inference.remote(
        model_entrypoint_file,
        prompt,
        n_predict,
        parsed_args,
        store_output=model.lower() == "deepseek-r1",
    )
    output_path = Path("/tmp") / f"llama-cpp-{model}.txt"
    output_path.parent.mkdir(parents=True, exist_ok=True)
    print(f"🦙 writing response to {output_path}")
    output_path.write_text(result)