Fabio
c36510c21e
Some checks failed
Linux Build / Linux cpu (push) Has been cancelled
Linux Build / Linux cpu-clang (push) Has been cancelled
Linux Build / Linux vulkan (push) Has been cancelled
mac-build / macOS CPU (push) Has been cancelled
Nix Build / Nix (cpu) (push) Has been cancelled
Nix Build / Nix (vulkan) (push) Has been cancelled
windows-build / Windows CPU (push) Has been cancelled
120 lines
3.8 KiB
Python
120 lines
3.8 KiB
Python
#!/usr/bin/env python3
|
|
"""Build a self-contained Maya1 GGUF from the original checkpoints."""
|
|
|
|
from __future__ import annotations
|
|
|
|
import argparse
|
|
import json
|
|
import os
|
|
import shutil
|
|
import subprocess
|
|
import tempfile
|
|
from pathlib import Path
|
|
|
|
import torch
|
|
from safetensors.torch import save_file
|
|
|
|
|
|
REPO = Path(__file__).resolve().parents[2]
|
|
CONVERTER = REPO / "build/debug/bin/audiocpp_gguf"
|
|
SPEC = REPO / "model_specs/maya1.json"
|
|
|
|
|
|
def main() -> None:
|
|
parser = argparse.ArgumentParser(description=__doc__)
|
|
parser.add_argument("--model-dir", type=Path, required=True)
|
|
parser.add_argument("--codec-dir", type=Path, required=True)
|
|
parser.add_argument("--output-dir", type=Path, required=True)
|
|
args = parser.parse_args()
|
|
|
|
model_dir = args.model_dir.resolve()
|
|
codec_dir = args.codec_dir.resolve()
|
|
output_dir = args.output_dir.resolve()
|
|
output_dir.mkdir(parents=True, exist_ok=True)
|
|
subprocess.run(
|
|
[
|
|
"cmake",
|
|
"--build",
|
|
str(REPO / "build/debug"),
|
|
"--target",
|
|
"audiocpp_gguf",
|
|
"-j",
|
|
str(os.cpu_count() or 8),
|
|
],
|
|
check=True,
|
|
)
|
|
|
|
with tempfile.TemporaryDirectory(prefix="maya1-convert-") as temporary:
|
|
staging = Path(temporary)
|
|
for name in (
|
|
"config.json",
|
|
"generation_config.json",
|
|
"tokenizer.json",
|
|
"tokenizer_config.json",
|
|
):
|
|
shutil.copyfile(model_dir / name, staging / name)
|
|
|
|
model_index = json.loads(
|
|
(model_dir / "model.safetensors.index.json").read_text(encoding="utf-8")
|
|
)
|
|
weight_map = dict(model_index["weight_map"])
|
|
for shard in set(weight_map.values()):
|
|
(staging / shard).symlink_to(model_dir / shard)
|
|
|
|
codec_state = torch.load(
|
|
codec_dir / "pytorch_model.bin", map_location="cpu", weights_only=True
|
|
)
|
|
codec_state = {
|
|
"codec." + name: tensor.contiguous()
|
|
for name, tensor in codec_state.items()
|
|
if name.startswith(("quantizer.quantizers.", "decoder."))
|
|
}
|
|
codec_shard = "codec.safetensors"
|
|
save_file(codec_state, staging / codec_shard)
|
|
weight_map.update({name: codec_shard for name in codec_state})
|
|
(staging / "model.safetensors.index.json").write_text(
|
|
json.dumps({"weight_map": weight_map}, sort_keys=True),
|
|
encoding="utf-8",
|
|
)
|
|
|
|
for storage_type, filename in (
|
|
("native", "maya1-orig.gguf"),
|
|
("q8_0", "maya1-q8_0.gguf"),
|
|
):
|
|
descriptor, temporary_output = tempfile.mkstemp(
|
|
prefix="maya1-", suffix=".gguf", dir=output_dir
|
|
)
|
|
os.close(descriptor)
|
|
temporary_output = Path(temporary_output)
|
|
try:
|
|
subprocess.run(
|
|
[
|
|
str(CONVERTER),
|
|
"--input",
|
|
str(staging / "model.safetensors.index.json"),
|
|
"--root",
|
|
str(staging),
|
|
"--output",
|
|
str(temporary_output),
|
|
"--type",
|
|
storage_type,
|
|
"--family",
|
|
"maya1",
|
|
"--model-spec",
|
|
str(SPEC),
|
|
"--overwrite",
|
|
],
|
|
check=True,
|
|
)
|
|
output = output_dir / filename
|
|
os.replace(temporary_output, output)
|
|
subprocess.run(
|
|
[str(CONVERTER), "--inspect", str(output)], check=True
|
|
)
|
|
print(output)
|
|
finally:
|
|
temporary_output.unlink(missing_ok=True)
|
|
|
|
|
|
if __name__ == "__main__":
|
|
main()
|