mlx-examples/llms/mixtral/convert.py

153 lines
4.3 KiB
Python
Raw Normal View History

2023-12-12 23:44:23 +08:00
# Copyright © 2023 Apple Inc.
import argparse
import copy
2023-12-15 07:30:32 +08:00
import glob
import json
import shutil
2023-12-12 23:44:23 +08:00
from pathlib import Path
import mlx.core as mx
import mlx.nn as nn
import numpy as np
2023-12-12 23:44:23 +08:00
import torch
from mixtral import Mixtral, ModelArgs
from mlx.utils import tree_flatten, tree_map, tree_unflatten
2023-12-12 23:44:23 +08:00
def convert(weights, config):
def convert_single(k, v):
v = v.to(torch.float16).numpy()
if "block_sparse_moe" not in k:
return [(k, v)]
if "gate" in k:
return [(k.replace("block_sparse_moe", "feed_forward"), v)]
2023-12-15 07:30:32 +08:00
# From: layers.N.block_sparse_moe.w
# To: layers.N.experts.M.w
num_experts = config["moe"]["num_experts"]
key_path = k.split(".")
v = np.split(v, num_experts, axis=0)
if key_path[-1] == "w2":
v = [u.T for u in v]
2023-12-15 07:30:32 +08:00
w_name = key_path.pop()
key_path[-1] = "feed_forward.experts"
return [
(".".join(key_path + [str(e), w_name, "weight"]), u)
for e, u in enumerate(v)
]
state = torch.load(tf)
weights = {}
for k, v in state.items():
weights.update(convert_single(k, v))
return weights
def quantize(weights, config, args):
quantized_config = copy.deepcopy(config)
# Load the model and update with the subset of weights:
config.pop("quantization", None)
model = Mixtral(ModelArgs(**config))
all_weights = dict(tree_flatten(model.parameters()))
weights = tree_map(mx.array, weights)
all_weights.update(weights)
all_weights = tree_unflatten(list(all_weights.items()))
model.update(all_weights)
# Quantize the model:
nn.QuantizedLinear.quantize_module(
model,
args.q_group_size,
args.q_bits,
# TODO: Quantize gate matrices when < 32 tiles supported
linear_class_predicate=lambda m: isinstance(m, nn.Linear)
and m.weight.shape[0] != 8,
)
# Extract the subset of quantized weights:
all_weights = dict(tree_flatten(model.parameters()))
quantized_weights = {}
for k, v in all_weights.items():
if k not in weights:
continue
quantized_weights[k] = v
prefix = k.split(".")[:-1]
for qw in ["scales", "biases"]:
if (k := ".".join(prefix + [qw])) in all_weights:
quantized_weights[k] = all_weights[k]
# Update the config:
quantized_config["quantization"] = {
"group_size": args.q_group_size,
"bits": args.q_bits,
}
return quantized_weights, quantized_config
2023-12-15 07:30:32 +08:00
2023-12-12 23:44:23 +08:00
if __name__ == "__main__":
parser = argparse.ArgumentParser(description="Convert Mixtral weights to MLX.")
parser.add_argument(
"--torch-path",
type=str,
default="Mixtral-8x7B-v0.1",
help="The path to the PyTorch model.",
)
parser.add_argument(
"--mlx-path",
2023-12-12 23:44:23 +08:00
type=str,
default="mlx_model",
help="The path to save the MLX model.",
)
parser.add_argument(
"-q",
"--quantize",
help="Generate a quantized model.",
action="store_true",
)
parser.add_argument(
"--q_group_size",
help="Group size for quantization.",
type=int,
default=64,
)
parser.add_argument(
"--q_bits",
help="Bits per weight for quantization.",
type=int,
default=4,
2023-12-12 23:44:23 +08:00
)
args = parser.parse_args()
torch_path = Path(args.torch_path)
mlx_path = Path(args.mlx_path)
mlx_path.mkdir(parents=True, exist_ok=True)
2023-12-15 07:30:32 +08:00
with open("params.json") as fid:
config = json.load(fid)
2023-12-15 07:30:32 +08:00
# Copy tokenizer
shutil.copyfile(
str(torch_path / "tokenizer.model"),
str(mlx_path / "tokenizer.model"),
)
# Convert and save model in shards
torch_files = glob.glob(str(torch_path / "consolidated.*.pt"))
2023-12-15 07:30:32 +08:00
torch_files = sorted(torch_files, key=lambda tf: int(tf.split(".")[-2]))
for e, tf in enumerate(torch_files):
print(f"[INFO] Converting file {e + 1}/{len(torch_files)}")
weights = convert(tf, config)
if args.quantize:
print("[INFO] Quantizing")
weights, config = quantize(weights, config, args)
np.savez(str(mlx_path / f"weights.{e}.npz"), **weights)
# Save updated config
with open(mlx_path / "config.json", "w") as f:
config["model_type"] = "mixtral"
json.dump(config, f, indent=4)