[Refactor] Remove dead code multiple places (#58002)

Signed-off-by: yewentao256 <zhyanwentao@126.com>
This commit is contained in:
Wentao Ye
2026-09-22 16:36:03 -04:00
committed by GitHub
parent 276db94fb2
commit fae5cbfc8b
12 changed files with 3 additions and 585 deletions
@@ -12,7 +12,6 @@ import os
import tempfile
import threading
import time
from collections import defaultdict
from collections.abc import Callable, Generator, Iterable
from contextlib import contextmanager
from pathlib import Path
@@ -23,7 +22,7 @@ import huggingface_hub.constants
import numpy as np
import regex as re
import torch
from safetensors.torch import load, load_file, safe_open, save_file
from safetensors.torch import load, load_file, safe_open
from tqdm.auto import tqdm
from transformers.utils import SAFE_WEIGHTS_INDEX_NAME
@@ -186,56 +185,6 @@ def maybe_download_from_modelscope(
return None
def _shared_pointers(tensors):
ptrs = defaultdict(list)
for k, v in tensors.items():
ptrs[v.data_ptr()].append(k)
failing = []
for _, names in ptrs.items():
if len(names) > 1:
failing.append(names)
return failing
def convert_bin_to_safetensor_file(
pt_filename: str,
sf_filename: str,
) -> None:
loaded = torch.load(pt_filename, map_location="cpu", weights_only=True)
if "state_dict" in loaded:
loaded = loaded["state_dict"]
shared = _shared_pointers(loaded)
for shared_weights in shared:
for name in shared_weights[1:]:
loaded.pop(name)
# For tensors to be contiguous
loaded = {k: v.contiguous() for k, v in loaded.items()}
dirname = os.path.dirname(sf_filename)
os.makedirs(dirname, exist_ok=True)
save_file(loaded, sf_filename, metadata={"format": "pt"})
# check file size
sf_size = os.stat(sf_filename).st_size
pt_size = os.stat(pt_filename).st_size
if (sf_size - pt_size) / pt_size > 0.01:
raise RuntimeError(
f"""The file size different is more than 1%:
- {sf_filename}: {sf_size}
- {pt_filename}: {pt_size}
"""
)
# check if the tensors are the same
reloaded = load_file(sf_filename)
for k in loaded:
pt_tensor = loaded[k]
sf_tensor = reloaded[k]
if not torch.equal(pt_tensor, sf_tensor):
raise RuntimeError(f"The output tensors do not match for key {k}")
# TODO(woosuk): Move this to other place.
def get_quant_config(
model_config: ModelConfig, load_config: LoadConfig
@@ -1313,29 +1262,6 @@ def composed_weight_loader(
return composed_loader
def initialize_dummy_weights(
model: torch.nn.Module,
model_config: ModelConfig,
low: float = -1e-3,
high: float = 1e-3,
seed: int = 1234,
) -> None:
"""Initialize model weights with random values.
The model weights must be randomly initialized for accurate performance
measurements. Additionally, the model weights should not cause NaNs in the
forward pass. We empirically found that initializing the weights with
values between -1e-3 and 1e-3 works well for most models.
We use per-parameter random seed, so that dummy weights are consistent,
even if the model is partitioned across multiple devices. When the seed
is fixed, the random values generated by this function only depends on
the parameter's number of elements and its data type.
"""
for param in model.state_dict().values():
initialize_single_dummy_weight(param, low, high, seed)
@torch.no_grad()
def initialize_single_dummy_weight(
param: torch.Tensor,