- config.py: action_dim默认值 8→18(对齐bot4cj 18列) - mt_parser.py: 轻量级.mt @Action解析器(支持-占位符前向填充) - pretrain_mt.py: 预训练入口(零图像占位,checkpoint兼容现有推理) - _run.sh: 新增pretrain子命令 - README.md: 新增预训练文档 + action_dim说明 - test_actpred.py: 测试断言更新为18
92 lines
2.5 KiB
Python
92 lines
2.5 KiB
Python
"""轻量级 .mt 脚本 @Action 矩阵解析器
|
|
|
|
从 .mt 文件中提取 @Action 块,解析为 (n_rows, n_cols) 的 float32 矩阵。
|
|
支持 `-` 占位符(沿用上一行同列的值)。
|
|
"""
|
|
|
|
import re
|
|
from pathlib import Path
|
|
|
|
import numpy as np
|
|
|
|
|
|
def _natural_key(p: Path):
|
|
"""自然排序键,确保 s1.mt < s2.mt < s10.mt"""
|
|
return [int(t) if t.isdigit() else t for t in re.split(r"(\d+)", p.name)]
|
|
|
|
|
|
def parse_mt_action(path) -> np.ndarray:
|
|
"""解析 .mt 文件中的 @Action 矩阵
|
|
|
|
Args:
|
|
path: .mt 文件路径
|
|
|
|
Returns:
|
|
(n_data_rows, n_cols) float32 矩阵,表头行已跳过,`-` 已前向填充
|
|
|
|
Raises:
|
|
ValueError: 文件中无 @Action 块或数据行不足
|
|
"""
|
|
text = Path(path).read_text(encoding="utf-8")
|
|
# 先去除整行注释(# 开头),避免注释中的 @Action 干扰正则匹配
|
|
text = re.sub(r"(?m)^[ \t]*#.*$", "", text)
|
|
|
|
match = re.search(r"@Action\s+(.*?)\s*@", text, re.DOTALL)
|
|
if not match:
|
|
raise ValueError(f"未找到 @Action 块: {path}")
|
|
|
|
block = match.group(1)
|
|
|
|
rows = []
|
|
for line in block.split("\n"):
|
|
line = re.sub(r"#.*$", "", line).strip()
|
|
if not line:
|
|
continue
|
|
if line.endswith(";"):
|
|
line = line[:-1].strip()
|
|
if not line:
|
|
continue
|
|
cols = re.split(r"\s+", line)
|
|
rows.append(cols)
|
|
|
|
if len(rows) < 2:
|
|
raise ValueError(
|
|
f"@Action 至少需要表头 + 1 行数据,实际 {len(rows)} 行: {path}"
|
|
)
|
|
|
|
# 第一行是表头(列名如 s1.A, s2.A, ...),跳过
|
|
data_rows = rows[1:]
|
|
n_cols = len(rows[0])
|
|
|
|
matrix = np.zeros((len(data_rows), n_cols), dtype=np.float32)
|
|
for i, row in enumerate(data_rows):
|
|
for j, val in enumerate(row):
|
|
if val == "-":
|
|
matrix[i, j] = matrix[i - 1, j] if i > 0 else 0.0
|
|
else:
|
|
matrix[i, j] = float(val)
|
|
|
|
return matrix
|
|
|
|
|
|
def parse_mt_directory(mt_dir, pattern="*.mt"):
|
|
"""解析目录下所有 .mt 文件的 @Action 矩阵
|
|
|
|
Args:
|
|
mt_dir: .mt 文件目录
|
|
pattern: 文件匹配模式
|
|
|
|
Returns:
|
|
list of (filepath, np.ndarray) 元组,按文件名自然排序
|
|
"""
|
|
dir_path = Path(mt_dir)
|
|
files = sorted(dir_path.glob(pattern), key=_natural_key)
|
|
results = []
|
|
for f in files:
|
|
try:
|
|
matrix = parse_mt_action(f)
|
|
results.append((str(f), matrix))
|
|
except ValueError:
|
|
pass
|
|
return results
|