Spaces:
Running on Zero
Running on Zero
File size: 7,172 Bytes
5f7afef | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 | from __future__ import annotations
import math
from typing import Any, Tuple, Optional
try:
import torch
HAS_TORCH = True
except ImportError:
HAS_TORCH = False
try:
import numpy as np
HAS_NUMPY = True
except ImportError:
HAS_NUMPY = False
def summarize_obj(obj: Any, max_len: int = 60) -> str:
"""Generate a clean, informative one-line summary of a tensor, array, or object."""
if HAS_TORCH and isinstance(obj, torch.Tensor):
return f"Tensor(shape={list(obj.shape)}, dtype={str(obj.dtype).replace('torch.', '')})"
if HAS_NUMPY and isinstance(obj, np.ndarray):
return f"ndarray(shape={list(obj.shape)}, dtype={obj.dtype})"
if isinstance(obj, (list, tuple)):
items_preview = ", ".join(summarize_obj(x, 20) for x in obj[:3])
if len(obj) > 3:
items_preview += f", ... (+{len(obj)-3} more)"
bracket = ("(", ")") if isinstance(obj, tuple) else ("[", "]")
return f"{bracket[0]}{items_preview}{bracket[1]}"
s = str(obj)
if len(s) > max_len:
return s[:max_len] + "..."
return s
def compare_outputs(
actual: Any,
expected: Any,
atol: float = 1e-4,
rtol: float = 1e-4,
) -> Tuple[bool, Optional[str], Optional[float]]:
"""
Compare actual output against expected output with numerical tolerance.
Returns:
(is_match, error_message, max_diff)
"""
# 1. Exact identity or both None
if actual is expected:
return True, None, 0.0
# 2. PyTorch Tensor comparison
if HAS_TORCH and (isinstance(actual, torch.Tensor) or isinstance(expected, torch.Tensor)):
if not isinstance(actual, torch.Tensor):
return False, f"Type mismatch: expected torch.Tensor, got {type(actual).__name__}", None
if not isinstance(expected, torch.Tensor):
return False, f"Type mismatch: expected {type(expected).__name__}, got torch.Tensor", None
# Check shape
if actual.shape != expected.shape:
return False, f"Shape mismatch: expected {list(expected.shape)}, got {list(actual.shape)}", None
# Check dtype
if actual.dtype != expected.dtype:
return False, f"Dtype mismatch: expected {expected.dtype}, got {actual.dtype}", None
act = actual.detach().cpu().float()
exp = expected.detach().cpu().float()
# Check for NaN / Inf
if torch.isnan(act).any() and not torch.isnan(exp).any():
return False, "Output contains unexpected NaN values", None
if torch.isinf(act).any() and not torch.isinf(exp).any():
return False, "Output contains unexpected Inf values", None
# Filter out identical infinite values (e.g. -inf and -inf) to avoid (-inf) - (-inf) => NaN
matching_neginf = torch.isneginf(act) & torch.isneginf(exp)
matching_posinf = torch.isposinf(act) & torch.isposinf(exp)
matching_inf = matching_neginf | matching_posinf
matching_nan = torch.isnan(act) & torch.isnan(exp)
ignore_mask = matching_inf | matching_nan
if ignore_mask.all():
max_diff = 0.0
else:
act_diff = torch.where(ignore_mask, torch.zeros_like(act), act)
exp_diff = torch.where(ignore_mask, torch.zeros_like(exp), exp)
diff = torch.abs(act_diff - exp_diff)
finite_diff = diff[~torch.isnan(diff) & ~torch.isinf(diff)]
max_diff = float(finite_diff.max().item()) if finite_diff.numel() > 0 else 0.0
if torch.allclose(act, exp, rtol=rtol, atol=atol, equal_nan=True):
return True, None, max_diff
else:
return (
False,
f"Numerical mismatch: max absolute diff {max_diff:.6e} > tolerance (atol={atol}, rtol={rtol})",
max_diff,
)
# 3. NumPy ndarray comparison
if HAS_NUMPY and (isinstance(actual, np.ndarray) or isinstance(expected, np.ndarray)):
if not isinstance(actual, np.ndarray):
return False, f"Type mismatch: expected np.ndarray, got {type(actual).__name__}", None
if not isinstance(expected, np.ndarray):
return False, f"Type mismatch: expected {type(expected).__name__}, got np.ndarray", None
if actual.shape != expected.shape:
return False, f"Shape mismatch: expected {list(expected.shape)}, got {list(actual.shape)}", None
if actual.dtype != expected.dtype:
return False, f"Dtype mismatch: expected {expected.dtype}, got {actual.dtype}", None
act = actual.astype(np.float64)
exp = expected.astype(np.float64)
if np.isnan(act).any() and not np.isnan(exp).any():
return False, "Output contains unexpected NaN values", None
if np.isinf(act).any() and not np.isinf(exp).any():
return False, "Output contains unexpected Inf values", None
matching_inf = np.isinf(act) & np.isinf(exp) & (np.sign(act) == np.sign(exp))
matching_nan = np.isnan(act) & np.isnan(exp)
ignore_mask = matching_inf | matching_nan
if ignore_mask.all():
max_diff = 0.0
else:
act_diff = np.where(ignore_mask, 0.0, act)
exp_diff = np.where(ignore_mask, 0.0, exp)
diff = np.abs(act_diff - exp_diff)
finite_diff = diff[~np.isnan(diff) & ~np.isinf(diff)]
max_diff = float(np.max(finite_diff)) if finite_diff.size > 0 else 0.0
if np.allclose(act, exp, rtol=rtol, atol=atol, equal_nan=True):
return True, None, max_diff
else:
return (
False,
f"Numerical mismatch: max absolute diff {max_diff:.6e} > tolerance (atol={atol}, rtol={rtol})",
max_diff,
)
# 4. Sequences (tuple / list)
if isinstance(expected, (tuple, list)):
if not isinstance(actual, (tuple, list)):
return False, f"Type mismatch: expected {type(expected).__name__}, got {type(actual).__name__}", None
if len(actual) != len(expected):
return False, f"Length mismatch: expected {len(expected)} elements, got {len(actual)}", None
max_sub_diff = 0.0
for i, (act_elem, exp_elem) in enumerate(zip(actual, expected)):
match, err, sub_diff = compare_outputs(act_elem, exp_elem, atol=atol, rtol=rtol)
if sub_diff is not None:
max_sub_diff = max(max_sub_diff, sub_diff)
if not match:
return False, f"Element [{i}] mismatch: {err}", max_sub_diff
return True, None, max_sub_diff
# 5. Scalar numbers (float, int)
if isinstance(expected, (float, int)) and isinstance(actual, (float, int)):
diff = abs(float(actual) - float(expected))
allowed = atol + rtol * abs(float(expected))
if diff <= allowed:
return True, None, diff
return False, f"Value mismatch: expected {expected}, got {actual} (diff {diff:.6e} > {allowed:.6e})", diff
# 6. Fallback exact equality
if actual == expected:
return True, None, 0.0
return False, f"Expected {expected!r}, but got {actual!r}", None
|