用 Codex 或 Claude 帮你安装 复制这段 Prompt,粘贴到 Codex、Claude 或其他助手里,让它检查 Skill 页面并帮你完成安装。
直接命令不会经过审查 Prompt;运行前请先检查来源。
npx skills add https://github.com/cxcscmu/SkillLearnBench --skill fixed-tensor-testing命令会保持在同一行。复制前请横向滚动并检查完整内容。
想先保存到本地?可下载 SkillsMP 当前能够提供的文件。
Handles reading, populating, and saving .docx files using the python-docx library. Use this skill for any tasks involving template filling or modifying Word documents.
Perform various data analysis on SEC 13-F and obtain some insights of fund activities such as number of holdings, AUM, and change of holdings between two quarters.
This skill includes search capability in 13F, such as fuzzy search a fund information using possibly inaccurate name, or fuzzy search a stock cusip info using its name.
基于 SOC 职业分类
正在显示 SKILL.md
| name | fixed-tensor-testing |
| description | Test ML functions with fixed input tensors for reproducibility. |
Fixed tensor testing ensures deterministic, reproducible results for loss functions and model outputs. Enables verification without training dependencies.
import torch
import numpy as np
# Set all random seeds
torch.manual_seed(42)
np.random.seed(42)
if torch.cuda.is_available():
torch.cuda.manual_seed_all(42)
# Fixed log probabilities (typically negative)
log_probs = torch.randn(batch_size, seq_len)
# Ensure reasonable log prob range (e.g., -2 to 0)
log_probs = torch.clamp(log_probs, min=-5.0, max=0.0)
# Fixed sequence lengths
seq_lengths = torch.randint(10, 100, (batch_size,))
# Fixed input IDs (for masking)
input_ids = torch.randint(0, vocab_size, (batch_size, seq_len))
batch_size = 4
seq_len = 10
# Generate fixed tensors
torch.manual_seed(123)
log_probs = torch.randn(batch_size, seq_len)
log_probs = torch.clamp(log_probs, -5.0, 0.0)
# Create mask (e.g., padding)
mask = torch.ones(batch_size, seq_len, dtype=torch.bool)
mask[:, seq_len-2:] = False # Last 2 tokens are padding
# Lengths accounting for mask
seq_lengths = mask.sum(dim=1).float()
import numpy as np
results = {
'losses': loss.cpu().detach().numpy(),
'log_probs': log_probs.cpu().detach().numpy(),
}
np.savez_compressed('/path/to/results.npz', **results)
data = np.load('/path/to/results.npz')
losses = data['losses']
print(f"Shape: {losses.shape}, dtype: {losses.dtype}")
# Loss should be finite and positive
assert torch.isfinite(loss).all(), "Loss contains NaN or Inf"
assert loss.item() > 0, "Loss should be positive"
# Shape validation
assert loss.shape == expected_shape, f"Shape mismatch: {loss.shape}"
# Range checks
assert loss.item() < 100, "Loss unreasonably large"
assert log_probs.min() >= -6.0, "Log probs too small"
# Run computation twice, should get same result
loss1 = compute_loss(fixed_tensors)
loss2 = compute_loss(fixed_tensors)
assert torch.allclose(loss1, loss2), "Loss not reproducible"
def test_simpo_loss():
# Setup fixed inputs
batch_size = 8
torch.manual_seed(42)
log_probs = torch.randn(batch_size, 20)
seq_lengths = torch.full((batch_size,), 20.0)
# Compute loss
beta = 2.0
gamma = 1.0
loss = compute_simpo_loss(log_probs, seq_lengths, beta, gamma)
# Assertions
assert loss.shape == torch.Size([])
assert torch.isfinite(loss)
assert loss.item() > 0
return loss.item()
# Ensure gradients can backpropagate
x = torch.randn(5, 10, requires_grad=True)
loss = some_loss_function(x)
loss.backward()
assert x.grad is not None
assert not torch.allclose(x.grad, torch.zeros_like(x.grad))
# Test with extreme values
extreme_inputs = [
torch.full((5,), -100.0), # Very negative log probs
torch.full((5,), 0.0), # Zero log probs
torch.zeros(5), # All zeros
]
for inp in extreme_inputs:
try:
loss = compute_loss(inp)
assert torch.isfinite(loss), f"Loss not finite for {inp}"
except Exception as e:
print(f"Failed with input {inp}: {e}")
# Print intermediate values
def debug_loss(log_probs, seq_lengths, beta, gamma):
rewards = beta * log_probs / seq_lengths.unsqueeze(1)
print(f"Rewards shape: {rewards.shape}, min: {rewards.min():.4f}, max: {rewards.max():.4f}")
# ... rest of computation
print(f"Final loss: {loss.item():.6f}")
return loss