AI-Hypercomputer / AI-Hypercomputer/maxtext
DEFAULT_MASK_VALUE causes gradient explosion and nan loss on deep models
- Ngôn ngữ chính
- Python
- Star
- 2.4k
- Fork
- 607
- Merge trung bình
- 2 ngày 19 giờ
- Pull request đã merge (30 ngày)
- 158
Mô tả
I was training a llama model on GPU, with a custom embedding. It worked fine with 12 layers, dim 1024, seq length 256, but loss would become nan after the first step if setting num_layers to more than 17. I debugged the gradients, and found after each layer their magnitude would increase by around 100x, until they hit float32_max at around the 18th layer and became inf, leading to nan loss.
The gradient explosion seemed to be coming from
`local_exps = jnp.exp(attn_weights - local_max)`
in attentions.py.
Changing
`DEFAULT_MASK_VALUE = -0.7 * float(jnp.finfo(jnp.dtype("float32")).max)`
to
`DEFAULT_MASK_VALUE = -jnp.inf`
fixed the issue, and the gradients' magnitude stopped increasing after each level.
Presumably the issue wasn't noticed during TPU training as that uses a separate codepath.
Hướng dẫn đóng góp
Đánh giá
Issue này chưa được đánh giá.