958 kB

Ctrl+K

1 contributor

Reductions refuse what they used to accept: GPU row-max non-finite guard, averageGrads length/NaN refusal, int32 accumulator bound, capacity-weighted cluster loss, finite reduced-gradient check, empty CUDA_VISIBLE_DEVICES

6e87247 verified 10 days ago

dashboard
Refuse NaN/Inf at every int8 quantize; requant rounds half-up (retrained); bounded units; README for September 14 days ago
verified
Refuse NaN/Inf at every int8 quantize; requant rounds half-up (retrained); bounded units; README for September 14 days ago
__init__.py

595 Bytes
Old-hardware training through emulated GPU logic 3 months ago
cluster.py

9.98 kB
Reductions refuse what they used to accept: GPU row-max non-finite guard, averageGrads length/NaN refusal, int32 accumulator bound, capacity-weighted cluster loss, finite reduced-gradient check, empty CUDA_VISIBLE_DEVICES 10 days ago
example_task.py

920 Bytes
Old-hardware training through emulated GPU logic 3 months ago
spikewhale_panel.py

11.3 kB
Refuse NaN/Inf at every int8 quantize; requant rounds half-up (retrained); bounded units; README for September 14 days ago
spikewhale_task.py

5.88 kB
Refuse NaN/Inf at every int8 quantize; requant rounds half-up (retrained); bounded units; README for September 14 days ago
task.py

1.26 kB
Old-hardware training through emulated GPU logic 3 months ago
train.py

3.28 kB
Old-hardware training through emulated GPU logic 3 months ago
verified_task.py

1.34 kB
Old-hardware training through emulated GPU logic 3 months ago