| .. |
|
1cycle_lr.png
|
b84a1fa410
Web edits (#147)
|
4 years ago |
|
3d-parallelism.png
|
65c2f974d8
Pipeline parallel training engine. (#392)
|
4 years ago |
|
DeepSpeed-vs-Megatron.png
|
f2ac7eafd5
ZeRO-2 (#217)
|
4 years ago |
|
adam-convergence.png
|
093f09ff27
Update documentation for 1-bit Adam (#388)
|
4 years ago |
|
bert-ib.png
|
093f09ff27
Update documentation for 1-bit Adam (#388)
|
4 years ago |
|
bert-large-training-time.png
|
5fb22a055a
Ported BERT pre-training tutorial (#184)
|
4 years ago |
|
bert-scaling.png
|
093f09ff27
Update documentation for 1-bit Adam (#388)
|
4 years ago |
|
bert-tcp.png
|
093f09ff27
Update documentation for 1-bit Adam (#388)
|
4 years ago |
|
convergence-table.png
|
093f09ff27
Update documentation for 1-bit Adam (#388)
|
4 years ago |
|
deepspeed-speedup.png
|
f2ac7eafd5
ZeRO-2 (#217)
|
4 years ago |
|
end-to-end-bert-training.PNG
|
734d8991c8
Transformer kernel release (#242)
|
4 years ago |
|
layernorm_animation.gif
|
734d8991c8
Transformer kernel release (#242)
|
4 years ago |
|
layernorm_deepspeed.gif
|
734d8991c8
Transformer kernel release (#242)
|
4 years ago |
|
layernorm_ds.png
|
734d8991c8
Transformer kernel release (#242)
|
4 years ago |
|
layernorm_pytorch.gif
|
734d8991c8
Transformer kernel release (#242)
|
4 years ago |
|
layernorm_torch.png
|
734d8991c8
Transformer kernel release (#242)
|
4 years ago |
|
loss_and_lr.png
|
b84a1fa410
Web edits (#147)
|
4 years ago |
|
lr_schedule.png
|
b84a1fa410
Web edits (#147)
|
4 years ago |
|
megatron-gpt2-perf-test.png
|
b84a1fa410
Web edits (#147)
|
4 years ago |
|
model_convergence.png
|
b84a1fa410
Web edits (#147)
|
4 years ago |
|
onebit-adam-overview.png
|
093f09ff27
Update documentation for 1-bit Adam (#388)
|
4 years ago |
|
onebit-convergence.png
|
093f09ff27
Update documentation for 1-bit Adam (#388)
|
4 years ago |
|
oom_dp8_1.5B_log.png
|
2dea61f285
ZeRO tutorials (#384)
|
4 years ago |
|
pipe-schedule.png
|
65c2f974d8
Pipeline parallel training engine. (#392)
|
4 years ago |
|
pp-lowbw-gpt2.png
|
a8a8b3d288
Landing page updates (#395)
|
4 years ago |
|
qkv_fusion.png
|
734d8991c8
Transformer kernel release (#242)
|
4 years ago |
|
sa_backward_pass.png
|
e5bbc2e559
Sparse attn + ops/runtime refactor + v0.3.0 (#343)
|
4 years ago |
|
sa_bert_base_time_result.png
|
e5bbc2e559
Sparse attn + ops/runtime refactor + v0.3.0 (#343)
|
4 years ago |
|
sa_bert_large_time_result.png
|
e5bbc2e559
Sparse attn + ops/runtime refactor + v0.3.0 (#343)
|
4 years ago |
|
sa_fixed_sparsity_structure.png
|
e5bbc2e559
Sparse attn + ops/runtime refactor + v0.3.0 (#343)
|
4 years ago |
|
sa_forward_pass.png
|
e5bbc2e559
Sparse attn + ops/runtime refactor + v0.3.0 (#343)
|
4 years ago |
|
sa_gpt2_time_result.png
|
e5bbc2e559
Sparse attn + ops/runtime refactor + v0.3.0 (#343)
|
4 years ago |
|
sa_long_document_comprehension_result.png
|
e5bbc2e559
Sparse attn + ops/runtime refactor + v0.3.0 (#343)
|
4 years ago |
|
sa_maximum_sequence_runnable_on_bert.png
|
e5bbc2e559
Sparse attn + ops/runtime refactor + v0.3.0 (#343)
|
4 years ago |
|
sa_variable_sparsity_structure.png
|
e5bbc2e559
Sparse attn + ops/runtime refactor + v0.3.0 (#343)
|
4 years ago |
|
softmax_animation.gif
|
734d8991c8
Transformer kernel release (#242)
|
4 years ago |
|
softmax_deepspeed.gif
|
734d8991c8
Transformer kernel release (#242)
|
4 years ago |
|
softmax_ds.png
|
734d8991c8
Transformer kernel release (#242)
|
4 years ago |
|
softmax_pytorch.gif
|
734d8991c8
Transformer kernel release (#242)
|
4 years ago |
|
softmax_torch.png
|
734d8991c8
Transformer kernel release (#242)
|
4 years ago |
|
squad-ib.png
|
093f09ff27
Update documentation for 1-bit Adam (#388)
|
4 years ago |
|
squad-scaling.png
|
093f09ff27
Update documentation for 1-bit Adam (#388)
|
4 years ago |
|
squad-tcp.png
|
093f09ff27
Update documentation for 1-bit Adam (#388)
|
4 years ago |
|
transformer_kernel_perf.png
|
734d8991c8
Transformer kernel release (#242)
|
4 years ago |
|
transformer_kernel_perf_seq128.PNG
|
734d8991c8
Transformer kernel release (#242)
|
4 years ago |
|
transformer_kernel_perf_seq512.PNG
|
734d8991c8
Transformer kernel release (#242)
|
4 years ago |
|
transformer_preln_arch.png
|
2e6d93e0e5
new transformer pre-ln image (#268)
|
4 years ago |
|
variable_sparsity_pattern.png
|
e5bbc2e559
Sparse attn + ops/runtime refactor + v0.3.0 (#343)
|
4 years ago |
|
webinar-aug2020.png
|
7ae8f8bc9b
DeepSpeed webinar announcement (#301)
|
4 years ago |
|
zero-full.png
|
f2ac7eafd5
ZeRO-2 (#217)
|
4 years ago |
|
zero1_dp8_1.5B_log.png
|
2dea61f285
ZeRO tutorials (#384)
|
4 years ago |
|
zero1_dp8_1.5B_smi.png
|
2dea61f285
ZeRO tutorials (#384)
|
4 years ago |
|
zero2_dp32_10B_log.png
|
2dea61f285
ZeRO tutorials (#384)
|
4 years ago |
|
zero2_dp32_10B_smi.png
|
2dea61f285
ZeRO tutorials (#384)
|
4 years ago |
|
zero_offload_dp1_10B_cpu.png
|
2dea61f285
ZeRO tutorials (#384)
|
4 years ago |
|
zero_offload_dp1_10B_log.png
|
2dea61f285
ZeRO tutorials (#384)
|
4 years ago |
|
zero_offload_dp1_10B_smi.png
|
2dea61f285
ZeRO tutorials (#384)
|
4 years ago |
Jeff Rasley
