OpenTransformer
/

AGILLM4-diffusionblocks

diffusionblocks

memory-efficient-training

block-wise-training

Model card Files Files and versions

AGILLM4-diffusionblocks

202 kB

Ctrl+K

Ctrl+K

3 contributors

History: 18 commits

Scott/Codex

Tune DBlock backward-math speed line

9c90574 7 days ago

.gitattributes

1.52 kB
initial commit 8 days ago
.gitignore

23 Bytes
Add improved sublinear anchor coverage 7 days ago
README.md

8.82 kB
Tune DBlock backward-math speed line 7 days ago
dblocks_agillm4.py

4.66 kB
AGILLM4-DiffusionBlocks: block-wise AR+SAT+NAT denoising, fused CE, tied heads 8 days ago
dblocks_agillm4_lm.py

6.35 kB
AGILLM4-DiffusionBlocks: block-wise AR+SAT+NAT denoising, fused CE, tied heads 8 days ago
dblocks_train.py

16.9 kB
Tune DBlock backward-math speed line 7 days ago
fused_ce.py

2.44 kB
Upgrade VRAM-first DiffusionBlocks trainer 8 days ago
nB300_agillm4_vram_dblock.py

147 kB
Tune DBlock backward-math speed line 7 days ago
relaunch_agillm4_dblock.sh

3.73 kB
Add stochastic sparse DBlock speed profile 7 days ago
relaunch_agillm4_dblock_sg2.sh

1.93 kB
Tune DBlock backward-math speed line 7 days ago
relaunch_agillm4_dblock_tied.sh

3.5 kB
Add stochastic sparse DBlock speed profile 7 days ago
sublinear_improved.py

3.08 kB
Add sublinear attention v2 improvements 7 days ago
sublinear_improved_snippet.py

1.55 kB
Add sublinear attention v2 improvements 7 days ago