Eval harness #212

TevenLeScao merged 51 commits into main from eval_harness
DanielHesslow
DanielHesslow Add functionality for running the evaluation harness on single gpu
71728c2f
DanielHesslow Add support for pipelining
97a2339c
DanielHesslow support tensor parallel
e72a18f5
DanielHesslow save the results
0a829651
DanielHesslow Minor cleanup
ceddfc5b
DanielHesslow
DanielHesslow DanielHesslow marked this pull request as ready for review 4 years ago
DanielHesslow DanielHesslow marked this pull request as draft 4 years ago
DanielHesslow Experimental Deepspeed support
c1e80225
DanielHesslow Proper deepspeed integration, now working on combined tp and pp
0f8c8c09
DanielHesslow Update model loading and clean up code.
37e6962e
DanielHesslow Add some options
ee0a1a90
DanielHesslow DanielHesslow marked this pull request as ready for review 4 years ago
DanielHesslow
DanielHesslow DanielHesslow assigned zphang zphang 4 years ago
DanielHesslow DanielHesslow assigned ibeltagy ibeltagy 4 years ago
DanielHesslow DanielHesslow unassigned ibeltagy ibeltagy 4 years ago
DanielHesslow DanielHesslow requested a review from StellaAthena StellaAthena 4 years ago
DanielHesslow DanielHesslow requested a review from zphang zphang 4 years ago
DanielHesslow Fix pipelining + fp32 evaluaiton.
cacc58f8
DanielHesslow DanielHesslow unassigned zphang zphang 4 years ago
thomasw21
thomasw21 commented on 2021-12-09
DanielHesslow Remove dummy paths in examples/run_evalharness.sh
778f2514
DanielHesslow Simplify offline loading with export HF_DATASETS_OFFLINE=1
3d90b18d
thomasw21
thomasw21 commented on 2021-12-09
SaulLu
SaulLu commented on 2021-12-14
DanielHesslow Remove accidental copy-paste.
2bb61ac7
thomasw21
thomasw21 commented on 2021-12-22
DanielHesslow Experimantel deepspeed evaluation-path
a362da38
stas00
stas00
thomasw21
DanielHesslow
stas00
DanielHesslow
stas00
stas00 commented on 2022-01-08
stas00 make it work with deepspeed; add instructions
9899be0b
stas00
stas00
stas00 improve
7ef5ba75
stas00 make adaptive_seq_len work with deepspeed
9527ad36
stas00
stas00 move to slurm
d4dacbe9
stas00 fixes
151e91ab
stas00 cleanup
92123d0a
thomasw21
thomasw21 commented on 2022-01-09
stas00 add instructions on how to import data into the spreadsheet
a6fab1f8
stas00 not tracking ppl/em
dedf111e
stas00 add task version
a12af5c0
stas00
stas00 make compatible with lm-eval@master
dbecf81c
stas00
stas00 switch to 16gb slurm; simplify; improve instructions
c04e3d06
DanielHesslow Deepspeed model loading hack
e6e4800f
DanielHesslow Restore correct zero state.
5e611bf1
stas00
stas00 commented on 2022-01-14
stas00 fix conversion script
7937eab0
stas00 simpler config
afd38142
stas00 Merge remote-tracking branch 'origin/main' into eval_harness
b1a54f33
stas00 corrections
9c60079f
stas00 add logiqa
d8611374
stas00 dealing with custom tokenizers
71587901
stas00 fix
f0da71de
conglongli
conglongli commented on 2022-02-18
stas00
stas00 commented on 2022-02-18
stas00 Update examples/run_evalharness_deepspeed.md
1e06f41e
stas00
stas00 Merge branch 'main' into eval_harness
a9221acc
stas00 Merge remote-tracking branch 'origin/main' into eval_harness
a7222590
stas00 check that the checkpoint path is valid
9ac9fadb
stas00 skip --abort_on_unmet_fused_kernel_constraints during eval
8ef90184
stas00 disable sanity check on layers-2%pp==0
a798d69f
stas00 sort skip_keys
5884dcf3
DanielHesslow DanielHesslow requested a review 3 years ago
TevenLeScao
TevenLeScao approved these changes on 2022-04-26
TevenLeScao TevenLeScao requested a review from TevenLeScao TevenLeScao 3 years ago
thomasw21
thomasw21 approved these changes on 2022-04-27
stas00 make the default path unique to avoid overwrite
45bd9c65
Muennighoff Add bootstrap_iters arg
f75e2326
Muennighoff
stas00
Muennighoff Explain bootstrap_iters flag
7bf75b94
Muennighoff Intermediate results flag
3f18e7be
stas00
stas00 commented on 2022-05-14
Muennighoff Add backup file
213317f1
Muennighoff Add arg to reduce bubble for pipeline parallel
1c11b107
Muennighoff Fix adaptive_seq_len via resetting activation shape
f3307058
Muennighoff Extract args.load prior to load_ds_checkpoint_and_setup_megatron
50820352
Muennighoff Parse args prior to loading function to get load_path
db203cce
Muennighoff Add run_evalharness-tr11-176b-ml slurm script
1d6c630a
stas00
Muennighoff
thomasw21
DanielHesslow
TevenLeScao
TevenLeScao TevenLeScao merged 3ab0ad18 into main 3 years ago
TevenLeScao TevenLeScao deleted the eval_harness branch 3 years ago

Login to write a write a comment.

Login via GitHub

Assignees
No one assigned
Labels
Milestone