diff --git a/users/zeyer/experiments/exp2024_04_23_baselines/ctc.py b/users/zeyer/experiments/exp2024_04_23_baselines/ctc.py index b837ef74f..7aa9ab13a 100644 --- a/users/zeyer/experiments/exp2024_04_23_baselines/ctc.py +++ b/users/zeyer/experiments/exp2024_04_23_baselines/ctc.py @@ -790,7 +790,7 @@ def py(): # (Baseline without lpNormedGrad: 5.77/6.03) # Log prob normed gradient (lpNormedGrad) (incl blank) with blank separated (blankSep) - train_exp( + train_exp( # 5.73/6.08 "v6-relPosAttDef-aedLoss-bhv20-11gb-f32-bs15k-accgrad1-mgpu4-pavg100-wd1e_2-lrlin1e_5_295k-featBN" "-speedpertV2-spm10k-bpeSample001-blankSep-lpNormedGradInclBlank", config_11gb_v6_f32_accgrad1_mgpu4_pavg100_wd1e_4, diff --git a/users/zeyer/experiments/exp2024_09_16_grad_align.py b/users/zeyer/experiments/exp2024_09_16_grad_align.py index 012af73d4..4ad27dc79 100644 --- a/users/zeyer/experiments/exp2024_09_16_grad_align.py +++ b/users/zeyer/experiments/exp2024_09_16_grad_align.py @@ -143,14 +143,14 @@ def py(): # "-featBN-speedpertV2-bpe10k-bpeSample001", # "bpe10k", # ), - # ( # ctc forced align: 84.9/64.2ms - # "base-bpe10k-blankSep", # 5.98/6.13 - # "v6-relPosAttDef" - # "-aedLoss-bhv20-11gb-f32-bs15k-accgrad1-mgpu4-pavg100-wd1e_2-lrlin1e_5_295k" - # "-featBN-speedpertV2-bpe10k-bpeSample001" - # "-blankSep", - # "bpe10k", - # ), + ( # ctc forced align: 84.9/64.2ms + "base-bpe10k-blankSep", # 5.98/6.13 + "v6-relPosAttDef" + "-aedLoss-bhv20-11gb-f32-bs15k-accgrad1-mgpu4-pavg100-wd1e_2-lrlin1e_5_295k" + "-featBN-speedpertV2-bpe10k-bpeSample001" + "-blankSep", + "bpe10k", + ), # ( # "ebranchformer", # 5.54/5.69 # # output/ctc/v6-EBranchformer-relPosAttDef-noBias-aedLoss-bhv20-11gb-f32-bs15k-accgrad1-mgpu4-pavg100-wd1e_2-lrlin1e_5_295k-featBN-speedpertV2-spm10k-bpeSample001/recog_results_best