GPT2からDecoder作成(causal_lmを使わないで学習)

[abePclWaseda]

What?

Why?

See also

[abePclWaseda]

38エポックまでで学習をとめて、40エポックまでで再開する予定

[abePclWaseda]

RESULTS

Environments

• date: Wed Nov 6 23:10:00 JST 2024
• python version: 3.8.19 (default, Mar 20 2024, 19:58:24) [GCC 11.2.0]
• espnet version: espnet 202402
• pytorch version: pytorch 1.13.1
• Git hash: 558de86a97881efd98962ca0ff20994a14b513b1
  • Commit date: Tue Nov 5 00:41:48 2024 +0900

exp/asr_train_asr_conformer_lr2e-3_warmup15k_amp_nondeterministic_raw_en_hugging_face_openai-community-gpt2_sp

WER

dataset	Snt	Wrd	Corr	Sub	Del	Ins	Err	S.Err
decode_asr_asr_model_valid.acc.ave/dev_clean	2703	54402	94.1	5.5	0.4	0.9	6.8	57.5
decode_asr_asr_model_valid.acc.ave/dev_other	2864	50948	84.8	13.8	1.4	2.1	17.4	80.1
decode_asr_asr_model_valid.acc.ave/test_clean	2620	52576	93.9	5.6	0.5	0.9	7.0	58.6
decode_asr_asr_model_valid.acc.ave/test_other	2939	52343	84.2	14.2	1.6	2.0	17.8	81.8

CER

dataset	Snt	Wrd	Corr	Sub	Del	Ins	Err	S.Err
decode_asr_asr_model_valid.acc.ave/dev_clean	2703	288456	98.1	1.2	0.8	0.8	2.7	57.5
decode_asr_asr_model_valid.acc.ave/dev_other	2864	265951	93.4	4.1	2.5	2.0	8.6	80.1
decode_asr_asr_model_valid.acc.ave/test_clean	2620	281530	98.0	1.1	0.8	0.7	2.7	58.6
decode_asr_asr_model_valid.acc.ave/test_other	2939	272758	93.3	4.1	2.7	1.9	8.7	81.8

TER

dataset	Snt	Wrd	Corr	Sub	Del	Ins	Err	S.Err

[abePclWaseda]

RESULTS

Environments

• date: Fri Nov 15 12:34:35 JST 2024
• python version: 3.8.19 (default, Mar 20 2024, 19:58:24) [GCC 11.2.0]
• espnet version: espnet 202402
• pytorch version: pytorch 1.13.1
• Git hash: 1a377ac094329466b2dd6c907c2e4f742bc4dc50
  • Commit date: Wed Nov 13 16:10:47 2024 +0900

exp/asr_train_asr_conformer_lr2e-3_warmup15k_amp_nondeterministic_raw_en_hugging_face_openai-community-gpt2_sp

WER

dataset	Snt	Wrd	Corr	Sub	Del	Ins	Err	S.Err
decode_asr_asr_model_valid.acc.ave/dev_clean	2703	54402	94.3	5.3	0.4	0.9	6.6	56.6
decode_asr_asr_model_valid.acc.ave/dev_other	2864	50948	85.1	13.5	1.4	2.0	16.9	79.7
decode_asr_asr_model_valid.acc.ave/test_clean	2620	52576	94.0	5.5	0.5	0.9	6.9	58.3
decode_asr_asr_model_valid.acc.ave/test_other	2939	52343	84.5	13.9	1.6	1.9	17.4	81.3
decode_asr_asr_model_valid.acc_1110.ave/dev_clean	2703	54402	94.1	5.5	0.4	0.9	6.8	57.5
decode_asr_asr_model_valid.acc_1110.ave/dev_other	2864	50948	84.8	13.8	1.4	2.1	17.4	80.1
decode_asr_asr_model_valid.acc_1110.ave/test_clean	2620	52576	93.9	5.6	0.5	0.9	7.0	58.6
decode_asr_asr_model_valid.acc_1110.ave/test_other	2939	52343	84.2	14.2	1.6	2.0	17.8	81.8
decode_asr_asr_model_valid.acc_1111Merge.ave/dev_clean	2703	54402	91.3	8.0	0.8	1.6	10.4	67.3
decode_asr_asr_model_valid.acc_1111Merge.ave/dev_other	2864	50948	81.2	16.7	2.2	2.6	21.4	84.8
decode_asr_asr_model_valid.acc_1111Merge.ave/test_clean	2620	52576	90.9	8.2	0.9	1.7	10.8	68.5
decode_asr_asr_model_valid.acc_1111Merge.ave/test_other	2939	52343	80.7	17.0	2.2	2.5	21.7	85.2

CER

dataset	Snt	Wrd	Corr	Sub	Del	Ins	Err	S.Err
decode_asr_asr_model_valid.acc.ave/dev_clean	2703	288456	98.1	1.2	0.7	0.7	2.6	56.6
decode_asr_asr_model_valid.acc.ave/dev_other	2864	265951	93.6	4.0	2.4	2.0	8.4	79.7
decode_asr_asr_model_valid.acc.ave/test_clean	2620	281530	98.1	1.1	0.8	0.7	2.7	58.3
decode_asr_asr_model_valid.acc.ave/test_other	2939	272758	93.4	4.0	2.6	1.9	8.5	81.3
decode_asr_asr_model_valid.acc_1110.ave/dev_clean	2703	288456	98.1	1.2	0.8	0.8	2.7	57.5
decode_asr_asr_model_valid.acc_1110.ave/dev_other	2864	265951	93.4	4.1	2.5	2.0	8.6	80.1
decode_asr_asr_model_valid.acc_1110.ave/test_clean	2620	281530	98.0	1.1	0.8	0.7	2.7	58.6
decode_asr_asr_model_valid.acc_1110.ave/test_other	2939	272758	93.3	4.1	2.7	1.9	8.7	81.8
decode_asr_asr_model_valid.acc_1111Merge.ave/dev_clean	2703	288456	97.4	1.5	1.1	1.9	4.5	67.3
decode_asr_asr_model_valid.acc_1111Merge.ave/dev_other	2864	265951	92.1	4.7	3.2	2.9	10.8	84.8
decode_asr_asr_model_valid.acc_1111Merge.ave/test_clean	2620	281530	97.3	1.5	1.2	2.0	4.7	68.5
decode_asr_asr_model_valid.acc_1111Merge.ave/test_other	2939	272758	92.2	4.6	3.2	2.8	10.6	85.2

TER

dataset	Snt	Wrd	Corr	Sub	Del	Ins	Err	S.Err

[abePclWaseda]

[abePclWaseda]

[abePclWaseda]

[abePclWaseda]

RESULTS

Environments

• date: Fri Nov 29 11:26:26 JST 2024
• python version: 3.8.19 (default, Mar 20 2024, 19:58:24) [GCC 11.2.0]
• espnet version: espnet 202402
• pytorch version: pytorch 1.13.1
• Git hash: 191ef1137aecb736adc31d4d3eb16302ee0e2284
  • Commit date: Fri Nov 29 07:33:58 2024 +0900

exp/asr_train_asr_conformer_lr2e-3_warmup15k_amp_nondeterministic_raw_en_hugging_face_openai-community-gpt2_sp

WER

dataset	Snt	Wrd	Corr	Sub	Del	Ins	Err	S.Err
decode_asr_asr_model_valid.acc.ave/dev_clean	2703	54402	94.2	5.4	0.4	1.0	6.7	56.8
decode_asr_asr_model_valid.acc.ave/dev_other	2864	50948	84.7	13.9	1.4	2.2	17.5	78.9
decode_asr_asr_model_valid.acc.ave/test_clean	2620	52576	94.0	5.6	0.5	0.9	6.9	58.1
decode_asr_asr_model_valid.acc.ave/test_other	2939	52343	84.3	14.2	1.5	2.1	17.8	80.9

CER

dataset	Snt	Wrd	Corr	Sub	Del	Ins	Err	S.Err
decode_asr_asr_model_valid.acc.ave/dev_clean	2703	288456	98.1	1.1	0.7	0.8	2.6	56.8
decode_asr_asr_model_valid.acc.ave/dev_other	2864	265951	93.3	4.2	2.4	2.1	8.7	78.9
decode_asr_asr_model_valid.acc.ave/test_clean	2620	281530	98.0	1.2	0.8	0.7	2.7	58.1
decode_asr_asr_model_valid.acc.ave/test_other	2939	272758	93.4	4.1	2.5	2.0	8.7	80.9

TER

dataset	Snt	Wrd	Corr	Sub	Del	Ins	Err	S.Err

[abePclWaseda]

デコーダフリーズさせたほう

RESULTS

Environments

• date: Sat Nov 30 00:31:37 JST 2024
• python version: 3.8.19 (default, Mar 20 2024, 19:58:24) [GCC 11.2.0]
• espnet version: espnet 202402
• pytorch version: pytorch 1.13.1
• Git hash: 82a49ad240f0872e4dd1b308dffa7d9dbed2484d
  • Commit date: Fri Nov 29 21:15:17 2024 +0900

exp/asr_train_asr_conformer_lr2e-3_warmup15k_amp_nondeterministic_raw_en_hugging_face_openai-community-gpt2_sp

WER

dataset	Snt	Wrd	Corr	Sub	Del	Ins	Err	S.Err
decode_asr_asr_model_valid.acc.ave/dev_clean	2703	54402	94.6	5.1	0.4	0.8	6.3	55.1
decode_asr_asr_model_valid.acc.ave/dev_other	2864	50948	85.0	13.7	1.4	2.0	17.1	78.9
decode_asr_asr_model_valid.acc.ave/test_clean	2620	52576	94.4	5.2	0.4	0.9	6.5	56.3
decode_asr_asr_model_valid.acc.ave/test_other	2939	52343	84.5	13.9	1.6	2.0	17.5	80.4

CER

dataset	Snt	Wrd	Corr	Sub	Del	Ins	Err	S.Err
decode_asr_asr_model_valid.acc.ave/dev_clean	2703	288456	98.2	1.1	0.7	0.7	2.5	55.1
decode_asr_asr_model_valid.acc.ave/dev_other	2864	265951	93.6	4.0	2.3	1.9	8.3	78.9
decode_asr_asr_model_valid.acc.ave/test_clean	2620	281530	98.2	1.1	0.7	0.7	2.4	56.3
decode_asr_asr_model_valid.acc.ave/test_other	2939	272758	93.4	4.0	2.5	1.9	8.4	80.4

TER

dataset	Snt	Wrd	Corr	Sub	Del	Ins	Err	S.Err