Last active
August 14, 2023 23:07
-
-
Save TimDettmers/385014b37f998c7857b15a3ea60b4cae to your computer and use it in GitHub Desktop.
Chip2 data evaluated with Rouge L and BERTScore for 4, 8, 16 bit LoRA and full finetuning
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
| Config: max_steps: 17500, lora_r: 64 , lr: 0.0002, bf16: False, lora_modules: all , bits: 4 , full_finetune: False, lora_dropout: 0.0 , warmup_steps: 100 , compress_statistics: True, dataset: NaN , gradient_accumulation_steps: NaN , learning_rate: NaN , quant_type: fp4 , adam_beta2: 0.999, update_freq: 6 | |
| eval_bert_f1 mean (SE): 64.8716 (21.8331). 95% CI (22.079, 107.664). Sample size: 4 | |
| eval_bert_f1 mean (SE): 64.8716 (21.8331). 95% CI (22.079, 107.664). Sample size: 4 | |
| eval_rougeL mean (SE): 33.1083 (19.1162). 95% CI (-4.359, 70.576). Sample size: 4 | |
| ================================================================================ | |
| Config: max_steps: 17500, lora_r: 64 , lr: 0.0002, bf16: False, lora_modules: all , bits: 4 , full_finetune: False, lora_dropout: 0.0 , warmup_steps: 100 , compress_statistics: False, dataset: NaN , gradient_accumulation_steps: NaN , learning_rate: NaN , quant_type: fp4 , adam_beta2: 0.999, update_freq: 6 | |
| eval_bert_f1 mean (SE): 67.0044 (22.3593). 95% CI (23.180, 110.829). Sample size: 4 | |
| eval_bert_f1 mean (SE): 67.0044 (22.3593). 95% CI (23.180, 110.829). Sample size: 4 | |
| eval_rougeL mean (SE): 45.2738 (15.7187). 95% CI (14.465, 76.082). Sample size: 4 | |
| ================================================================================ | |
| Config: max_steps: 17500, lora_r: 64 , lr: 0.0002, bf16: False, lora_modules: all , bits: 4 , full_finetune: False, lora_dropout: 0.0 , warmup_steps: 100 , compress_statistics: False, dataset: NaN , gradient_accumulation_steps: NaN , learning_rate: NaN , quant_type: nf4 , adam_beta2: 0.999, update_freq: 6 | |
| eval_bert_f1 mean (SE): 67.2176 (22.4145). 95% CI (23.285, 111.150). Sample size: 4 | |
| eval_bert_f1 mean (SE): 67.2176 (22.4145). 95% CI (23.285, 111.150). Sample size: 4 | |
| eval_rougeL mean (SE): 45.4117 (15.3900). 95% CI (15.247, 75.576). Sample size: 4 | |
| ================================================================================ | |
| Config: max_steps: 5000, lora_r: 64 , lr: 0.0002, bf16: True, lora_modules: all , bits: 4 , full_finetune: False, lora_dropout: 0.0 , warmup_steps: 100 , compress_statistics: False, dataset: NaN , gradient_accumulation_steps: NaN , learning_rate: NaN , quant_type: fp4 , adam_beta2: 0.999, update_freq: 6 | |
| eval_bert_f1 mean (SE): 82.5676 (4.6489). 95% CI (73.456, 91.679). Sample size: 4 | |
| eval_bert_f1 mean (SE): 82.5676 (4.6489). 95% CI (73.456, 91.679). Sample size: 4 | |
| eval_rougeL mean (SE): 32.4291 (18.7233). 95% CI (-4.268, 69.127). Sample size: 4 | |
| ================================================================================ | |
| Config: max_steps: 5000, lora_r: 64 , lr: 0.0002, bf16: True, lora_modules: all , bits: 4 , full_finetune: False, lora_dropout: 0.0 , warmup_steps: 100 , compress_statistics: False, dataset: NaN , gradient_accumulation_steps: NaN , learning_rate: NaN , quant_type: nf4 , adam_beta2: 0.999, update_freq: 6 | |
| eval_bert_f1 mean (SE): 83.0177 (4.3380). 95% CI (74.515, 91.520). Sample size: 4 | |
| eval_bert_f1 mean (SE): 83.0177 (4.3380). 95% CI (74.515, 91.520). Sample size: 4 | |
| eval_rougeL mean (SE): 35.2325 (17.1500). 95% CI (1.619, 68.846). Sample size: 4 | |
| ================================================================================ | |
| Config: max_steps: 17500, lora_r: 64 , lr: 0.0002, bf16: True, lora_modules: all , bits: 4 , full_finetune: False, lora_dropout: 0.0 , warmup_steps: 100 , compress_statistics: True, dataset: NaN , gradient_accumulation_steps: NaN , learning_rate: NaN , quant_type: nf4 , adam_beta2: 0.999, update_freq: 6 | |
| eval_bert_f1 mean (SE): 83.2916 (4.3216). 95% CI (74.821, 91.762). Sample size: 4 | |
| eval_bert_f1 mean (SE): 83.2916 (4.3216). 95% CI (74.821, 91.762). Sample size: 4 | |
| eval_rougeL mean (SE): 37.2928 (16.5428). 95% CI (4.869, 69.717). Sample size: 4 | |
| ================================================================================ | |
| Config: max_steps: 17500, lora_r: 64 , lr: 0.0002, bf16: False, lora_modules: all , bits: 4 , full_finetune: False, lora_dropout: 0.0 , warmup_steps: 100 , compress_statistics: True, dataset: NaN , gradient_accumulation_steps: NaN , learning_rate: NaN , quant_type: nf4 , adam_beta2: 0.999, update_freq: 6 | |
| eval_bert_f1 mean (SE): 83.5084 (3.7384). 95% CI (76.181, 90.836). Sample size: 4 | |
| eval_bert_f1 mean (SE): 83.5084 (3.7384). 95% CI (76.181, 90.836). Sample size: 4 | |
| eval_rougeL mean (SE): 32.2683 (18.0691). 95% CI (-3.147, 67.684). Sample size: 4 | |
| ================================================================================ | |
| Config: max_steps: 5000, lora_r: 64 , lr: 0.0002, bf16: True, lora_modules: all , bits: 4 , full_finetune: False, lora_dropout: 0.0 , warmup_steps: 100 , compress_statistics: True, dataset: NaN , gradient_accumulation_steps: NaN , learning_rate: NaN , quant_type: fp4 , adam_beta2: 0.999, update_freq: 6 | |
| eval_bert_f1 mean (SE): 84.7429 (4.9394). 95% CI (75.062, 94.424). Sample size: 4 | |
| eval_bert_f1 mean (SE): 84.7429 (4.9394). 95% CI (75.062, 94.424). Sample size: 4 | |
| eval_rougeL mean (SE): 45.2540 (15.4220). 95% CI (15.027, 75.481). Sample size: 4 | |
| ================================================================================ | |
| Config: max_steps: 4500, lora_r: 64 , lr: 0.0002, bf16: True, lora_modules: all , bits: 16 , full_finetune: False, lora_dropout: 0.0 , warmup_steps: 100 , compress_statistics: False, dataset: alpaca, gradient_accumulation_steps: NaN , learning_rate: NaN , quant_type: fp4 , adam_beta2: 0.999, update_freq: 6 | |
| eval_bert_f1 mean (SE): 85.2494 (nan). 95% CI (nan, nan). Sample size: 1 | |
| eval_bert_f1 mean (SE): 85.2494 (nan). 95% CI (nan, nan). Sample size: 1 | |
| eval_rougeL mean (SE): 36.5082 (nan). 95% CI (nan, nan). Sample size: 1 | |
| ================================================================================ | |
| Config: max_steps: 5000, lora_r: 64 , lr: 0.0002, bf16: True, lora_modules: all , bits: 4 , full_finetune: False, lora_dropout: 0.0 , warmup_steps: 100 , compress_statistics: True, dataset: NaN , gradient_accumulation_steps: NaN , learning_rate: NaN , quant_type: nf4 , adam_beta2: 0.999, update_freq: 6 | |
| eval_bert_f1 mean (SE): 85.5598 (3.7924). 95% CI (78.127, 92.993). Sample size: 4 | |
| eval_bert_f1 mean (SE): 85.5598 (3.7924). 95% CI (78.127, 92.993). Sample size: 4 | |
| eval_rougeL mean (SE): 44.9017 (14.4743). 95% CI (16.532, 73.271). Sample size: 4 | |
| ================================================================================ | |
| Config: max_steps: 4000, lora_r: 64 , lr: 0.0002, bf16: False, lora_modules: all , bits: 4 , full_finetune: False, lora_dropout: 0.0 , warmup_steps: 100 , compress_statistics: NaN , dataset: NaN , gradient_accumulation_steps: NaN , learning_rate: NaN , quant_type: NaN , adam_beta2: NaN , update_freq: 6 | |
| eval_bert_f1 mean (SE): 86.5179 (nan). 95% CI (nan, nan). Sample size: 1 | |
| eval_bert_f1 mean (SE): 86.5179 (nan). 95% CI (nan, nan). Sample size: 1 | |
| eval_rougeL mean (SE): 57.2240 (nan). 95% CI (nan, nan). Sample size: 1 | |
| ================================================================================ | |
| Config: max_steps: 5000, lora_r: 64 , lr: 0.0003, bf16: False, lora_modules: ffn , bits: 4 , full_finetune: False, lora_dropout: 0.05, warmup_steps: NaN , compress_statistics: NaN , dataset: NaN , gradient_accumulation_steps: NaN , learning_rate: NaN , quant_type: NaN , adam_beta2: NaN , update_freq: 4 | |
| eval_bert_f1 mean (SE): 87.2168 (0.0000). 95% CI (87.217, 87.217). Sample size: 2 | |
| eval_bert_f1 mean (SE): 87.2168 (0.0000). 95% CI (87.217, 87.217). Sample size: 2 | |
| eval_rougeL mean (SE): 57.1546 (0.0000). 95% CI (57.155, 57.155). Sample size: 2 | |
| ================================================================================ | |
| Config: max_steps: 5000, lora_r: 128 , lr: 0.0003, bf16: True, lora_modules: ffn , bits: 16 , full_finetune: True, lora_dropout: 0.0 , warmup_steps: NaN , compress_statistics: NaN , dataset: NaN , gradient_accumulation_steps: NaN , learning_rate: NaN , quant_type: NaN , adam_beta2: NaN , update_freq: 4 | |
| eval_bert_f1 mean (SE): 87.9641 (0.0648). 95% CI (87.837, 88.091). Sample size: 6 | |
| eval_bert_f1 mean (SE): 87.9641 (0.0648). 95% CI (87.837, 88.091). Sample size: 6 | |
| eval_rougeL mean (SE): 53.2109 (0.2020). 95% CI (52.815, 53.607). Sample size: 6 | |
| ================================================================================ | |
| Config: max_steps: 5000, lora_r: 128 , lr: 0.0003, bf16: True, lora_modules: ffn , bits: 16 , full_finetune: True, lora_dropout: 0.0 , warmup_steps: 500 , compress_statistics: NaN , dataset: NaN , gradient_accumulation_steps: NaN , learning_rate: NaN , quant_type: NaN , adam_beta2: NaN , update_freq: 4 | |
| eval_bert_f1 mean (SE): 88.2319 (0.0000). 95% CI (88.232, 88.232). Sample size: 2 | |
| eval_bert_f1 mean (SE): 88.2319 (0.0000). 95% CI (88.232, 88.232). Sample size: 2 | |
| eval_rougeL mean (SE): 53.1711 (0.0000). 95% CI (53.171, 53.171). Sample size: 2 | |
| ================================================================================ | |
| Config: max_steps: 5000, lora_r: 128 , lr: 0.0003, bf16: True, lora_modules: ffn , bits: 16 , full_finetune: True, lora_dropout: 0.0 , warmup_steps: 100 , compress_statistics: NaN , dataset: NaN , gradient_accumulation_steps: NaN , learning_rate: NaN , quant_type: NaN , adam_beta2: NaN , update_freq: 4 | |
| eval_bert_f1 mean (SE): 88.3811 (0.0000). 95% CI (88.381, 88.381). Sample size: 2 | |
| eval_bert_f1 mean (SE): 88.3811 (0.0000). 95% CI (88.381, 88.381). Sample size: 2 | |
| eval_rougeL mean (SE): 53.5289 (0.0000). 95% CI (53.529, 53.529). Sample size: 2 | |
| ================================================================================ | |
| Config: max_steps: 2500, lora_r: 128 , lr: 0.0003, bf16: True, lora_modules: ffn , bits: 16 , full_finetune: True, lora_dropout: 0.0 , warmup_steps: 100 , compress_statistics: NaN , dataset: NaN , gradient_accumulation_steps: NaN , learning_rate: NaN , quant_type: NaN , adam_beta2: NaN , update_freq: 8 | |
| eval_bert_f1 mean (SE): 88.6400 (0.0000). 95% CI (88.640, 88.640). Sample size: 2 | |
| eval_bert_f1 mean (SE): 88.6400 (0.0000). 95% CI (88.640, 88.640). Sample size: 2 | |
| eval_rougeL mean (SE): 54.6076 (0.0000). 95% CI (54.608, 54.608). Sample size: 2 | |
| ================================================================================ | |
| Config: max_steps: 17500, lora_r: 64 , lr: 0.0002, bf16: True, lora_modules: all , bits: 4 , full_finetune: False, lora_dropout: 0.0 , warmup_steps: 100 , compress_statistics: True, dataset: NaN , gradient_accumulation_steps: NaN , learning_rate: NaN , quant_type: fp4 , adam_beta2: 0.999, update_freq: 6 | |
| eval_bert_f1 mean (SE): 88.9557 (0.9238). 95% CI (87.145, 90.766). Sample size: 4 | |
| eval_bert_f1 mean (SE): 88.9557 (0.9238). 95% CI (87.145, 90.766). Sample size: 4 | |
| eval_rougeL mean (SE): 57.8024 (4.6491). 95% CI (48.690, 66.915). Sample size: 4 | |
| ================================================================================ | |
| Config: max_steps: 17500, lora_r: 64 , lr: 0.0002, bf16: True, lora_modules: all , bits: 4 , full_finetune: False, lora_dropout: 0.0 , warmup_steps: 100 , compress_statistics: False, dataset: NaN , gradient_accumulation_steps: NaN , learning_rate: NaN , quant_type: fp4 , adam_beta2: 0.999, update_freq: 6 | |
| eval_bert_f1 mean (SE): 89.0261 (0.8125). 95% CI (87.434, 90.619). Sample size: 4 | |
| eval_bert_f1 mean (SE): 89.0261 (0.8125). 95% CI (87.434, 90.619). Sample size: 4 | |
| eval_rougeL mean (SE): 58.6140 (3.3605). 95% CI (52.027, 65.201). Sample size: 4 | |
| ================================================================================ | |
| Config: max_steps: 5000, lora_r: 128 , lr: 0.001, bf16: False, lora_modules: all , bits: 4 , full_finetune: False, lora_dropout: 0.05, warmup_steps: NaN , compress_statistics: NaN , dataset: NaN , gradient_accumulation_steps: NaN , learning_rate: NaN , quant_type: NaN , adam_beta2: NaN , update_freq: 4 | |
| eval_bert_f1 mean (SE): 89.0370 (0.0000). 95% CI (89.037, 89.037). Sample size: 2 | |
| eval_bert_f1 mean (SE): 89.0370 (0.0000). 95% CI (89.037, 89.037). Sample size: 2 | |
| eval_rougeL mean (SE): 62.0205 (0.0000). 95% CI (62.021, 62.021). Sample size: 2 | |
| ================================================================================ | |
| Config: max_steps: 17500, lora_r: 64 , lr: 0.0002, bf16: True, lora_modules: all , bits: 4 , full_finetune: False, lora_dropout: 0.0 , warmup_steps: 100 , compress_statistics: False, dataset: NaN , gradient_accumulation_steps: NaN , learning_rate: NaN , quant_type: nf4 , adam_beta2: 0.999, update_freq: 6 | |
| eval_bert_f1 mean (SE): 89.1721 (0.9192). 95% CI (87.370, 90.974). Sample size: 4 | |
| eval_bert_f1 mean (SE): 89.1721 (0.9192). 95% CI (87.370, 90.974). Sample size: 4 | |
| eval_rougeL mean (SE): 59.7464 (3.4698). 95% CI (52.946, 66.547). Sample size: 4 | |
| ================================================================================ | |
| Config: max_steps: 6000, lora_r: 64 , lr: 0.0002, bf16: True, lora_modules: all , bits: 4 , full_finetune: False, lora_dropout: 0.0 , warmup_steps: 100 , compress_statistics: NaN , dataset: NaN , gradient_accumulation_steps: NaN , learning_rate: NaN , quant_type: NaN , adam_beta2: NaN , update_freq: 6 | |
| eval_bert_f1 mean (SE): 89.1988 (nan). 95% CI (nan, nan). Sample size: 1 | |
| eval_bert_f1 mean (SE): 89.1988 (nan). 95% CI (nan, nan). Sample size: 1 | |
| eval_rougeL mean (SE): 57.9034 (nan). 95% CI (nan, nan). Sample size: 1 | |
| ================================================================================ | |
| Config: max_steps: 5000, lora_r: 128 , lr: 3e-06, bf16: True, lora_modules: ffn , bits: 16 , full_finetune: True, lora_dropout: 0.0 , warmup_steps: 100 , compress_statistics: NaN , dataset: NaN , gradient_accumulation_steps: NaN , learning_rate: NaN , quant_type: NaN , adam_beta2: NaN , update_freq: 4 | |
| eval_bert_f1 mean (SE): 89.2989 (0.0000). 95% CI (89.299, 89.299). Sample size: 2 | |
| eval_bert_f1 mean (SE): 89.2989 (0.0000). 95% CI (89.299, 89.299). Sample size: 2 | |
| eval_rougeL mean (SE): 60.1265 (0.0000). 95% CI (60.126, 60.126). Sample size: 2 | |
| ================================================================================ | |
| Config: max_steps: 2500, lora_r: 128 , lr: 3e-06, bf16: True, lora_modules: ffn , bits: 16 , full_finetune: True, lora_dropout: 0.0 , warmup_steps: 100 , compress_statistics: NaN , dataset: NaN , gradient_accumulation_steps: NaN , learning_rate: NaN , quant_type: NaN , adam_beta2: NaN , update_freq: 8 | |
| eval_bert_f1 mean (SE): 89.3073 (0.0000). 95% CI (89.307, 89.307). Sample size: 2 | |
| eval_bert_f1 mean (SE): 89.3073 (0.0000). 95% CI (89.307, 89.307). Sample size: 2 | |
| eval_rougeL mean (SE): 60.0922 (0.0000). 95% CI (60.092, 60.092). Sample size: 2 | |
| ================================================================================ | |
| Config: max_steps: 5000, lora_r: 128 , lr: 3e-06, bf16: True, lora_modules: ffn , bits: 16 , full_finetune: True, lora_dropout: 0.0 , warmup_steps: 500 , compress_statistics: NaN , dataset: NaN , gradient_accumulation_steps: NaN , learning_rate: NaN , quant_type: NaN , adam_beta2: NaN , update_freq: 4 | |
| eval_bert_f1 mean (SE): 89.4275 (0.0000). 95% CI (89.427, 89.427). Sample size: 2 | |
| eval_bert_f1 mean (SE): 89.4275 (0.0000). 95% CI (89.427, 89.427). Sample size: 2 | |
| eval_rougeL mean (SE): 60.1178 (0.0000). 95% CI (60.118, 60.118). Sample size: 2 | |
| ================================================================================ | |
| Config: max_steps: 2500, lora_r: 128 , lr: 3e-06, bf16: True, lora_modules: ffn , bits: 16 , full_finetune: True, lora_dropout: 0.0 , warmup_steps: 500 , compress_statistics: NaN , dataset: NaN , gradient_accumulation_steps: NaN , learning_rate: NaN , quant_type: NaN , adam_beta2: NaN , update_freq: 8 | |
| eval_bert_f1 mean (SE): 89.4301 (0.0000). 95% CI (89.430, 89.430). Sample size: 2 | |
| eval_bert_f1 mean (SE): 89.4301 (0.0000). 95% CI (89.430, 89.430). Sample size: 2 | |
| eval_rougeL mean (SE): 59.9984 (0.0000). 95% CI (59.998, 59.998). Sample size: 2 | |
| ================================================================================ | |
| Config: max_steps: 2500, lora_r: 128 , lr: 0.0003, bf16: True, lora_modules: ffn , bits: 16 , full_finetune: True, lora_dropout: 0.0 , warmup_steps: 500 , compress_statistics: NaN , dataset: NaN , gradient_accumulation_steps: NaN , learning_rate: NaN , quant_type: NaN , adam_beta2: NaN , update_freq: 8 | |
| eval_bert_f1 mean (SE): 89.4670 (0.0000). 95% CI (89.467, 89.467). Sample size: 2 | |
| eval_bert_f1 mean (SE): 89.4670 (0.0000). 95% CI (89.467, 89.467). Sample size: 2 | |
| eval_rougeL mean (SE): 60.0147 (0.0000). 95% CI (60.015, 60.015). Sample size: 2 | |
| ================================================================================ | |
| Config: max_steps: 312 , lora_r: 8 , lr: 0.0003, bf16: False, lora_modules: baseline, bits: 8 , full_finetune: False, lora_dropout: 0.05, warmup_steps: NaN , compress_statistics: NaN , dataset: NaN , gradient_accumulation_steps: NaN , learning_rate: NaN , quant_type: NaN , adam_beta2: NaN , update_freq: 64 | |
| eval_bert_f1 mean (SE): 89.4972 (0.1115). 95% CI (89.279, 89.716). Sample size: 6 | |
| eval_bert_f1 mean (SE): 89.4972 (0.1115). 95% CI (89.279, 89.716). Sample size: 6 | |
| eval_rougeL mean (SE): 60.6429 (0.3112). 95% CI (60.033, 61.253). Sample size: 6 | |
| ================================================================================ | |
| Config: max_steps: 5000, lora_r: 64 , lr: 0.001, bf16: False, lora_modules: all , bits: 4 , full_finetune: False, lora_dropout: 0.0 , warmup_steps: NaN , compress_statistics: NaN , dataset: NaN , gradient_accumulation_steps: NaN , learning_rate: NaN , quant_type: NaN , adam_beta2: NaN , update_freq: 4 | |
| eval_bert_f1 mean (SE): 89.5113 (0.0095). 95% CI (89.493, 89.530). Sample size: 4 | |
| eval_bert_f1 mean (SE): 89.5113 (0.0095). 95% CI (89.493, 89.530). Sample size: 4 | |
| eval_rougeL mean (SE): 60.1750 (0.1485). 95% CI (59.884, 60.466). Sample size: 4 | |
| ================================================================================ | |
| Config: max_steps: 2500, lora_r: 128 , lr: 0.0001, bf16: False, lora_modules: attn, bits: 4 , full_finetune: False, lora_dropout: 0.0 , warmup_steps: NaN , compress_statistics: NaN , dataset: NaN , gradient_accumulation_steps: NaN , learning_rate: NaN , quant_type: NaN , adam_beta2: NaN , update_freq: 8 | |
| eval_bert_f1 mean (SE): 89.5207 (0.2994). 95% CI (88.934, 90.108). Sample size: 4 | |
| eval_bert_f1 mean (SE): 89.5207 (0.2994). 95% CI (88.934, 90.108). Sample size: 4 | |
| eval_rougeL mean (SE): 60.5209 (1.1890). 95% CI (58.190, 62.851). Sample size: 4 | |
| ================================================================================ | |
| Config: max_steps: 5000, lora_r: 128 , lr: 0.001, bf16: False, lora_modules: ffn , bits: 4 , full_finetune: False, lora_dropout: 0.0 , warmup_steps: NaN , compress_statistics: NaN , dataset: NaN , gradient_accumulation_steps: NaN , learning_rate: NaN , quant_type: NaN , adam_beta2: NaN , update_freq: 4 | |
| eval_bert_f1 mean (SE): 89.6191 (0.0000). 95% CI (89.619, 89.619). Sample size: 2 | |
| eval_bert_f1 mean (SE): 89.6191 (0.0000). 95% CI (89.619, 89.619). Sample size: 2 | |
| eval_rougeL mean (SE): 60.5949 (0.0000). 95% CI (60.595, 60.595). Sample size: 2 | |
| ================================================================================ | |
| Config: max_steps: 5000, lora_r: 64 , lr: 0.001, bf16: False, lora_modules: ffn , bits: 4 , full_finetune: False, lora_dropout: 0.0 , warmup_steps: NaN , compress_statistics: NaN , dataset: NaN , gradient_accumulation_steps: NaN , learning_rate: NaN , quant_type: NaN , adam_beta2: NaN , update_freq: 4 | |
| eval_bert_f1 mean (SE): 89.6576 (0.0000). 95% CI (89.658, 89.658). Sample size: 2 | |
| eval_bert_f1 mean (SE): 89.6576 (0.0000). 95% CI (89.658, 89.658). Sample size: 2 | |
| eval_rougeL mean (SE): 60.5405 (0.0000). 95% CI (60.541, 60.541). Sample size: 2 | |
| ================================================================================ | |
| Config: max_steps: 5000, lora_r: 128 , lr: 0.001, bf16: False, lora_modules: ffn , bits: 4 , full_finetune: False, lora_dropout: 0.05, warmup_steps: NaN , compress_statistics: NaN , dataset: NaN , gradient_accumulation_steps: NaN , learning_rate: NaN , quant_type: NaN , adam_beta2: NaN , update_freq: 4 | |
| eval_bert_f1 mean (SE): 89.6644 (0.0000). 95% CI (89.664, 89.664). Sample size: 2 | |
| eval_bert_f1 mean (SE): 89.6644 (0.0000). 95% CI (89.664, 89.664). Sample size: 2 | |
| eval_rougeL mean (SE): 60.5080 (0.0000). 95% CI (60.508, 60.508). Sample size: 2 | |
| ================================================================================ | |
| Config: max_steps: 5000, lora_r: 64 , lr: 0.001, bf16: False, lora_modules: attn, bits: 4 , full_finetune: False, lora_dropout: 0.0 , warmup_steps: NaN , compress_statistics: NaN , dataset: NaN , gradient_accumulation_steps: NaN , learning_rate: NaN , quant_type: NaN , adam_beta2: NaN , update_freq: 4 | |
| eval_bert_f1 mean (SE): 89.6770 (0.0000). 95% CI (89.677, 89.677). Sample size: 2 | |
| eval_bert_f1 mean (SE): 89.6770 (0.0000). 95% CI (89.677, 89.677). Sample size: 2 | |
| eval_rougeL mean (SE): 61.1801 (0.0000). 95% CI (61.180, 61.180). Sample size: 2 | |
| ================================================================================ | |
| Config: max_steps: 5000, lora_r: 128 , lr: 0.0003, bf16: False, lora_modules: all , bits: 4 , full_finetune: False, lora_dropout: 0.0 , warmup_steps: NaN , compress_statistics: NaN , dataset: NaN , gradient_accumulation_steps: NaN , learning_rate: NaN , quant_type: NaN , adam_beta2: NaN , update_freq: 4 | |
| eval_bert_f1 mean (SE): 89.6805 (0.0025). 95% CI (89.676, 89.685). Sample size: 4 | |
| eval_bert_f1 mean (SE): 89.6805 (0.0025). 95% CI (89.676, 89.685). Sample size: 4 | |
| eval_rougeL mean (SE): 61.0410 (0.0164). 95% CI (61.009, 61.073). Sample size: 4 | |
| ================================================================================ | |
| Config: max_steps: 17500, lora_r: 64 , lr: 0.0002, bf16: False, lora_modules: all , bits: 4 , full_finetune: False, lora_dropout: 0.0 , warmup_steps: 100 , compress_statistics: True, dataset: NaN , gradient_accumulation_steps: NaN , learning_rate: NaN , quant_type: nf4 , adam_beta2: 0.95, update_freq: 6 | |
| eval_bert_f1 mean (SE): 89.6974 (0.2965). 95% CI (89.116, 90.279). Sample size: 4 | |
| eval_bert_f1 mean (SE): 89.6974 (0.2965). 95% CI (89.116, 90.279). Sample size: 4 | |
| eval_rougeL mean (SE): 64.1488 (1.1610). 95% CI (61.873, 66.424). Sample size: 4 | |
| ================================================================================ | |
| Config: max_steps: 5000, lora_r: 128 , lr: 0.0003, bf16: False, lora_modules: all , bits: 4 , full_finetune: False, lora_dropout: 0.05, warmup_steps: NaN , compress_statistics: NaN , dataset: NaN , gradient_accumulation_steps: NaN , learning_rate: NaN , quant_type: NaN , adam_beta2: NaN , update_freq: 4 | |
| eval_bert_f1 mean (SE): 89.7082 (0.0000). 95% CI (89.708, 89.708). Sample size: 2 | |
| eval_bert_f1 mean (SE): 89.7082 (0.0000). 95% CI (89.708, 89.708). Sample size: 2 | |
| eval_rougeL mean (SE): 61.0613 (0.0000). 95% CI (61.061, 61.061). Sample size: 2 | |
| ================================================================================ | |
| Config: max_steps: 5000, lora_r: 64 , lr: 0.001, bf16: False, lora_modules: ffn , bits: 4 , full_finetune: False, lora_dropout: 0.05, warmup_steps: NaN , compress_statistics: NaN , dataset: NaN , gradient_accumulation_steps: NaN , learning_rate: NaN , quant_type: NaN , adam_beta2: NaN , update_freq: 4 | |
| eval_bert_f1 mean (SE): 89.7292 (0.0000). 95% CI (89.729, 89.729). Sample size: 2 | |
| eval_bert_f1 mean (SE): 89.7292 (0.0000). 95% CI (89.729, 89.729). Sample size: 2 | |
| eval_rougeL mean (SE): 60.7493 (0.0000). 95% CI (60.749, 60.749). Sample size: 2 | |
| ================================================================================ | |
| Config: max_steps: 2500, lora_r: 128 , lr: 0.0001, bf16: False, lora_modules: all , bits: 4 , full_finetune: False, lora_dropout: 0.0 , warmup_steps: NaN , compress_statistics: NaN , dataset: NaN , gradient_accumulation_steps: NaN , learning_rate: NaN , quant_type: NaN , adam_beta2: NaN , update_freq: 8 | |
| eval_bert_f1 mean (SE): 89.7932 (0.3093). 95% CI (89.187, 90.399). Sample size: 4 | |
| eval_bert_f1 mean (SE): 89.7932 (0.3093). 95% CI (89.187, 90.399). Sample size: 4 | |
| eval_rougeL mean (SE): 61.3837 (1.3568). 95% CI (58.724, 64.043). Sample size: 4 | |
| ================================================================================ | |
| Config: max_steps: 5000, lora_r: 128 , lr: 3e-05, bf16: True, lora_modules: ffn , bits: 16 , full_finetune: True, lora_dropout: 0.0 , warmup_steps: 500 , compress_statistics: NaN , dataset: NaN , gradient_accumulation_steps: NaN , learning_rate: NaN , quant_type: NaN , adam_beta2: NaN , update_freq: 4 | |
| eval_bert_f1 mean (SE): 89.7989 (0.0000). 95% CI (89.799, 89.799). Sample size: 2 | |
| eval_bert_f1 mean (SE): 89.7989 (0.0000). 95% CI (89.799, 89.799). Sample size: 2 | |
| eval_rougeL mean (SE): 62.5366 (0.0000). 95% CI (62.537, 62.537). Sample size: 2 | |
| ================================================================================ | |
| Config: max_steps: 17500, lora_r: 64 , lr: 0.0002, bf16: False, lora_modules: all , bits: 4 , full_finetune: False, lora_dropout: 0.0 , warmup_steps: 100 , compress_statistics: True, dataset: NaN , gradient_accumulation_steps: NaN , learning_rate: NaN , quant_type: fp4 , adam_beta2: 0.95, update_freq: 6 | |
| eval_bert_f1 mean (SE): 89.8283 (0.4397). 95% CI (88.966, 90.690). Sample size: 4 | |
| eval_bert_f1 mean (SE): 89.8283 (0.4397). 95% CI (88.966, 90.690). Sample size: 4 | |
| eval_rougeL mean (SE): 64.5086 (1.2929). 95% CI (61.975, 67.043). Sample size: 4 | |
| ================================================================================ | |
| Config: max_steps: 2500, lora_r: 64 , lr: 0.001, bf16: False, lora_modules: all , bits: 4 , full_finetune: False, lora_dropout: 0.0 , warmup_steps: NaN , compress_statistics: NaN , dataset: NaN , gradient_accumulation_steps: NaN , learning_rate: NaN , quant_type: NaN , adam_beta2: NaN , update_freq: 8 | |
| eval_bert_f1 mean (SE): 89.8341 (0.0000). 95% CI (89.834, 89.834). Sample size: 2 | |
| eval_bert_f1 mean (SE): 89.8341 (0.0000). 95% CI (89.834, 89.834). Sample size: 2 | |
| eval_rougeL mean (SE): 61.4766 (0.0000). 95% CI (61.477, 61.477). Sample size: 2 | |
| ================================================================================ | |
| Config: max_steps: 5000, lora_r: 128 , lr: 0.001, bf16: False, lora_modules: attn, bits: 4 , full_finetune: False, lora_dropout: 0.05, warmup_steps: NaN , compress_statistics: NaN , dataset: NaN , gradient_accumulation_steps: NaN , learning_rate: NaN , quant_type: NaN , adam_beta2: NaN , update_freq: 4 | |
| eval_bert_f1 mean (SE): 89.8997 (0.0000). 95% CI (89.900, 89.900). Sample size: 2 | |
| eval_bert_f1 mean (SE): 89.8997 (0.0000). 95% CI (89.900, 89.900). Sample size: 2 | |
| eval_rougeL mean (SE): 61.6869 (0.0000). 95% CI (61.687, 61.687). Sample size: 2 | |
| ================================================================================ | |
| Config: max_steps: 2500, lora_r: 128 , lr: 0.001, bf16: False, lora_modules: attn, bits: 4 , full_finetune: False, lora_dropout: 0.0 , warmup_steps: NaN , compress_statistics: NaN , dataset: NaN , gradient_accumulation_steps: NaN , learning_rate: NaN , quant_type: NaN , adam_beta2: NaN , update_freq: 8 | |
| eval_bert_f1 mean (SE): 89.9110 (0.0000). 95% CI (89.911, 89.911). Sample size: 2 | |
| eval_bert_f1 mean (SE): 89.9110 (0.0000). 95% CI (89.911, 89.911). Sample size: 2 | |
| eval_rougeL mean (SE): 61.4808 (0.0000). 95% CI (61.481, 61.481). Sample size: 2 | |
| ================================================================================ | |
| Config: max_steps: 2500, lora_r: 64 , lr: 0.001, bf16: False, lora_modules: ffn , bits: 4 , full_finetune: False, lora_dropout: 0.05, warmup_steps: NaN , compress_statistics: NaN , dataset: NaN , gradient_accumulation_steps: NaN , learning_rate: NaN , quant_type: NaN , adam_beta2: NaN , update_freq: 8 | |
| eval_bert_f1 mean (SE): 89.9213 (0.2109). 95% CI (89.508, 90.335). Sample size: 4 | |
| eval_bert_f1 mean (SE): 89.9213 (0.2109). 95% CI (89.508, 90.335). Sample size: 4 | |
| eval_rougeL mean (SE): 62.2937 (0.7943). 95% CI (60.737, 63.850). Sample size: 4 | |
| ================================================================================ | |
| Config: max_steps: 2500, lora_r: 64 , lr: 0.001, bf16: False, lora_modules: all , bits: 4 , full_finetune: False, lora_dropout: 0.05, warmup_steps: NaN , compress_statistics: NaN , dataset: NaN , gradient_accumulation_steps: NaN , learning_rate: NaN , quant_type: NaN , adam_beta2: NaN , update_freq: 8 | |
| eval_bert_f1 mean (SE): 89.9853 (0.0000). 95% CI (89.985, 89.985). Sample size: 2 | |
| eval_bert_f1 mean (SE): 89.9853 (0.0000). 95% CI (89.985, 89.985). Sample size: 2 | |
| eval_rougeL mean (SE): 62.3881 (0.0000). 95% CI (62.388, 62.388). Sample size: 2 | |
| ================================================================================ | |
| Config: max_steps: 2500, lora_r: 64 , lr: 0.0001, bf16: False, lora_modules: attn, bits: 4 , full_finetune: False, lora_dropout: 0.05, warmup_steps: NaN , compress_statistics: NaN , dataset: NaN , gradient_accumulation_steps: NaN , learning_rate: NaN , quant_type: NaN , adam_beta2: NaN , update_freq: 8 | |
| eval_bert_f1 mean (SE): 90.0105 (0.0071). 95% CI (89.997, 90.024). Sample size: 4 | |
| eval_bert_f1 mean (SE): 90.0105 (0.0071). 95% CI (89.997, 90.024). Sample size: 4 | |
| eval_rougeL mean (SE): 62.4882 (0.0079). 95% CI (62.473, 62.504). Sample size: 4 | |
| ================================================================================ | |
| Config: max_steps: 2500, lora_r: 128 , lr: 0.0001, bf16: False, lora_modules: attn, bits: 4 , full_finetune: False, lora_dropout: 0.05, warmup_steps: NaN , compress_statistics: NaN , dataset: NaN , gradient_accumulation_steps: NaN , learning_rate: NaN , quant_type: NaN , adam_beta2: NaN , update_freq: 8 | |
| eval_bert_f1 mean (SE): 90.0159 (0.0000). 95% CI (90.016, 90.016). Sample size: 2 | |
| eval_bert_f1 mean (SE): 90.0159 (0.0000). 95% CI (90.016, 90.016). Sample size: 2 | |
| eval_rougeL mean (SE): 62.5617 (0.0000). 95% CI (62.562, 62.562). Sample size: 2 | |
| ================================================================================ | |
| Config: max_steps: 5000, lora_r: 128 , lr: 0.0003, bf16: False, lora_modules: ffn , bits: 4 , full_finetune: False, lora_dropout: 0.0 , warmup_steps: NaN , compress_statistics: NaN , dataset: NaN , gradient_accumulation_steps: NaN , learning_rate: NaN , quant_type: NaN , adam_beta2: NaN , update_freq: 4 | |
| eval_bert_f1 mean (SE): 90.0186 (0.2255). 95% CI (89.577, 90.461). Sample size: 4 | |
| eval_bert_f1 mean (SE): 90.0186 (0.2255). 95% CI (89.577, 90.461). Sample size: 4 | |
| eval_rougeL mean (SE): 62.6738 (0.8986). 95% CI (60.912, 64.435). Sample size: 4 | |
| ================================================================================ | |
| Config: max_steps: 5000, lora_r: 64 , lr: 0.0001, bf16: False, lora_modules: attn, bits: 4 , full_finetune: False, lora_dropout: 0.0 , warmup_steps: NaN , compress_statistics: NaN , dataset: NaN , gradient_accumulation_steps: NaN , learning_rate: NaN , quant_type: NaN , adam_beta2: NaN , update_freq: 4 | |
| eval_bert_f1 mean (SE): 90.0220 (0.0152). 95% CI (89.992, 90.052). Sample size: 4 | |
| eval_bert_f1 mean (SE): 90.0220 (0.0152). 95% CI (89.992, 90.052). Sample size: 4 | |
| eval_rougeL mean (SE): 62.5605 (0.0155). 95% CI (62.530, 62.591). Sample size: 4 | |
| ================================================================================ | |
| Config: max_steps: 2500, lora_r: 16 , lr: 0.0002, bf16: False, lora_modules: all , bits: 4 , full_finetune: False, lora_dropout: 0.0 , warmup_steps: 400 , compress_statistics: NaN , dataset: NaN , gradient_accumulation_steps: NaN , learning_rate: NaN , quant_type: NaN , adam_beta2: NaN , update_freq: 8 | |
| eval_bert_f1 mean (SE): 90.0251 (0.3552). 95% CI (89.329, 90.721). Sample size: 2 | |
| eval_bert_f1 mean (SE): 90.0251 (0.3552). 95% CI (89.329, 90.721). Sample size: 2 | |
| eval_rougeL mean (SE): 62.4433 (1.6534). 95% CI (59.203, 65.684). Sample size: 2 | |
| ================================================================================ | |
| Config: max_steps: 2500, lora_r: 64 , lr: 0.0001, bf16: False, lora_modules: attn, bits: 4 , full_finetune: False, lora_dropout: 0.0 , warmup_steps: NaN , compress_statistics: NaN , dataset: NaN , gradient_accumulation_steps: NaN , learning_rate: NaN , quant_type: NaN , adam_beta2: NaN , update_freq: 8 | |
| eval_bert_f1 mean (SE): 90.0283 (0.0020). 95% CI (90.024, 90.032). Sample size: 4 | |
| eval_bert_f1 mean (SE): 90.0283 (0.0020). 95% CI (90.024, 90.032). Sample size: 4 | |
| eval_rougeL mean (SE): 62.5920 (0.0202). 95% CI (62.552, 62.632). Sample size: 4 | |
| ================================================================================ | |
| Config: max_steps: 2500, lora_r: 128 , lr: 0.001, bf16: False, lora_modules: attn, bits: 4 , full_finetune: False, lora_dropout: 0.05, warmup_steps: NaN , compress_statistics: NaN , dataset: NaN , gradient_accumulation_steps: NaN , learning_rate: NaN , quant_type: NaN , adam_beta2: NaN , update_freq: 8 | |
| eval_bert_f1 mean (SE): 90.0394 (0.0000). 95% CI (90.039, 90.039). Sample size: 2 | |
| eval_bert_f1 mean (SE): 90.0394 (0.0000). 95% CI (90.039, 90.039). Sample size: 2 | |
| eval_rougeL mean (SE): 62.7824 (0.0000). 95% CI (62.782, 62.782). Sample size: 2 | |
| ================================================================================ | |
| Config: max_steps: 5000, lora_r: 64 , lr: 0.0002, bf16: True, lora_modules: all , bits: 4 , full_finetune: False, lora_dropout: 0.0 , warmup_steps: 100 , compress_statistics: True, dataset: NaN , gradient_accumulation_steps: NaN , learning_rate: NaN , quant_type: nf4 , adam_beta2: 0.95, update_freq: 6 | |
| eval_bert_f1 mean (SE): 90.0401 (0.2407). 95% CI (89.568, 90.512). Sample size: 4 | |
| eval_bert_f1 mean (SE): 90.0401 (0.2407). 95% CI (89.568, 90.512). Sample size: 4 | |
| eval_rougeL mean (SE): 63.8055 (0.3053). 95% CI (63.207, 64.404). Sample size: 4 | |
| ================================================================================ | |
| Config: max_steps: 5000, lora_r: 128 , lr: 0.0001, bf16: False, lora_modules: attn, bits: 4 , full_finetune: False, lora_dropout: 0.05, warmup_steps: NaN , compress_statistics: NaN , dataset: NaN , gradient_accumulation_steps: NaN , learning_rate: NaN , quant_type: NaN , adam_beta2: NaN , update_freq: 4 | |
| eval_bert_f1 mean (SE): 90.0454 (0.0000). 95% CI (90.045, 90.045). Sample size: 2 | |
| eval_bert_f1 mean (SE): 90.0454 (0.0000). 95% CI (90.045, 90.045). Sample size: 2 | |
| eval_rougeL mean (SE): 62.5829 (0.0000). 95% CI (62.583, 62.583). Sample size: 2 | |
| ================================================================================ | |
| Config: max_steps: 2500, lora_r: 128 , lr: 1e-05, bf16: True, lora_modules: ffn , bits: 16 , full_finetune: True, lora_dropout: 0.0 , warmup_steps: 500 , compress_statistics: NaN , dataset: NaN , gradient_accumulation_steps: NaN , learning_rate: NaN , quant_type: NaN , adam_beta2: NaN , update_freq: 8 | |
| eval_bert_f1 mean (SE): 90.0468 (0.0000). 95% CI (90.047, 90.047). Sample size: 4 | |
| eval_bert_f1 mean (SE): 90.0468 (0.0000). 95% CI (90.047, 90.047). Sample size: 4 | |
| eval_rougeL mean (SE): 62.6748 (0.0005). 95% CI (62.674, 62.676). Sample size: 4 | |
| ================================================================================ | |
| Config: max_steps: 5000, lora_r: 64 , lr: 0.0001, bf16: False, lora_modules: attn, bits: 4 , full_finetune: False, lora_dropout: 0.05, warmup_steps: NaN , compress_statistics: NaN , dataset: NaN , gradient_accumulation_steps: NaN , learning_rate: NaN , quant_type: NaN , adam_beta2: NaN , update_freq: 4 | |
| eval_bert_f1 mean (SE): 90.0514 (0.0000). 95% CI (90.051, 90.051). Sample size: 2 | |
| eval_bert_f1 mean (SE): 90.0514 (0.0000). 95% CI (90.051, 90.051). Sample size: 2 | |
| eval_rougeL mean (SE): 62.5382 (0.0000). 95% CI (62.538, 62.538). Sample size: 2 | |
| ================================================================================ | |
| Config: max_steps: 2500, lora_r: 64 , lr: 0.001, bf16: False, lora_modules: attn, bits: 4 , full_finetune: False, lora_dropout: 0.05, warmup_steps: NaN , compress_statistics: NaN , dataset: NaN , gradient_accumulation_steps: NaN , learning_rate: NaN , quant_type: NaN , adam_beta2: NaN , update_freq: 8 | |
| eval_bert_f1 mean (SE): 90.0547 (0.0296). 95% CI (89.997, 90.113). Sample size: 4 | |
| eval_bert_f1 mean (SE): 90.0547 (0.0296). 95% CI (89.997, 90.113). Sample size: 4 | |
| eval_rougeL mean (SE): 62.5070 (0.3897). 95% CI (61.743, 63.271). Sample size: 4 | |
| ================================================================================ | |
| Config: max_steps: 5000, lora_r: 128 , lr: 0.0001, bf16: False, lora_modules: attn, bits: 4 , full_finetune: False, lora_dropout: 0.0 , warmup_steps: NaN , compress_statistics: NaN , dataset: NaN , gradient_accumulation_steps: NaN , learning_rate: NaN , quant_type: NaN , adam_beta2: NaN , update_freq: 4 | |
| eval_bert_f1 mean (SE): 90.0595 (0.0000). 95% CI (90.060, 90.060). Sample size: 2 | |
| eval_bert_f1 mean (SE): 90.0595 (0.0000). 95% CI (90.060, 90.060). Sample size: 2 | |
| eval_rougeL mean (SE): 62.6029 (0.0000). 95% CI (62.603, 62.603). Sample size: 2 | |
| ================================================================================ | |
| Config: max_steps: 5000, lora_r: 128 , lr: 0.001, bf16: False, lora_modules: attn, bits: 4 , full_finetune: False, lora_dropout: 0.0 , warmup_steps: NaN , compress_statistics: NaN , dataset: NaN , gradient_accumulation_steps: NaN , learning_rate: NaN , quant_type: NaN , adam_beta2: NaN , update_freq: 4 | |
| eval_bert_f1 mean (SE): 90.0675 (0.0000). 95% CI (90.067, 90.067). Sample size: 2 | |
| eval_bert_f1 mean (SE): 90.0675 (0.0000). 95% CI (90.067, 90.067). Sample size: 2 | |
| eval_rougeL mean (SE): 63.4846 (0.0000). 95% CI (63.485, 63.485). Sample size: 2 | |
| ================================================================================ | |
| Config: max_steps: 2000, lora_r: 64 , lr: 3e-05, bf16: False, lora_modules: all , bits: 4 , full_finetune: False, lora_dropout: 0.0 , warmup_steps: 100 , compress_statistics: NaN , dataset: NaN , gradient_accumulation_steps: NaN , learning_rate: NaN , quant_type: NaN , adam_beta2: NaN , update_freq: 10 | |
| eval_bert_f1 mean (SE): 90.0744 (nan). 95% CI (nan, nan). Sample size: 1 | |
| eval_bert_f1 mean (SE): 90.0744 (nan). 95% CI (nan, nan). Sample size: 1 | |
| eval_rougeL mean (SE): 62.6029 (nan). 95% CI (nan, nan). Sample size: 1 | |
| ================================================================================ | |
| Config: max_steps: 17500, lora_r: 64 , lr: 0.0002, bf16: True, lora_modules: all , bits: 4 , full_finetune: False, lora_dropout: 0.0 , warmup_steps: 100 , compress_statistics: True, dataset: NaN , gradient_accumulation_steps: NaN , learning_rate: NaN , quant_type: fp4 , adam_beta2: 0.95, update_freq: 6 | |
| eval_bert_f1 mean (SE): 90.0962 (0.1754). 95% CI (89.752, 90.440). Sample size: 4 | |
| eval_bert_f1 mean (SE): 90.0962 (0.1754). 95% CI (89.752, 90.440). Sample size: 4 | |
| eval_rougeL mean (SE): 64.7923 (0.9330). 95% CI (62.964, 66.621). Sample size: 4 | |
| ================================================================================ | |
| Config: max_steps: 2500, lora_r: 64 , lr: 3e-05, bf16: False, lora_modules: all , bits: 4 , full_finetune: False, lora_dropout: 0.0 , warmup_steps: 100 , compress_statistics: NaN , dataset: NaN , gradient_accumulation_steps: NaN , learning_rate: NaN , quant_type: NaN , adam_beta2: NaN , update_freq: 8 | |
| eval_bert_f1 mean (SE): 90.1059 (nan). 95% CI (nan, nan). Sample size: 1 | |
| eval_bert_f1 mean (SE): 90.1059 (nan). 95% CI (nan, nan). Sample size: 1 | |
| eval_rougeL mean (SE): 62.9100 (nan). 95% CI (nan, nan). Sample size: 1 | |
| ================================================================================ | |
| Config: max_steps: 5000, lora_r: 64 , lr: 0.0002, bf16: True, lora_modules: all , bits: 4 , full_finetune: False, lora_dropout: 0.0 , warmup_steps: 100 , compress_statistics: False, dataset: NaN , gradient_accumulation_steps: NaN , learning_rate: NaN , quant_type: nf4 , adam_beta2: 0.95, update_freq: 6 | |
| eval_bert_f1 mean (SE): 90.1083 (0.1817). 95% CI (89.752, 90.464). Sample size: 4 | |
| eval_bert_f1 mean (SE): 90.1083 (0.1817). 95% CI (89.752, 90.464). Sample size: 4 | |
| eval_rougeL mean (SE): 63.3727 (0.4927). 95% CI (62.407, 64.338). Sample size: 4 | |
| ================================================================================ | |
| Config: max_steps: 17500, lora_r: 64 , lr: 0.0002, bf16: False, lora_modules: all , bits: 4 , full_finetune: False, lora_dropout: 0.0 , warmup_steps: 100 , compress_statistics: False, dataset: NaN , gradient_accumulation_steps: NaN , learning_rate: NaN , quant_type: nf4 , adam_beta2: 0.95, update_freq: 6 | |
| eval_bert_f1 mean (SE): 90.1147 (0.1922). 95% CI (89.738, 90.491). Sample size: 4 | |
| eval_bert_f1 mean (SE): 90.1147 (0.1922). 95% CI (89.738, 90.491). Sample size: 4 | |
| eval_rougeL mean (SE): 65.0393 (0.7645). 95% CI (63.541, 66.538). Sample size: 4 | |
| ================================================================================ | |
| Config: max_steps: 5000, lora_r: 64 , lr: 0.001, bf16: False, lora_modules: attn, bits: 4 , full_finetune: False, lora_dropout: 0.05, warmup_steps: NaN , compress_statistics: NaN , dataset: NaN , gradient_accumulation_steps: NaN , learning_rate: NaN , quant_type: NaN , adam_beta2: NaN , update_freq: 4 | |
| eval_bert_f1 mean (SE): 90.1163 (0.0000). 95% CI (90.116, 90.116). Sample size: 2 | |
| eval_bert_f1 mean (SE): 90.1163 (0.0000). 95% CI (90.116, 90.116). Sample size: 2 | |
| eval_rougeL mean (SE): 62.5445 (0.0000). 95% CI (62.544, 62.544). Sample size: 2 | |
| ================================================================================ | |
| Config: max_steps: 5000, lora_r: 64 , lr: 0.001, bf16: False, lora_modules: all , bits: 4 , full_finetune: False, lora_dropout: 0.05, warmup_steps: NaN , compress_statistics: NaN , dataset: NaN , gradient_accumulation_steps: NaN , learning_rate: NaN , quant_type: NaN , adam_beta2: NaN , update_freq: 4 | |
| eval_bert_f1 mean (SE): 90.1317 (0.0000). 95% CI (90.132, 90.132). Sample size: 2 | |
| eval_bert_f1 mean (SE): 90.1317 (0.0000). 95% CI (90.132, 90.132). Sample size: 2 | |
| eval_rougeL mean (SE): 64.0623 (0.0000). 95% CI (64.062, 64.062). Sample size: 2 | |
| ================================================================================ | |
| Config: max_steps: 3333, lora_r: 64 , lr: 3e-05, bf16: False, lora_modules: all , bits: 4 , full_finetune: False, lora_dropout: 0.0 , warmup_steps: 100 , compress_statistics: NaN , dataset: NaN , gradient_accumulation_steps: NaN , learning_rate: NaN , quant_type: NaN , adam_beta2: NaN , update_freq: 6 | |
| eval_bert_f1 mean (SE): 90.1332 (nan). 95% CI (nan, nan). Sample size: 1 | |
| eval_bert_f1 mean (SE): 90.1332 (nan). 95% CI (nan, nan). Sample size: 1 | |
| eval_rougeL mean (SE): 62.9626 (nan). 95% CI (nan, nan). Sample size: 1 | |
| ================================================================================ | |
| Config: max_steps: 5000, lora_r: 128 , lr: 1e-05, bf16: True, lora_modules: ffn , bits: 16 , full_finetune: True, lora_dropout: 0.0 , warmup_steps: 500 , compress_statistics: NaN , dataset: NaN , gradient_accumulation_steps: NaN , learning_rate: NaN , quant_type: NaN , adam_beta2: NaN , update_freq: 4 | |
| eval_bert_f1 mean (SE): 90.1441 (0.0000). 95% CI (90.144, 90.144). Sample size: 4 | |
| eval_bert_f1 mean (SE): 90.1441 (0.0000). 95% CI (90.144, 90.144). Sample size: 4 | |
| eval_rougeL mean (SE): 62.6560 (0.0031). 95% CI (62.650, 62.662). Sample size: 4 | |
| ================================================================================ | |
| Config: max_steps: 3200, lora_r: 64 , lr: 3e-05, bf16: False, lora_modules: all , bits: 4 , full_finetune: False, lora_dropout: 0.0 , warmup_steps: 100 , compress_statistics: NaN , dataset: NaN , gradient_accumulation_steps: NaN , learning_rate: NaN , quant_type: NaN , adam_beta2: NaN , update_freq: 10 | |
| eval_bert_f1 mean (SE): 90.1614 (nan). 95% CI (nan, nan). Sample size: 1 | |
| eval_bert_f1 mean (SE): 90.1614 (nan). 95% CI (nan, nan). Sample size: 1 | |
| eval_rougeL mean (SE): 63.0538 (nan). 95% CI (nan, nan). Sample size: 1 | |
| ================================================================================ | |
| Config: max_steps: 2500, lora_r: 64 , lr: 0.001, bf16: False, lora_modules: attn, bits: 4 , full_finetune: False, lora_dropout: 0.0 , warmup_steps: NaN , compress_statistics: NaN , dataset: NaN , gradient_accumulation_steps: NaN , learning_rate: NaN , quant_type: NaN , adam_beta2: NaN , update_freq: 8 | |
| eval_bert_f1 mean (SE): 90.1729 (0.0932). 95% CI (89.990, 90.356). Sample size: 4 | |
| eval_bert_f1 mean (SE): 90.1729 (0.0932). 95% CI (89.990, 90.356). Sample size: 4 | |
| eval_rougeL mean (SE): 63.0638 (0.6259). 95% CI (61.837, 64.291). Sample size: 4 | |
| ================================================================================ | |
| Config: max_steps: 5000, lora_r: 64 , lr: 0.0003, bf16: False, lora_modules: attn, bits: 4 , full_finetune: False, lora_dropout: 0.0 , warmup_steps: NaN , compress_statistics: NaN , dataset: NaN , gradient_accumulation_steps: NaN , learning_rate: NaN , quant_type: NaN , adam_beta2: NaN , update_freq: 4 | |
| eval_bert_f1 mean (SE): 90.1744 (0.0000). 95% CI (90.174, 90.174). Sample size: 2 | |
| eval_bert_f1 mean (SE): 90.1744 (0.0000). 95% CI (90.174, 90.174). Sample size: 2 | |
| eval_rougeL mean (SE): 62.8048 (0.0000). 95% CI (62.805, 62.805). Sample size: 2 | |
| ================================================================================ | |
| Config: max_steps: 2500, lora_r: 128 , lr: 0.0003, bf16: False, lora_modules: attn, bits: 4 , full_finetune: False, lora_dropout: 0.05, warmup_steps: NaN , compress_statistics: NaN , dataset: NaN , gradient_accumulation_steps: NaN , learning_rate: NaN , quant_type: NaN , adam_beta2: NaN , update_freq: 8 | |
| eval_bert_f1 mean (SE): 90.1762 (0.0040). 95% CI (90.168, 90.184). Sample size: 4 | |
| eval_bert_f1 mean (SE): 90.1762 (0.0040). 95% CI (90.168, 90.184). Sample size: 4 | |
| eval_rougeL mean (SE): 62.7251 (0.0659). 95% CI (62.596, 62.854). Sample size: 4 | |
| ================================================================================ | |
| Config: max_steps: 5000, lora_r: 128 , lr: 0.0003, bf16: False, lora_modules: attn, bits: 4 , full_finetune: False, lora_dropout: 0.0 , warmup_steps: NaN , compress_statistics: NaN , dataset: NaN , gradient_accumulation_steps: NaN , learning_rate: NaN , quant_type: NaN , adam_beta2: NaN , update_freq: 4 | |
| eval_bert_f1 mean (SE): 90.1768 (0.0000). 95% CI (90.177, 90.177). Sample size: 2 | |
| eval_bert_f1 mean (SE): 90.1768 (0.0000). 95% CI (90.177, 90.177). Sample size: 2 | |
| eval_rougeL mean (SE): 62.8331 (0.0000). 95% CI (62.833, 62.833). Sample size: 2 | |
| ================================================================================ | |
| Config: max_steps: 5000, lora_r: 64 , lr: 0.0002, bf16: True, lora_modules: all , bits: 4 , full_finetune: False, lora_dropout: 0.0 , warmup_steps: 100 , compress_statistics: False, dataset: NaN , gradient_accumulation_steps: NaN , learning_rate: NaN , quant_type: fp4 , adam_beta2: 0.95, update_freq: 6 | |
| eval_bert_f1 mean (SE): 90.1768 (0.2433). 95% CI (89.700, 90.654). Sample size: 4 | |
| eval_bert_f1 mean (SE): 90.1768 (0.2433). 95% CI (89.700, 90.654). Sample size: 4 | |
| eval_rougeL mean (SE): 63.1856 (1.3049). 95% CI (60.628, 65.743). Sample size: 4 | |
| ================================================================================ | |
| Config: max_steps: 5000, lora_r: 64 , lr: 0.0001, bf16: False, lora_modules: ffn , bits: 4 , full_finetune: False, lora_dropout: 0.0 , warmup_steps: NaN , compress_statistics: NaN , dataset: NaN , gradient_accumulation_steps: NaN , learning_rate: NaN , quant_type: NaN , adam_beta2: NaN , update_freq: 4 | |
| eval_bert_f1 mean (SE): 90.1780 (0.0396). 95% CI (90.100, 90.256). Sample size: 4 | |
| eval_bert_f1 mean (SE): 90.1780 (0.0396). 95% CI (90.100, 90.256). Sample size: 4 | |
| eval_rougeL mean (SE): 63.1610 (0.2294). 95% CI (62.711, 63.611). Sample size: 4 | |
| ================================================================================ | |
| Config: max_steps: 5000, lora_r: 128 , lr: 0.001, bf16: False, lora_modules: all , bits: 4 , full_finetune: False, lora_dropout: 0.0 , warmup_steps: NaN , compress_statistics: NaN , dataset: NaN , gradient_accumulation_steps: NaN , learning_rate: NaN , quant_type: NaN , adam_beta2: NaN , update_freq: 4 | |
| eval_bert_f1 mean (SE): 90.1790 (0.0000). 95% CI (90.179, 90.179). Sample size: 2 | |
| eval_bert_f1 mean (SE): 90.1790 (0.0000). 95% CI (90.179, 90.179). Sample size: 2 | |
| eval_rougeL mean (SE): 63.9024 (0.0000). 95% CI (63.902, 63.902). Sample size: 2 | |
| ================================================================================ | |
| Config: max_steps: 2500, lora_r: 64 , lr: 0.0003, bf16: False, lora_modules: attn, bits: 4 , full_finetune: False, lora_dropout: 0.05, warmup_steps: NaN , compress_statistics: NaN , dataset: NaN , gradient_accumulation_steps: NaN , learning_rate: NaN , quant_type: NaN , adam_beta2: NaN , update_freq: 8 | |
| eval_bert_f1 mean (SE): 90.1795 (0.0000). 95% CI (90.179, 90.179). Sample size: 2 | |
| eval_bert_f1 mean (SE): 90.1795 (0.0000). 95% CI (90.179, 90.179). Sample size: 2 | |
| eval_rougeL mean (SE): 62.6180 (0.0000). 95% CI (62.618, 62.618). Sample size: 2 | |
| ================================================================================ | |
| Config: max_steps: 2500, lora_r: 128 , lr: 1e-05, bf16: True, lora_modules: ffn , bits: 16 , full_finetune: True, lora_dropout: 0.0 , warmup_steps: 100 , compress_statistics: NaN , dataset: NaN , gradient_accumulation_steps: NaN , learning_rate: NaN , quant_type: NaN , adam_beta2: NaN , update_freq: 8 | |
| eval_bert_f1 mean (SE): 90.1854 (0.0000). 95% CI (90.185, 90.185). Sample size: 4 | |
| eval_bert_f1 mean (SE): 90.1854 (0.0000). 95% CI (90.185, 90.185). Sample size: 4 | |
| eval_rougeL mean (SE): 62.8517 (0.0035). 95% CI (62.845, 62.858). Sample size: 4 | |
| ================================================================================ | |
| Config: max_steps: 2500, lora_r: 64 , lr: 0.0003, bf16: False, lora_modules: attn, bits: 4 , full_finetune: False, lora_dropout: 0.0 , warmup_steps: NaN , compress_statistics: NaN , dataset: NaN , gradient_accumulation_steps: NaN , learning_rate: NaN , quant_type: NaN , adam_beta2: NaN , update_freq: 8 | |
| eval_bert_f1 mean (SE): 90.1895 (0.0016). 95% CI (90.186, 90.193). Sample size: 4 | |
| eval_bert_f1 mean (SE): 90.1895 (0.0016). 95% CI (90.186, 90.193). Sample size: 4 | |
| eval_rougeL mean (SE): 62.8498 (0.0119). 95% CI (62.826, 62.873). Sample size: 4 | |
| ================================================================================ | |
| Config: max_steps: 2500, lora_r: 128 , lr: 0.0003, bf16: False, lora_modules: attn, bits: 4 , full_finetune: False, lora_dropout: 0.0 , warmup_steps: NaN , compress_statistics: NaN , dataset: NaN , gradient_accumulation_steps: NaN , learning_rate: NaN , quant_type: NaN , adam_beta2: NaN , update_freq: 8 | |
| eval_bert_f1 mean (SE): 90.1935 (0.0000). 95% CI (90.193, 90.193). Sample size: 2 | |
| eval_bert_f1 mean (SE): 90.1935 (0.0000). 95% CI (90.193, 90.193). Sample size: 2 | |
| eval_rougeL mean (SE): 62.6587 (0.0000). 95% CI (62.659, 62.659). Sample size: 2 | |
| ================================================================================ | |
| Config: max_steps: 4000, lora_r: 64 , lr: 3e-05, bf16: False, lora_modules: all , bits: 4 , full_finetune: False, lora_dropout: 0.0 , warmup_steps: 100 , compress_statistics: NaN , dataset: NaN , gradient_accumulation_steps: NaN , learning_rate: NaN , quant_type: NaN , adam_beta2: NaN , update_freq: 8 | |
| eval_bert_f1 mean (SE): 90.1983 (nan). 95% CI (nan, nan). Sample size: 1 | |
| eval_bert_f1 mean (SE): 90.1983 (nan). 95% CI (nan, nan). Sample size: 1 | |
| eval_rougeL mean (SE): 63.2548 (nan). 95% CI (nan, nan). Sample size: 1 | |
| ================================================================================ | |
| Config: max_steps: 2500, lora_r: 128 , lr: 0.001, bf16: False, lora_modules: all , bits: 4 , full_finetune: False, lora_dropout: 0.0 , warmup_steps: NaN , compress_statistics: NaN , dataset: NaN , gradient_accumulation_steps: NaN , learning_rate: NaN , quant_type: NaN , adam_beta2: NaN , update_freq: 8 | |
| eval_bert_f1 mean (SE): 90.1988 (0.0525). 95% CI (90.096, 90.302). Sample size: 4 | |
| eval_bert_f1 mean (SE): 90.1988 (0.0525). 95% CI (90.096, 90.302). Sample size: 4 | |
| eval_rougeL mean (SE): 63.4968 (0.6637). 95% CI (62.196, 64.798). Sample size: 4 | |
| ================================================================================ | |
| Config: max_steps: 2500, lora_r: 64 , lr: 0.0001, bf16: False, lora_modules: ffn , bits: 4 , full_finetune: False, lora_dropout: 0.05, warmup_steps: NaN , compress_statistics: NaN , dataset: NaN , gradient_accumulation_steps: NaN , learning_rate: NaN , quant_type: NaN , adam_beta2: NaN , update_freq: 8 | |
| eval_bert_f1 mean (SE): 90.2026 (0.0000). 95% CI (90.203, 90.203). Sample size: 2 | |
| eval_bert_f1 mean (SE): 90.2026 (0.0000). 95% CI (90.203, 90.203). Sample size: 2 | |
| eval_rougeL mean (SE): 63.3859 (0.0000). 95% CI (63.386, 63.386). Sample size: 2 | |
| ================================================================================ | |
| Config: max_steps: 4000, lora_r: 64 , lr: 3e-05, bf16: False, lora_modules: all , bits: 4 , full_finetune: False, lora_dropout: 0.0 , warmup_steps: 100 , compress_statistics: NaN , dataset: NaN , gradient_accumulation_steps: NaN , learning_rate: NaN , quant_type: NaN , adam_beta2: NaN , update_freq: 10 | |
| eval_bert_f1 mean (SE): 90.2043 (nan). 95% CI (nan, nan). Sample size: 1 | |
| eval_bert_f1 mean (SE): 90.2043 (nan). 95% CI (nan, nan). Sample size: 1 | |
| eval_rougeL mean (SE): 63.3455 (nan). 95% CI (nan, nan). Sample size: 1 | |
| ================================================================================ | |
| Config: max_steps: 5000, lora_r: 64 , lr: 0.0003, bf16: False, lora_modules: attn, bits: 4 , full_finetune: False, lora_dropout: 0.05, warmup_steps: NaN , compress_statistics: NaN , dataset: NaN , gradient_accumulation_steps: NaN , learning_rate: NaN , quant_type: NaN , adam_beta2: NaN , update_freq: 4 | |
| eval_bert_f1 mean (SE): 90.2045 (0.0126). 95% CI (90.180, 90.229). Sample size: 4 | |
| eval_bert_f1 mean (SE): 90.2045 (0.0126). 95% CI (90.180, 90.229). Sample size: 4 | |
| eval_rougeL mean (SE): 63.0460 (0.1153). 95% CI (62.820, 63.272). Sample size: 4 | |
| ================================================================================ | |
| Config: max_steps: 2500, lora_r: 64 , lr: 0.0001, bf16: False, lora_modules: ffn , bits: 4 , full_finetune: False, lora_dropout: 0.0 , warmup_steps: NaN , compress_statistics: NaN , dataset: NaN , gradient_accumulation_steps: NaN , learning_rate: NaN , quant_type: NaN , adam_beta2: NaN , update_freq: 8 | |
| eval_bert_f1 mean (SE): 90.2102 (0.0000). 95% CI (90.210, 90.210). Sample size: 2 | |
| eval_bert_f1 mean (SE): 90.2102 (0.0000). 95% CI (90.210, 90.210). Sample size: 2 | |
| eval_rougeL mean (SE): 63.3770 (0.0000). 95% CI (63.377, 63.377). Sample size: 2 | |
| ================================================================================ | |
| Config: max_steps: 5333, lora_r: 64 , lr: 3e-05, bf16: False, lora_modules: all , bits: 4 , full_finetune: False, lora_dropout: 0.0 , warmup_steps: 100 , compress_statistics: NaN , dataset: NaN , gradient_accumulation_steps: NaN , learning_rate: NaN , quant_type: NaN , adam_beta2: NaN , update_freq: 6 | |
| eval_bert_f1 mean (SE): 90.2154 (nan). 95% CI (nan, nan). Sample size: 1 | |
| eval_bert_f1 mean (SE): 90.2154 (nan). 95% CI (nan, nan). Sample size: 1 | |
| eval_rougeL mean (SE): 63.4377 (nan). 95% CI (nan, nan). Sample size: 1 | |
| ================================================================================ | |
| Config: max_steps: 5000, lora_r: 128 , lr: 1e-05, bf16: True, lora_modules: ffn , bits: 16 , full_finetune: True, lora_dropout: 0.0 , warmup_steps: 100 , compress_statistics: NaN , dataset: NaN , gradient_accumulation_steps: NaN , learning_rate: NaN , quant_type: NaN , adam_beta2: NaN , update_freq: 4 | |
| eval_bert_f1 mean (SE): 90.2160 (0.0000). 95% CI (90.216, 90.216). Sample size: 4 | |
| eval_bert_f1 mean (SE): 90.2160 (0.0000). 95% CI (90.216, 90.216). Sample size: 4 | |
| eval_rougeL mean (SE): 63.1442 (0.0118). 95% CI (63.121, 63.167). Sample size: 4 | |
| ================================================================================ | |
| Config: max_steps: 2500, lora_r: 128 , lr: 0.0001, bf16: False, lora_modules: ffn , bits: 4 , full_finetune: False, lora_dropout: 0.0 , warmup_steps: NaN , compress_statistics: NaN , dataset: NaN , gradient_accumulation_steps: NaN , learning_rate: NaN , quant_type: NaN , adam_beta2: NaN , update_freq: 8 | |
| eval_bert_f1 mean (SE): 90.2170 (0.0000). 95% CI (90.217, 90.217). Sample size: 2 | |
| eval_bert_f1 mean (SE): 90.2170 (0.0000). 95% CI (90.217, 90.217). Sample size: 2 | |
| eval_rougeL mean (SE): 63.3327 (0.0000). 95% CI (63.333, 63.333). Sample size: 2 | |
| ================================================================================ | |
| Config: max_steps: 2500, lora_r: 128 , lr: 0.0001, bf16: False, lora_modules: ffn , bits: 4 , full_finetune: False, lora_dropout: 0.05, warmup_steps: NaN , compress_statistics: NaN , dataset: NaN , gradient_accumulation_steps: NaN , learning_rate: NaN , quant_type: NaN , adam_beta2: NaN , update_freq: 8 | |
| eval_bert_f1 mean (SE): 90.2219 (0.0000). 95% CI (90.222, 90.222). Sample size: 2 | |
| eval_bert_f1 mean (SE): 90.2219 (0.0000). 95% CI (90.222, 90.222). Sample size: 2 | |
| eval_rougeL mean (SE): 63.3905 (0.0000). 95% CI (63.390, 63.390). Sample size: 2 | |
| ================================================================================ | |
| Config: max_steps: 5000, lora_r: 128 , lr: 0.0003, bf16: False, lora_modules: attn, bits: 4 , full_finetune: False, lora_dropout: 0.05, warmup_steps: NaN , compress_statistics: NaN , dataset: NaN , gradient_accumulation_steps: NaN , learning_rate: NaN , quant_type: NaN , adam_beta2: NaN , update_freq: 4 | |
| eval_bert_f1 mean (SE): 90.2261 (0.0000). 95% CI (90.226, 90.226). Sample size: 2 | |
| eval_bert_f1 mean (SE): 90.2261 (0.0000). 95% CI (90.226, 90.226). Sample size: 2 | |
| eval_rougeL mean (SE): 63.2395 (0.0000). 95% CI (63.239, 63.239). Sample size: 2 | |
| ================================================================================ | |
| Config: max_steps: 5000, lora_r: 64 , lr: 3e-05, bf16: False, lora_modules: all , bits: 4 , full_finetune: False, lora_dropout: 0.0 , warmup_steps: 100 , compress_statistics: NaN , dataset: NaN , gradient_accumulation_steps: NaN , learning_rate: NaN , quant_type: NaN , adam_beta2: NaN , update_freq: 8 | |
| eval_bert_f1 mean (SE): 90.2354 (nan). 95% CI (nan, nan). Sample size: 1 | |
| eval_bert_f1 mean (SE): 90.2354 (nan). 95% CI (nan, nan). Sample size: 1 | |
| eval_rougeL mean (SE): 63.4383 (nan). 95% CI (nan, nan). Sample size: 1 | |
| ================================================================================ | |
| Config: max_steps: 17500, lora_r: 64 , lr: 0.0002, bf16: True, lora_modules: all , bits: 4 , full_finetune: False, lora_dropout: 0.0 , warmup_steps: 100 , compress_statistics: False, dataset: NaN , gradient_accumulation_steps: NaN , learning_rate: NaN , quant_type: nf4 , adam_beta2: 0.95, update_freq: 6 | |
| eval_bert_f1 mean (SE): 90.2424 (0.1353). 95% CI (89.977, 90.508). Sample size: 6 | |
| eval_bert_f1 mean (SE): 90.2424 (0.1353). 95% CI (89.977, 90.508). Sample size: 6 | |
| eval_rougeL mean (SE): 64.8326 (0.4831). 95% CI (63.886, 65.779). Sample size: 6 | |
| ================================================================================ | |
| Config: max_steps: 5000, lora_r: 64 , lr: 0.0001, bf16: False, lora_modules: ffn , bits: 4 , full_finetune: False, lora_dropout: 0.05, warmup_steps: NaN , compress_statistics: NaN , dataset: NaN , gradient_accumulation_steps: NaN , learning_rate: NaN , quant_type: NaN , adam_beta2: NaN , update_freq: 4 | |
| eval_bert_f1 mean (SE): 90.2478 (0.0000). 95% CI (90.248, 90.248). Sample size: 2 | |
| eval_bert_f1 mean (SE): 90.2478 (0.0000). 95% CI (90.248, 90.248). Sample size: 2 | |
| eval_rougeL mean (SE): 63.6017 (0.0000). 95% CI (63.602, 63.602). Sample size: 2 | |
| ================================================================================ | |
| Config: max_steps: 5000, lora_r: 128 , lr: 0.0001, bf16: False, lora_modules: ffn , bits: 4 , full_finetune: False, lora_dropout: 0.05, warmup_steps: NaN , compress_statistics: NaN , dataset: NaN , gradient_accumulation_steps: NaN , learning_rate: NaN , quant_type: NaN , adam_beta2: NaN , update_freq: 4 | |
| eval_bert_f1 mean (SE): 90.2567 (0.0000). 95% CI (90.257, 90.257). Sample size: 2 | |
| eval_bert_f1 mean (SE): 90.2567 (0.0000). 95% CI (90.257, 90.257). Sample size: 2 | |
| eval_rougeL mean (SE): 63.5939 (0.0000). 95% CI (63.594, 63.594). Sample size: 2 | |
| ================================================================================ | |
| Config: max_steps: 2000, lora_r: 32 , lr: 0.0002, bf16: False, lora_modules: all , bits: 4 , full_finetune: False, lora_dropout: 0.0 , warmup_steps: 400 , compress_statistics: NaN , dataset: NaN , gradient_accumulation_steps: NaN , learning_rate: NaN , quant_type: NaN , adam_beta2: NaN , update_freq: 16 | |
| eval_bert_f1 mean (SE): 90.2623 (0.1462). 95% CI (89.976, 90.549). Sample size: 2 | |
| eval_bert_f1 mean (SE): 90.2623 (0.1462). 95% CI (89.976, 90.549). Sample size: 2 | |
| eval_rougeL mean (SE): 63.4570 (0.8202). 95% CI (61.849, 65.065). Sample size: 2 | |
| ================================================================================ | |
| Config: max_steps: 2500, lora_r: 128 , lr: 0.001, bf16: False, lora_modules: all , bits: 4 , full_finetune: False, lora_dropout: 0.05, warmup_steps: NaN , compress_statistics: NaN , dataset: NaN , gradient_accumulation_steps: NaN , learning_rate: NaN , quant_type: NaN , adam_beta2: NaN , update_freq: 8 | |
| eval_bert_f1 mean (SE): 90.2643 (0.0000). 95% CI (90.264, 90.264). Sample size: 2 | |
| eval_bert_f1 mean (SE): 90.2643 (0.0000). 95% CI (90.264, 90.264). Sample size: 2 | |
| eval_rougeL mean (SE): 64.7841 (0.0000). 95% CI (64.784, 64.784). Sample size: 2 | |
| ================================================================================ | |
| Config: max_steps: 5000, lora_r: 128 , lr: 0.0001, bf16: False, lora_modules: ffn , bits: 4 , full_finetune: False, lora_dropout: 0.0 , warmup_steps: NaN , compress_statistics: NaN , dataset: NaN , gradient_accumulation_steps: NaN , learning_rate: NaN , quant_type: NaN , adam_beta2: NaN , update_freq: 4 | |
| eval_bert_f1 mean (SE): 90.2645 (0.0000). 95% CI (90.265, 90.265). Sample size: 2 | |
| eval_bert_f1 mean (SE): 90.2645 (0.0000). 95% CI (90.265, 90.265). Sample size: 2 | |
| eval_rougeL mean (SE): 63.6360 (0.0000). 95% CI (63.636, 63.636). Sample size: 2 | |
| ================================================================================ | |
| Config: max_steps: 2000, lora_r: 64 , lr: 0.0001, bf16: False, lora_modules: all , bits: 4 , full_finetune: False, lora_dropout: 0.0 , warmup_steps: 100 , compress_statistics: NaN , dataset: NaN , gradient_accumulation_steps: NaN , learning_rate: NaN , quant_type: NaN , adam_beta2: NaN , update_freq: 10 | |
| eval_bert_f1 mean (SE): 90.2900 (nan). 95% CI (nan, nan). Sample size: 1 | |
| eval_bert_f1 mean (SE): 90.2900 (nan). 95% CI (nan, nan). Sample size: 1 | |
| eval_rougeL mean (SE): 63.6535 (nan). 95% CI (nan, nan). Sample size: 1 | |
| ================================================================================ | |
| Config: max_steps: 6667, lora_r: 64 , lr: 3e-05, bf16: False, lora_modules: all , bits: 4 , full_finetune: False, lora_dropout: 0.0 , warmup_steps: 100 , compress_statistics: NaN , dataset: NaN , gradient_accumulation_steps: NaN , learning_rate: NaN , quant_type: NaN , adam_beta2: NaN , update_freq: 6 | |
| eval_bert_f1 mean (SE): 90.2911 (nan). 95% CI (nan, nan). Sample size: 1 | |
| eval_bert_f1 mean (SE): 90.2911 (nan). 95% CI (nan, nan). Sample size: 1 | |
| eval_rougeL mean (SE): 63.6195 (nan). 95% CI (nan, nan). Sample size: 1 | |
| ================================================================================ | |
| Config: max_steps: 2500, lora_r: 64 , lr: 0.0001, bf16: False, lora_modules: all , bits: 4 , full_finetune: False, lora_dropout: 0.05, warmup_steps: NaN , compress_statistics: NaN , dataset: NaN , gradient_accumulation_steps: NaN , learning_rate: NaN , quant_type: NaN , adam_beta2: NaN , update_freq: 8 | |
| eval_bert_f1 mean (SE): 90.3140 (0.0120). 95% CI (90.290, 90.338). Sample size: 4 | |
| eval_bert_f1 mean (SE): 90.3140 (0.0120). 95% CI (90.290, 90.338). Sample size: 4 | |
| eval_rougeL mean (SE): 63.6113 (0.0600). 95% CI (63.494, 63.729). Sample size: 4 | |
| ================================================================================ | |
| Config: max_steps: 17500, lora_r: 64 , lr: 0.0002, bf16: True, lora_modules: all , bits: 4 , full_finetune: False, lora_dropout: 0.0 , warmup_steps: 100 , compress_statistics: True, dataset: NaN , gradient_accumulation_steps: NaN , learning_rate: NaN , quant_type: nf4 , adam_beta2: 0.95, update_freq: 6 | |
| eval_bert_f1 mean (SE): 90.3148 (0.1176). 95% CI (90.084, 90.545). Sample size: 6 | |
| eval_bert_f1 mean (SE): 90.3148 (0.1176). 95% CI (90.084, 90.545). Sample size: 6 | |
| eval_rougeL mean (SE): 64.9091 (0.4367). 95% CI (64.053, 65.765). Sample size: 6 | |
| ================================================================================ | |
| Config: max_steps: 3333, lora_r: 64 , lr: 0.0001, bf16: False, lora_modules: all , bits: 4 , full_finetune: False, lora_dropout: 0.0 , warmup_steps: 100 , compress_statistics: NaN , dataset: NaN , gradient_accumulation_steps: NaN , learning_rate: NaN , quant_type: NaN , adam_beta2: NaN , update_freq: 6 | |
| eval_bert_f1 mean (SE): 90.3307 (nan). 95% CI (nan, nan). Sample size: 1 | |
| eval_bert_f1 mean (SE): 90.3307 (nan). 95% CI (nan, nan). Sample size: 1 | |
| eval_rougeL mean (SE): 63.7904 (nan). 95% CI (nan, nan). Sample size: 1 | |
| ================================================================================ | |
| Config: max_steps: 2500, lora_r: 128 , lr: 0.0001, bf16: False, lora_modules: all , bits: 4 , full_finetune: False, lora_dropout: 0.05, warmup_steps: NaN , compress_statistics: NaN , dataset: NaN , gradient_accumulation_steps: NaN , learning_rate: NaN , quant_type: NaN , adam_beta2: NaN , update_freq: 8 | |
| eval_bert_f1 mean (SE): 90.3322 (0.0000). 95% CI (90.332, 90.332). Sample size: 2 | |
| eval_bert_f1 mean (SE): 90.3322 (0.0000). 95% CI (90.332, 90.332). Sample size: 2 | |
| eval_rougeL mean (SE): 63.7222 (0.0000). 95% CI (63.722, 63.722). Sample size: 2 | |
| ================================================================================ | |
| Config: max_steps: 2500, lora_r: 64 , lr: 0.0001, bf16: False, lora_modules: all , bits: 4 , full_finetune: False, lora_dropout: 0.0 , warmup_steps: 100 , compress_statistics: NaN , dataset: NaN , gradient_accumulation_steps: NaN , learning_rate: NaN , quant_type: NaN , adam_beta2: NaN , update_freq: 8 | |
| eval_bert_f1 mean (SE): 90.3330 (nan). 95% CI (nan, nan). Sample size: 1 | |
| eval_bert_f1 mean (SE): 90.3330 (nan). 95% CI (nan, nan). Sample size: 1 | |
| eval_rougeL mean (SE): 63.6808 (nan). 95% CI (nan, nan). Sample size: 1 | |
| ================================================================================ | |
| Config: max_steps: 2500, lora_r: 64 , lr: 0.0001, bf16: False, lora_modules: all , bits: 4 , full_finetune: False, lora_dropout: 0.0 , warmup_steps: NaN , compress_statistics: NaN , dataset: NaN , gradient_accumulation_steps: NaN , learning_rate: NaN , quant_type: NaN , adam_beta2: NaN , update_freq: 8 | |
| eval_bert_f1 mean (SE): 90.3357 (0.0000). 95% CI (90.336, 90.336). Sample size: 2 | |
| eval_bert_f1 mean (SE): 90.3357 (0.0000). 95% CI (90.336, 90.336). Sample size: 2 | |
| eval_rougeL mean (SE): 63.7327 (0.0000). 95% CI (63.733, 63.733). Sample size: 2 | |
| ================================================================================ | |
| Config: max_steps: 5000, lora_r: 128 , lr: 0.0003, bf16: True, lora_modules: ffn , bits: 16 , full_finetune: False, lora_dropout: 0.0 , warmup_steps: NaN , compress_statistics: NaN , dataset: NaN , gradient_accumulation_steps: NaN , learning_rate: NaN , quant_type: NaN , adam_beta2: NaN , update_freq: 4 | |
| eval_bert_f1 mean (SE): 90.3517 (0.0196). 95% CI (90.313, 90.390). Sample size: 6 | |
| eval_bert_f1 mean (SE): 90.3517 (0.0196). 95% CI (90.313, 90.390). Sample size: 6 | |
| eval_rougeL mean (SE): 64.0017 (0.0245). 95% CI (63.954, 64.050). Sample size: 6 | |
| ================================================================================ | |
| Config: max_steps: 2500, lora_r: 64 , lr: 0.0003, bf16: False, lora_modules: ffn , bits: 4 , full_finetune: False, lora_dropout: 0.0 , warmup_steps: NaN , compress_statistics: NaN , dataset: NaN , gradient_accumulation_steps: NaN , learning_rate: NaN , quant_type: NaN , adam_beta2: NaN , update_freq: 8 | |
| eval_bert_f1 mean (SE): 90.3521 (0.0000). 95% CI (90.352, 90.352). Sample size: 2 | |
| eval_bert_f1 mean (SE): 90.3521 (0.0000). 95% CI (90.352, 90.352). Sample size: 2 | |
| eval_rougeL mean (SE): 63.9203 (0.0000). 95% CI (63.920, 63.920). Sample size: 2 | |
| ================================================================================ | |
| Config: max_steps: 2500, lora_r: 128 , lr: 0.0003, bf16: False, lora_modules: ffn , bits: 4 , full_finetune: False, lora_dropout: 0.0 , warmup_steps: NaN , compress_statistics: NaN , dataset: NaN , gradient_accumulation_steps: NaN , learning_rate: NaN , quant_type: NaN , adam_beta2: NaN , update_freq: 8 | |
| eval_bert_f1 mean (SE): 90.3659 (0.0000). 95% CI (90.366, 90.366). Sample size: 2 | |
| eval_bert_f1 mean (SE): 90.3659 (0.0000). 95% CI (90.366, 90.366). Sample size: 2 | |
| eval_rougeL mean (SE): 63.9706 (0.0000). 95% CI (63.971, 63.971). Sample size: 2 | |
| ================================================================================ | |
| Config: max_steps: 6000, lora_r: 64 , lr: 0.0002, bf16: True, lora_modules: all , bits: 4 , full_finetune: False, lora_dropout: 0.0 , warmup_steps: 100 , compress_statistics: False, dataset: NaN , gradient_accumulation_steps: NaN , learning_rate: NaN , quant_type: nf4 , adam_beta2: 0.95, update_freq: 6 | |
| eval_bert_f1 mean (SE): 90.3668 (0.0666). 95% CI (90.236, 90.497). Sample size: 6 | |
| eval_bert_f1 mean (SE): 90.3668 (0.0666). 95% CI (90.236, 90.497). Sample size: 6 | |
| eval_rougeL mean (SE): 63.3372 (0.4999). 95% CI (62.357, 64.317). Sample size: 6 | |
| ================================================================================ | |
| Config: max_steps: 2500, lora_r: 64 , lr: 0.0003, bf16: False, lora_modules: ffn , bits: 4 , full_finetune: False, lora_dropout: 0.05, warmup_steps: NaN , compress_statistics: NaN , dataset: NaN , gradient_accumulation_steps: NaN , learning_rate: NaN , quant_type: NaN , adam_beta2: NaN , update_freq: 8 | |
| eval_bert_f1 mean (SE): 90.3693 (0.0000). 95% CI (90.369, 90.369). Sample size: 2 | |
| eval_bert_f1 mean (SE): 90.3693 (0.0000). 95% CI (90.369, 90.369). Sample size: 2 | |
| eval_rougeL mean (SE): 63.9818 (0.0000). 95% CI (63.982, 63.982). Sample size: 2 | |
| ================================================================================ | |
| Config: max_steps: 17500, lora_r: 64 , lr: 0.0002, bf16: False, lora_modules: all , bits: 4 , full_finetune: False, lora_dropout: 0.0 , warmup_steps: 100 , compress_statistics: False, dataset: NaN , gradient_accumulation_steps: NaN , learning_rate: NaN , quant_type: fp4 , adam_beta2: 0.95, update_freq: 6 | |
| eval_bert_f1 mean (SE): 90.3713 (0.2573). 95% CI (89.867, 90.876). Sample size: 4 | |
| eval_bert_f1 mean (SE): 90.3713 (0.2573). 95% CI (89.867, 90.876). Sample size: 4 | |
| eval_rougeL mean (SE): 64.9940 (0.8270). 95% CI (63.373, 66.615). Sample size: 4 | |
| ================================================================================ | |
| Config: max_steps: 2500, lora_r: 128 , lr: 0.0003, bf16: False, lora_modules: ffn , bits: 4 , full_finetune: False, lora_dropout: 0.05, warmup_steps: NaN , compress_statistics: NaN , dataset: NaN , gradient_accumulation_steps: NaN , learning_rate: NaN , quant_type: NaN , adam_beta2: NaN , update_freq: 8 | |
| eval_bert_f1 mean (SE): 90.3771 (0.0000). 95% CI (90.377, 90.377). Sample size: 2 | |
| eval_bert_f1 mean (SE): 90.3771 (0.0000). 95% CI (90.377, 90.377). Sample size: 2 | |
| eval_rougeL mean (SE): 63.9350 (0.0000). 95% CI (63.935, 63.935). Sample size: 2 | |
| ================================================================================ | |
| Config: max_steps: 5000, lora_r: 128 , lr: 0.0003, bf16: False, lora_modules: ffn , bits: 4 , full_finetune: False, lora_dropout: 0.05, warmup_steps: NaN , compress_statistics: NaN , dataset: NaN , gradient_accumulation_steps: NaN , learning_rate: NaN , quant_type: NaN , adam_beta2: NaN , update_freq: 4 | |
| eval_bert_f1 mean (SE): 90.3772 (0.0000). 95% CI (90.377, 90.377). Sample size: 2 | |
| eval_bert_f1 mean (SE): 90.3772 (0.0000). 95% CI (90.377, 90.377). Sample size: 2 | |
| eval_rougeL mean (SE): 64.2891 (0.0000). 95% CI (64.289, 64.289). Sample size: 2 | |
| ================================================================================ | |
| Config: max_steps: 5000, lora_r: 128 , lr: 0.0001, bf16: False, lora_modules: all , bits: 4 , full_finetune: False, lora_dropout: 0.0 , warmup_steps: NaN , compress_statistics: NaN , dataset: NaN , gradient_accumulation_steps: NaN , learning_rate: NaN , quant_type: NaN , adam_beta2: NaN , update_freq: 4 | |
| eval_bert_f1 mean (SE): 90.3848 (0.0000). 95% CI (90.385, 90.385). Sample size: 2 | |
| eval_bert_f1 mean (SE): 90.3848 (0.0000). 95% CI (90.385, 90.385). Sample size: 2 | |
| eval_rougeL mean (SE): 63.8816 (0.0000). 95% CI (63.882, 63.882). Sample size: 2 | |
| ================================================================================ | |
| Config: max_steps: 6000, lora_r: 64 , lr: 0.0002, bf16: True, lora_modules: all , bits: 4 , full_finetune: False, lora_dropout: 0.0 , warmup_steps: 100 , compress_statistics: True, dataset: NaN , gradient_accumulation_steps: NaN , learning_rate: NaN , quant_type: nf4 , adam_beta2: 0.999, update_freq: 6 | |
| eval_bert_f1 mean (SE): 90.3889 (0.0500). 95% CI (90.291, 90.487). Sample size: 6 | |
| eval_bert_f1 mean (SE): 90.3889 (0.0500). 95% CI (90.291, 90.487). Sample size: 6 | |
| eval_rougeL mean (SE): 63.3567 (0.4462). 95% CI (62.482, 64.231). Sample size: 6 | |
| ================================================================================ | |
| Config: max_steps: 5000, lora_r: 64 , lr: 0.0001, bf16: False, lora_modules: all , bits: 4 , full_finetune: False, lora_dropout: 0.0 , warmup_steps: NaN , compress_statistics: NaN , dataset: NaN , gradient_accumulation_steps: NaN , learning_rate: NaN , quant_type: NaN , adam_beta2: NaN , update_freq: 4 | |
| eval_bert_f1 mean (SE): 90.3895 (0.0000). 95% CI (90.390, 90.390). Sample size: 2 | |
| eval_bert_f1 mean (SE): 90.3895 (0.0000). 95% CI (90.390, 90.390). Sample size: 2 | |
| eval_rougeL mean (SE): 63.9320 (0.0000). 95% CI (63.932, 63.932). Sample size: 2 | |
| ================================================================================ | |
| Config: max_steps: 2500, lora_r: 8 , lr: 0.0002, bf16: False, lora_modules: all , bits: 4 , full_finetune: False, lora_dropout: 0.0 , warmup_steps: 400 , compress_statistics: NaN , dataset: NaN , gradient_accumulation_steps: NaN , learning_rate: NaN , quant_type: NaN , adam_beta2: NaN , update_freq: 8 | |
| eval_bert_f1 mean (SE): 90.3923 (0.0086). 95% CI (90.375, 90.409). Sample size: 2 | |
| eval_bert_f1 mean (SE): 90.3923 (0.0086). 95% CI (90.375, 90.409). Sample size: 2 | |
| eval_rougeL mean (SE): 64.0548 (0.0309). 95% CI (63.994, 64.115). Sample size: 2 | |
| ================================================================================ | |
| Config: max_steps: 5000, lora_r: 128 , lr: 0.0001, bf16: False, lora_modules: all , bits: 4 , full_finetune: False, lora_dropout: 0.05, warmup_steps: NaN , compress_statistics: NaN , dataset: NaN , gradient_accumulation_steps: NaN , learning_rate: NaN , quant_type: NaN , adam_beta2: NaN , update_freq: 4 | |
| eval_bert_f1 mean (SE): 90.3934 (0.0000). 95% CI (90.393, 90.393). Sample size: 2 | |
| eval_bert_f1 mean (SE): 90.3934 (0.0000). 95% CI (90.393, 90.393). Sample size: 2 | |
| eval_rougeL mean (SE): 63.9398 (0.0000). 95% CI (63.940, 63.940). Sample size: 2 | |
| ================================================================================ | |
| Config: max_steps: 2500, lora_r: 64 , lr: 0.0003, bf16: False, lora_modules: all , bits: 4 , full_finetune: False, lora_dropout: 0.0 , warmup_steps: NaN , compress_statistics: NaN , dataset: NaN , gradient_accumulation_steps: NaN , learning_rate: NaN , quant_type: NaN , adam_beta2: NaN , update_freq: 8 | |
| eval_bert_f1 mean (SE): 90.3942 (0.0651). 95% CI (90.267, 90.522). Sample size: 4 | |
| eval_bert_f1 mean (SE): 90.3942 (0.0651). 95% CI (90.267, 90.522). Sample size: 4 | |
| eval_rougeL mean (SE): 63.8966 (0.3530). 95% CI (63.205, 64.588). Sample size: 4 | |
| ================================================================================ | |
| Config: max_steps: 5000, lora_r: 64 , lr: 0.0001, bf16: False, lora_modules: all , bits: 4 , full_finetune: False, lora_dropout: 0.05, warmup_steps: NaN , compress_statistics: NaN , dataset: NaN , gradient_accumulation_steps: NaN , learning_rate: NaN , quant_type: NaN , adam_beta2: NaN , update_freq: 4 | |
| eval_bert_f1 mean (SE): 90.3974 (0.0000). 95% CI (90.397, 90.397). Sample size: 2 | |
| eval_bert_f1 mean (SE): 90.3974 (0.0000). 95% CI (90.397, 90.397). Sample size: 2 | |
| eval_rougeL mean (SE): 63.9389 (0.0000). 95% CI (63.939, 63.939). Sample size: 2 | |
| ================================================================================ | |
| Config: max_steps: 2500, lora_r: 32 , lr: 0.0002, bf16: False, lora_modules: all , bits: 4 , full_finetune: False, lora_dropout: 0.0 , warmup_steps: 400 , compress_statistics: NaN , dataset: NaN , gradient_accumulation_steps: NaN , learning_rate: NaN , quant_type: NaN , adam_beta2: NaN , update_freq: 8 | |
| eval_bert_f1 mean (SE): 90.4017 (0.0030). 95% CI (90.396, 90.408). Sample size: 2 | |
| eval_bert_f1 mean (SE): 90.4017 (0.0030). 95% CI (90.396, 90.408). Sample size: 2 | |
| eval_rougeL mean (SE): 64.0156 (0.1635). 95% CI (63.695, 64.336). Sample size: 2 | |
| ================================================================================ | |
| Config: max_steps: 2500, lora_r: 64 , lr: 0.0002, bf16: False, lora_modules: all , bits: 4 , full_finetune: False, lora_dropout: 0.0 , warmup_steps: 400 , compress_statistics: NaN , dataset: NaN , gradient_accumulation_steps: NaN , learning_rate: NaN , quant_type: NaN , adam_beta2: NaN , update_freq: 8 | |
| eval_bert_f1 mean (SE): 90.4050 (0.0038). 95% CI (90.398, 90.413). Sample size: 2 | |
| eval_bert_f1 mean (SE): 90.4050 (0.0038). 95% CI (90.398, 90.413). Sample size: 2 | |
| eval_rougeL mean (SE): 64.0486 (0.1259). 95% CI (63.802, 64.295). Sample size: 2 | |
| ================================================================================ | |
| Config: max_steps: 5000, lora_r: 64 , lr: 0.0003, bf16: False, lora_modules: ffn , bits: 4 , full_finetune: False, lora_dropout: 0.0 , warmup_steps: NaN , compress_statistics: NaN , dataset: NaN , gradient_accumulation_steps: NaN , learning_rate: NaN , quant_type: NaN , adam_beta2: NaN , update_freq: 4 | |
| eval_bert_f1 mean (SE): 90.4086 (0.0000). 95% CI (90.409, 90.409). Sample size: 2 | |
| eval_bert_f1 mean (SE): 90.4086 (0.0000). 95% CI (90.409, 90.409). Sample size: 2 | |
| eval_rougeL mean (SE): 64.3748 (0.0000). 95% CI (64.375, 64.375). Sample size: 2 | |
| ================================================================================ | |
| Config: max_steps: 2000, lora_r: 8 , lr: 0.0002, bf16: False, lora_modules: all , bits: 4 , full_finetune: False, lora_dropout: 0.0 , warmup_steps: 400 , compress_statistics: NaN , dataset: NaN , gradient_accumulation_steps: NaN , learning_rate: NaN , quant_type: NaN , adam_beta2: NaN , update_freq: 16 | |
| eval_bert_f1 mean (SE): 90.4097 (0.0117). 95% CI (90.387, 90.433). Sample size: 2 | |
| eval_bert_f1 mean (SE): 90.4097 (0.0117). 95% CI (90.387, 90.433). Sample size: 2 | |
| eval_rougeL mean (SE): 64.3900 (0.0112). 95% CI (64.368, 64.412). Sample size: 2 | |
| ================================================================================ | |
| Config: max_steps: 4000, lora_r: 16 , lr: 0.0002, bf16: False, lora_modules: all , bits: 4 , full_finetune: False, lora_dropout: 0.0 , warmup_steps: 400 , compress_statistics: NaN , dataset: NaN , gradient_accumulation_steps: NaN , learning_rate: NaN , quant_type: NaN , adam_beta2: NaN , update_freq: 8 | |
| eval_bert_f1 mean (SE): 90.4110 (0.0049). 95% CI (90.401, 90.421). Sample size: 2 | |
| eval_bert_f1 mean (SE): 90.4110 (0.0049). 95% CI (90.401, 90.421). Sample size: 2 | |
| eval_rougeL mean (SE): 64.3389 (0.2461). 95% CI (63.857, 64.821). Sample size: 2 | |
| ================================================================================ | |
| Config: max_steps: 2500, lora_r: 64 , lr: 0.001, bf16: False, lora_modules: ffn , bits: 4 , full_finetune: False, lora_dropout: 0.0 , warmup_steps: NaN , compress_statistics: NaN , dataset: NaN , gradient_accumulation_steps: NaN , learning_rate: NaN , quant_type: NaN , adam_beta2: NaN , update_freq: 8 | |
| eval_bert_f1 mean (SE): 90.4122 (0.0235). 95% CI (90.366, 90.458). Sample size: 4 | |
| eval_bert_f1 mean (SE): 90.4122 (0.0235). 95% CI (90.366, 90.458). Sample size: 4 | |
| eval_rougeL mean (SE): 64.0394 (0.2378). 95% CI (63.573, 64.505). Sample size: 4 | |
| ================================================================================ | |
| Config: max_steps: 6000, lora_r: 64 , lr: 0.0002, bf16: True, lora_modules: all , bits: 4 , full_finetune: False, lora_dropout: 0.0 , warmup_steps: 100 , compress_statistics: True, dataset: NaN , gradient_accumulation_steps: NaN , learning_rate: NaN , quant_type: nf4 , adam_beta2: 0.95, update_freq: 6 | |
| eval_bert_f1 mean (SE): 90.4165 (0.0688). 95% CI (90.282, 90.551). Sample size: 6 | |
| eval_bert_f1 mean (SE): 90.4165 (0.0688). 95% CI (90.282, 90.551). Sample size: 6 | |
| eval_rougeL mean (SE): 63.8624 (0.5239). 95% CI (62.835, 64.889). Sample size: 6 | |
| ================================================================================ | |
| Config: max_steps: 2500, lora_r: 128 , lr: 0.001, bf16: False, lora_modules: ffn , bits: 4 , full_finetune: False, lora_dropout: 0.05, warmup_steps: NaN , compress_statistics: NaN , dataset: NaN , gradient_accumulation_steps: NaN , learning_rate: NaN , quant_type: NaN , adam_beta2: NaN , update_freq: 8 | |
| eval_bert_f1 mean (SE): 90.4180 (0.0000). 95% CI (90.418, 90.418). Sample size: 2 | |
| eval_bert_f1 mean (SE): 90.4180 (0.0000). 95% CI (90.418, 90.418). Sample size: 2 | |
| eval_rougeL mean (SE): 64.3409 (0.0000). 95% CI (64.341, 64.341). Sample size: 2 | |
| ================================================================================ | |
| Config: max_steps: 4000, lora_r: 8 , lr: 0.0002, bf16: False, lora_modules: all , bits: 4 , full_finetune: False, lora_dropout: 0.0 , warmup_steps: 100 , compress_statistics: NaN , dataset: NaN , gradient_accumulation_steps: NaN , learning_rate: NaN , quant_type: NaN , adam_beta2: NaN , update_freq: 8 | |
| eval_bert_f1 mean (SE): 90.4191 (0.0583). 95% CI (90.305, 90.533). Sample size: 2 | |
| eval_bert_f1 mean (SE): 90.4191 (0.0583). 95% CI (90.305, 90.533). Sample size: 2 | |
| eval_rougeL mean (SE): 64.3956 (0.1553). 95% CI (64.091, 64.700). Sample size: 2 | |
| ================================================================================ | |
| Config: max_steps: 5000, lora_r: 8 , lr: 0.0002, bf16: False, lora_modules: all , bits: 4 , full_finetune: False, lora_dropout: 0.0 , warmup_steps: 100 , compress_statistics: NaN , dataset: NaN , gradient_accumulation_steps: NaN , learning_rate: NaN , quant_type: NaN , adam_beta2: NaN , update_freq: 8 | |
| eval_bert_f1 mean (SE): 90.4198 (0.0232). 95% CI (90.374, 90.465). Sample size: 2 | |
| eval_bert_f1 mean (SE): 90.4198 (0.0232). 95% CI (90.374, 90.465). Sample size: 2 | |
| eval_rougeL mean (SE): 64.8022 (0.0298). 95% CI (64.744, 64.860). Sample size: 2 | |
| ================================================================================ | |
| Config: max_steps: 2000, lora_r: 8 , lr: 0.0002, bf16: False, lora_modules: all , bits: 4 , full_finetune: False, lora_dropout: 0.0 , warmup_steps: 100 , compress_statistics: NaN , dataset: NaN , gradient_accumulation_steps: NaN , learning_rate: NaN , quant_type: NaN , adam_beta2: NaN , update_freq: 16 | |
| eval_bert_f1 mean (SE): 90.4239 (0.0505). 95% CI (90.325, 90.523). Sample size: 2 | |
| eval_bert_f1 mean (SE): 90.4239 (0.0505). 95% CI (90.325, 90.523). Sample size: 2 | |
| eval_rougeL mean (SE): 64.3116 (0.0236). 95% CI (64.265, 64.358). Sample size: 2 | |
| ================================================================================ | |
| Config: max_steps: 6000, lora_r: 64 , lr: 0.0002, bf16: True, lora_modules: all , bits: 4 , full_finetune: False, lora_dropout: 0.0 , warmup_steps: 100 , compress_statistics: True, dataset: NaN , gradient_accumulation_steps: NaN , learning_rate: NaN , quant_type: fp4 , adam_beta2: 0.999, update_freq: 6 | |
| eval_bert_f1 mean (SE): 90.4249 (0.0469). 95% CI (90.333, 90.517). Sample size: 6 | |
| eval_bert_f1 mean (SE): 90.4249 (0.0469). 95% CI (90.333, 90.517). Sample size: 6 | |
| eval_rougeL mean (SE): 63.3294 (0.5269). 95% CI (62.297, 64.362). Sample size: 6 | |
| ================================================================================ | |
| Config: max_steps: 2500, lora_r: 16 , lr: 0.0002, bf16: False, lora_modules: all , bits: 4 , full_finetune: False, lora_dropout: 0.0 , warmup_steps: 100 , compress_statistics: NaN , dataset: NaN , gradient_accumulation_steps: NaN , learning_rate: NaN , quant_type: NaN , adam_beta2: NaN , update_freq: 8 | |
| eval_bert_f1 mean (SE): 90.4264 (0.0123). 95% CI (90.402, 90.451). Sample size: 2 | |
| eval_bert_f1 mean (SE): 90.4264 (0.0123). 95% CI (90.402, 90.451). Sample size: 2 | |
| eval_rougeL mean (SE): 64.1011 (0.1272). 95% CI (63.852, 64.350). Sample size: 2 | |
| ================================================================================ | |
| Config: max_steps: 5000, lora_r: 128 , lr: 3e-05, bf16: True, lora_modules: ffn , bits: 16 , full_finetune: True, lora_dropout: 0.0 , warmup_steps: 100 , compress_statistics: NaN , dataset: NaN , gradient_accumulation_steps: NaN , learning_rate: NaN , quant_type: NaN , adam_beta2: NaN , update_freq: 4 | |
| eval_bert_f1 mean (SE): 90.4274 (0.0000). 95% CI (90.427, 90.427). Sample size: 2 | |
| eval_bert_f1 mean (SE): 90.4274 (0.0000). 95% CI (90.427, 90.427). Sample size: 2 | |
| eval_rougeL mean (SE): 64.4043 (0.0000). 95% CI (64.404, 64.404). Sample size: 2 | |
| ================================================================================ | |
| Config: max_steps: 3200, lora_r: 64 , lr: 0.0001, bf16: False, lora_modules: all , bits: 4 , full_finetune: False, lora_dropout: 0.0 , warmup_steps: 100 , compress_statistics: NaN , dataset: NaN , gradient_accumulation_steps: NaN , learning_rate: NaN , quant_type: NaN , adam_beta2: NaN , update_freq: 10 | |
| eval_bert_f1 mean (SE): 90.4320 (nan). 95% CI (nan, nan). Sample size: 1 | |
| eval_bert_f1 mean (SE): 90.4320 (nan). 95% CI (nan, nan). Sample size: 1 | |
| eval_rougeL mean (SE): 64.1086 (nan). 95% CI (nan, nan). Sample size: 1 | |
| ================================================================================ | |
| Config: max_steps: 5000, lora_r: 64 , lr: 0.0002, bf16: True, lora_modules: all , bits: 4 , full_finetune: False, lora_dropout: 0.0 , warmup_steps: 100 , compress_statistics: True, dataset: NaN , gradient_accumulation_steps: NaN , learning_rate: NaN , quant_type: fp4 , adam_beta2: 0.95, update_freq: 6 | |
| eval_bert_f1 mean (SE): 90.4321 (0.2337). 95% CI (89.974, 90.890). Sample size: 4 | |
| eval_bert_f1 mean (SE): 90.4321 (0.2337). 95% CI (89.974, 90.890). Sample size: 4 | |
| eval_rougeL mean (SE): 64.6114 (0.1770). 95% CI (64.265, 64.958). Sample size: 4 | |
| ================================================================================ | |
| Config: max_steps: 2500, lora_r: 8 , lr: 0.0002, bf16: False, lora_modules: all , bits: 4 , full_finetune: False, lora_dropout: 0.0 , warmup_steps: 100 , compress_statistics: NaN , dataset: NaN , gradient_accumulation_steps: NaN , learning_rate: NaN , quant_type: NaN , adam_beta2: NaN , update_freq: 8 | |
| eval_bert_f1 mean (SE): 90.4325 (0.0047). 95% CI (90.423, 90.442). Sample size: 2 | |
| eval_bert_f1 mean (SE): 90.4325 (0.0047). 95% CI (90.423, 90.442). Sample size: 2 | |
| eval_rougeL mean (SE): 64.0860 (0.1174). 95% CI (63.856, 64.316). Sample size: 2 | |
| ================================================================================ | |
| Config: max_steps: 6000, lora_r: 64 , lr: 0.0002, bf16: True, lora_modules: all , bits: 4 , full_finetune: False, lora_dropout: 0.0 , warmup_steps: 100 , compress_statistics: False, dataset: NaN , gradient_accumulation_steps: NaN , learning_rate: NaN , quant_type: nf4 , adam_beta2: 0.999, update_freq: 6 | |
| eval_bert_f1 mean (SE): 90.4352 (0.0546). 95% CI (90.328, 90.542). Sample size: 6 | |
| eval_bert_f1 mean (SE): 90.4352 (0.0546). 95% CI (90.328, 90.542). Sample size: 6 | |
| eval_rougeL mean (SE): 63.6589 (0.5879). 95% CI (62.507, 64.811). Sample size: 6 | |
| ================================================================================ | |
| Config: max_steps: 4000, lora_r: 64 , lr: 0.0002, bf16: False, lora_modules: all , bits: 4 , full_finetune: False, lora_dropout: 0.0 , warmup_steps: 400 , compress_statistics: NaN , dataset: NaN , gradient_accumulation_steps: NaN , learning_rate: NaN , quant_type: NaN , adam_beta2: NaN , update_freq: 8 | |
| eval_bert_f1 mean (SE): 90.4377 (0.0129). 95% CI (90.413, 90.463). Sample size: 2 | |
| eval_bert_f1 mean (SE): 90.4377 (0.0129). 95% CI (90.413, 90.463). Sample size: 2 | |
| eval_rougeL mean (SE): 64.2244 (0.0805). 95% CI (64.067, 64.382). Sample size: 2 | |
| ================================================================================ | |
| Config: max_steps: 2500, lora_r: 32 , lr: 0.0002, bf16: False, lora_modules: all , bits: 4 , full_finetune: False, lora_dropout: 0.0 , warmup_steps: 100 , compress_statistics: NaN , dataset: NaN , gradient_accumulation_steps: NaN , learning_rate: NaN , quant_type: NaN , adam_beta2: NaN , update_freq: 8 | |
| eval_bert_f1 mean (SE): 90.4387 (0.0097). 95% CI (90.420, 90.458). Sample size: 2 | |
| eval_bert_f1 mean (SE): 90.4387 (0.0097). 95% CI (90.420, 90.458). Sample size: 2 | |
| eval_rougeL mean (SE): 64.0705 (0.0506). 95% CI (63.971, 64.170). Sample size: 2 | |
| ================================================================================ | |
| Config: max_steps: 4000, lora_r: 32 , lr: 0.0002, bf16: False, lora_modules: all , bits: 4 , full_finetune: False, lora_dropout: 0.0 , warmup_steps: 400 , compress_statistics: NaN , dataset: NaN , gradient_accumulation_steps: NaN , learning_rate: NaN , quant_type: NaN , adam_beta2: NaN , update_freq: 8 | |
| eval_bert_f1 mean (SE): 90.4399 (0.0090). 95% CI (90.422, 90.458). Sample size: 2 | |
| eval_bert_f1 mean (SE): 90.4399 (0.0090). 95% CI (90.422, 90.458). Sample size: 2 | |
| eval_rougeL mean (SE): 64.3609 (0.1783). 95% CI (64.011, 64.711). Sample size: 2 | |
| ================================================================================ | |
| Config: max_steps: 2500, lora_r: 64 , lr: 0.0002, bf16: False, lora_modules: all , bits: 4 , full_finetune: False, lora_dropout: 0.0 , warmup_steps: 100 , compress_statistics: NaN , dataset: NaN , gradient_accumulation_steps: NaN , learning_rate: NaN , quant_type: NaN , adam_beta2: NaN , update_freq: 8 | |
| eval_bert_f1 mean (SE): 90.4422 (0.0013). 95% CI (90.440, 90.445). Sample size: 2 | |
| eval_bert_f1 mean (SE): 90.4422 (0.0013). 95% CI (90.440, 90.445). Sample size: 2 | |
| eval_rougeL mean (SE): 64.0452 (0.0268). 95% CI (63.993, 64.098). Sample size: 2 | |
| ================================================================================ | |
| Config: max_steps: 2000, lora_r: 64 , lr: 0.0002, bf16: False, lora_modules: all , bits: 4 , full_finetune: False, lora_dropout: 0.0 , warmup_steps: 400 , compress_statistics: NaN , dataset: NaN , gradient_accumulation_steps: NaN , learning_rate: NaN , quant_type: NaN , adam_beta2: NaN , update_freq: 16 | |
| eval_bert_f1 mean (SE): 90.4442 (0.0103). 95% CI (90.424, 90.464). Sample size: 2 | |
| eval_bert_f1 mean (SE): 90.4442 (0.0103). 95% CI (90.424, 90.464). Sample size: 2 | |
| eval_rougeL mean (SE): 64.3317 (0.0305). 95% CI (64.272, 64.391). Sample size: 2 | |
| ================================================================================ | |
| Config: max_steps: 2000, lora_r: 32 , lr: 0.0002, bf16: False, lora_modules: all , bits: 4 , full_finetune: False, lora_dropout: 0.0 , warmup_steps: 100 , compress_statistics: NaN , dataset: NaN , gradient_accumulation_steps: NaN , learning_rate: NaN , quant_type: NaN , adam_beta2: NaN , update_freq: 16 | |
| eval_bert_f1 mean (SE): 90.4460 (0.0177). 95% CI (90.411, 90.481). Sample size: 2 | |
| eval_bert_f1 mean (SE): 90.4460 (0.0177). 95% CI (90.411, 90.481). Sample size: 2 | |
| eval_rougeL mean (SE): 64.3792 (0.0581). 95% CI (64.265, 64.493). Sample size: 2 | |
| ================================================================================ | |
| Config: max_steps: 5000, lora_r: 8 , lr: 0.0002, bf16: False, lora_modules: all , bits: 4 , full_finetune: False, lora_dropout: 0.0 , warmup_steps: 400 , compress_statistics: NaN , dataset: NaN , gradient_accumulation_steps: NaN , learning_rate: NaN , quant_type: NaN , adam_beta2: NaN , update_freq: 8 | |
| eval_bert_f1 mean (SE): 90.4461 (0.0602). 95% CI (90.328, 90.564). Sample size: 2 | |
| eval_bert_f1 mean (SE): 90.4461 (0.0602). 95% CI (90.328, 90.564). Sample size: 2 | |
| eval_rougeL mean (SE): 64.4576 (0.3699). 95% CI (63.732, 65.183). Sample size: 2 | |
| ================================================================================ | |
| Config: max_steps: 2000, lora_r: 64 , lr: 0.0002, bf16: False, lora_modules: all , bits: 4 , full_finetune: False, lora_dropout: 0.0 , warmup_steps: 100 , compress_statistics: NaN , dataset: NaN , gradient_accumulation_steps: NaN , learning_rate: NaN , quant_type: NaN , adam_beta2: NaN , update_freq: 16 | |
| eval_bert_f1 mean (SE): 90.4474 (0.0028). 95% CI (90.442, 90.453). Sample size: 2 | |
| eval_bert_f1 mean (SE): 90.4474 (0.0028). 95% CI (90.442, 90.453). Sample size: 2 | |
| eval_rougeL mean (SE): 64.3734 (0.0645). 95% CI (64.247, 64.500). Sample size: 2 | |
| ================================================================================ | |
| Config: max_steps: 4000, lora_r: 8 , lr: 0.0002, bf16: False, lora_modules: all , bits: 4 , full_finetune: False, lora_dropout: 0.0 , warmup_steps: 400 , compress_statistics: NaN , dataset: NaN , gradient_accumulation_steps: NaN , learning_rate: NaN , quant_type: NaN , adam_beta2: NaN , update_freq: 8 | |
| eval_bert_f1 mean (SE): 90.4494 (0.0366). 95% CI (90.378, 90.521). Sample size: 2 | |
| eval_bert_f1 mean (SE): 90.4494 (0.0366). 95% CI (90.378, 90.521). Sample size: 2 | |
| eval_rougeL mean (SE): 64.5652 (0.0819). 95% CI (64.405, 64.726). Sample size: 2 | |
| ================================================================================ | |
| Config: max_steps: 5333, lora_r: 64 , lr: 0.0001, bf16: False, lora_modules: all , bits: 4 , full_finetune: False, lora_dropout: 0.0 , warmup_steps: 100 , compress_statistics: NaN , dataset: NaN , gradient_accumulation_steps: NaN , learning_rate: NaN , quant_type: NaN , adam_beta2: NaN , update_freq: 6 | |
| eval_bert_f1 mean (SE): 90.4495 (nan). 95% CI (nan, nan). Sample size: 1 | |
| eval_bert_f1 mean (SE): 90.4495 (nan). 95% CI (nan, nan). Sample size: 1 | |
| eval_rougeL mean (SE): 64.2885 (nan). 95% CI (nan, nan). Sample size: 1 | |
| ================================================================================ | |
| Config: max_steps: 2000, lora_r: 16 , lr: 0.0002, bf16: False, lora_modules: all , bits: 4 , full_finetune: False, lora_dropout: 0.0 , warmup_steps: 100 , compress_statistics: NaN , dataset: NaN , gradient_accumulation_steps: NaN , learning_rate: NaN , quant_type: NaN , adam_beta2: NaN , update_freq: 16 | |
| eval_bert_f1 mean (SE): 90.4514 (0.0028). 95% CI (90.446, 90.457). Sample size: 2 | |
| eval_bert_f1 mean (SE): 90.4514 (0.0028). 95% CI (90.446, 90.457). Sample size: 2 | |
| eval_rougeL mean (SE): 64.3890 (0.0216). 95% CI (64.347, 64.431). Sample size: 2 | |
| ================================================================================ | |
| Config: max_steps: 2500, lora_r: 128 , lr: 3e-05, bf16: True, lora_modules: ffn , bits: 16 , full_finetune: True, lora_dropout: 0.0 , warmup_steps: 500 , compress_statistics: NaN , dataset: NaN , gradient_accumulation_steps: NaN , learning_rate: NaN , quant_type: NaN , adam_beta2: NaN , update_freq: 8 | |
| eval_bert_f1 mean (SE): 90.4519 (0.0000). 95% CI (90.452, 90.452). Sample size: 2 | |
| eval_bert_f1 mean (SE): 90.4519 (0.0000). 95% CI (90.452, 90.452). Sample size: 2 | |
| eval_rougeL mean (SE): 64.2359 (0.0000). 95% CI (64.236, 64.236). Sample size: 2 | |
| ================================================================================ | |
| Config: max_steps: 6000, lora_r: 64 , lr: 0.0002, bf16: True, lora_modules: all , bits: 4 , full_finetune: False, lora_dropout: 0.0 , warmup_steps: 100 , compress_statistics: False, dataset: NaN , gradient_accumulation_steps: NaN , learning_rate: NaN , quant_type: fp4 , adam_beta2: 0.95, update_freq: 6 | |
| eval_bert_f1 mean (SE): 90.4548 (0.0459). 95% CI (90.365, 90.545). Sample size: 6 | |
| eval_bert_f1 mean (SE): 90.4548 (0.0459). 95% CI (90.365, 90.545). Sample size: 6 | |
| eval_rougeL mean (SE): 63.4767 (0.4950). 95% CI (62.507, 64.447). Sample size: 6 | |
| ================================================================================ | |
| Config: max_steps: 4000, lora_r: 32 , lr: 0.0002, bf16: False, lora_modules: all , bits: 4 , full_finetune: False, lora_dropout: 0.0 , warmup_steps: 100 , compress_statistics: NaN , dataset: NaN , gradient_accumulation_steps: NaN , learning_rate: NaN , quant_type: NaN , adam_beta2: NaN , update_freq: 8 | |
| eval_bert_f1 mean (SE): 90.4554 (0.0126). 95% CI (90.431, 90.480). Sample size: 2 | |
| eval_bert_f1 mean (SE): 90.4554 (0.0126). 95% CI (90.431, 90.480). Sample size: 2 | |
| eval_rougeL mean (SE): 64.5199 (0.1353). 95% CI (64.255, 64.785). Sample size: 2 | |
| ================================================================================ | |
| Config: max_steps: 4000, lora_r: 64 , lr: 0.0001, bf16: False, lora_modules: all , bits: 4 , full_finetune: False, lora_dropout: 0.0 , warmup_steps: 100 , compress_statistics: NaN , dataset: NaN , gradient_accumulation_steps: NaN , learning_rate: NaN , quant_type: NaN , adam_beta2: NaN , update_freq: 8 | |
| eval_bert_f1 mean (SE): 90.4555 (nan). 95% CI (nan, nan). Sample size: 1 | |
| eval_bert_f1 mean (SE): 90.4555 (nan). 95% CI (nan, nan). Sample size: 1 | |
| eval_rougeL mean (SE): 64.2258 (nan). 95% CI (nan, nan). Sample size: 1 | |
| ================================================================================ | |
| Config: max_steps: 5000, lora_r: 32 , lr: 0.0002, bf16: False, lora_modules: all , bits: 4 , full_finetune: False, lora_dropout: 0.0 , warmup_steps: 100 , compress_statistics: NaN , dataset: NaN , gradient_accumulation_steps: NaN , learning_rate: NaN , quant_type: NaN , adam_beta2: NaN , update_freq: 8 | |
| eval_bert_f1 mean (SE): 90.4585 (0.0002). 95% CI (90.458, 90.459). Sample size: 2 | |
| eval_bert_f1 mean (SE): 90.4585 (0.0002). 95% CI (90.458, 90.459). Sample size: 2 | |
| eval_rougeL mean (SE): 64.8562 (0.0711). 95% CI (64.717, 64.996). Sample size: 2 | |
| ================================================================================ | |
| Config: max_steps: 5000, lora_r: 64 , lr: 0.0002, bf16: False, lora_modules: all , bits: 4 , full_finetune: False, lora_dropout: 0.0 , warmup_steps: 400 , compress_statistics: NaN , dataset: NaN , gradient_accumulation_steps: NaN , learning_rate: NaN , quant_type: NaN , adam_beta2: NaN , update_freq: 8 | |
| eval_bert_f1 mean (SE): 90.4587 (0.0085). 95% CI (90.442, 90.475). Sample size: 2 | |
| eval_bert_f1 mean (SE): 90.4587 (0.0085). 95% CI (90.442, 90.475). Sample size: 2 | |
| eval_rougeL mean (SE): 64.7650 (0.0132). 95% CI (64.739, 64.791). Sample size: 2 | |
| ================================================================================ | |
| Config: max_steps: 2500, lora_r: 128 , lr: 0.0003, bf16: False, lora_modules: all , bits: 4 , full_finetune: False, lora_dropout: 0.05, warmup_steps: NaN , compress_statistics: NaN , dataset: NaN , gradient_accumulation_steps: NaN , learning_rate: NaN , quant_type: NaN , adam_beta2: NaN , update_freq: 8 | |
| eval_bert_f1 mean (SE): 90.4607 (0.0187). 95% CI (90.424, 90.497). Sample size: 4 | |
| eval_bert_f1 mean (SE): 90.4607 (0.0187). 95% CI (90.424, 90.497). Sample size: 4 | |
| eval_rougeL mean (SE): 64.3726 (0.0532). 95% CI (64.268, 64.477). Sample size: 4 | |
| ================================================================================ | |
| Config: max_steps: 4000, lora_r: 64 , lr: 0.0002, bf16: False, lora_modules: all , bits: 4 , full_finetune: False, lora_dropout: 0.0 , warmup_steps: 100 , compress_statistics: NaN , dataset: NaN , gradient_accumulation_steps: NaN , learning_rate: NaN , quant_type: NaN , adam_beta2: NaN , update_freq: 8 | |
| eval_bert_f1 mean (SE): 90.4619 (0.0011). 95% CI (90.460, 90.464). Sample size: 2 | |
| eval_bert_f1 mean (SE): 90.4619 (0.0011). 95% CI (90.460, 90.464). Sample size: 2 | |
| eval_rougeL mean (SE): 64.5099 (0.1572). 95% CI (64.202, 64.818). Sample size: 2 | |
| ================================================================================ | |
| Config: max_steps: 5000, lora_r: 64 , lr: 0.0002, bf16: False, lora_modules: all , bits: 4 , full_finetune: False, lora_dropout: 0.0 , warmup_steps: 100 , compress_statistics: NaN , dataset: NaN , gradient_accumulation_steps: NaN , learning_rate: NaN , quant_type: NaN , adam_beta2: NaN , update_freq: 8 | |
| eval_bert_f1 mean (SE): 90.4624 (0.0199). 95% CI (90.423, 90.501). Sample size: 2 | |
| eval_bert_f1 mean (SE): 90.4624 (0.0199). 95% CI (90.423, 90.501). Sample size: 2 | |
| eval_rougeL mean (SE): 64.8780 (0.0419). 95% CI (64.796, 64.960). Sample size: 2 | |
| ================================================================================ | |
| Config: max_steps: 2000, lora_r: 16 , lr: 0.0002, bf16: False, lora_modules: all , bits: 4 , full_finetune: False, lora_dropout: 0.0 , warmup_steps: 400 , compress_statistics: NaN , dataset: NaN , gradient_accumulation_steps: NaN , learning_rate: NaN , quant_type: NaN , adam_beta2: NaN , update_freq: 16 | |
| eval_bert_f1 mean (SE): 90.4642 (0.0180). 95% CI (90.429, 90.499). Sample size: 2 | |
| eval_bert_f1 mean (SE): 90.4642 (0.0180). 95% CI (90.429, 90.499). Sample size: 2 | |
| eval_rougeL mean (SE): 64.3903 (0.0014). 95% CI (64.387, 64.393). Sample size: 2 | |
| ================================================================================ | |
| Config: max_steps: 2500, lora_r: 128 , lr: 3e-05, bf16: True, lora_modules: ffn , bits: 16 , full_finetune: True, lora_dropout: 0.0 , warmup_steps: 100 , compress_statistics: NaN , dataset: NaN , gradient_accumulation_steps: NaN , learning_rate: NaN , quant_type: NaN , adam_beta2: NaN , update_freq: 8 | |
| eval_bert_f1 mean (SE): 90.4645 (0.0000). 95% CI (90.465, 90.465). Sample size: 2 | |
| eval_bert_f1 mean (SE): 90.4645 (0.0000). 95% CI (90.465, 90.465). Sample size: 2 | |
| eval_rougeL mean (SE): 64.3108 (0.0000). 95% CI (64.311, 64.311). Sample size: 2 | |
| ================================================================================ | |
| Config: max_steps: 4000, lora_r: 64 , lr: 0.0001, bf16: False, lora_modules: all , bits: 4 , full_finetune: False, lora_dropout: 0.0 , warmup_steps: 100 , compress_statistics: NaN , dataset: NaN , gradient_accumulation_steps: NaN , learning_rate: NaN , quant_type: NaN , adam_beta2: NaN , update_freq: 10 | |
| eval_bert_f1 mean (SE): 90.4688 (nan). 95% CI (nan, nan). Sample size: 1 | |
| eval_bert_f1 mean (SE): 90.4688 (nan). 95% CI (nan, nan). Sample size: 1 | |
| eval_rougeL mean (SE): 64.2497 (nan). 95% CI (nan, nan). Sample size: 1 | |
| ================================================================================ | |
| Config: max_steps: 5000, lora_r: 32 , lr: 0.0002, bf16: False, lora_modules: all , bits: 4 , full_finetune: False, lora_dropout: 0.0 , warmup_steps: 400 , compress_statistics: NaN , dataset: NaN , gradient_accumulation_steps: NaN , learning_rate: NaN , quant_type: NaN , adam_beta2: NaN , update_freq: 8 | |
| eval_bert_f1 mean (SE): 90.4763 (0.0249). 95% CI (90.427, 90.525). Sample size: 2 | |
| eval_bert_f1 mean (SE): 90.4763 (0.0249). 95% CI (90.427, 90.525). Sample size: 2 | |
| eval_rougeL mean (SE): 64.7678 (0.0439). 95% CI (64.682, 64.854). Sample size: 2 | |
| ================================================================================ | |
| Config: max_steps: 5000, lora_r: 16 , lr: 0.0002, bf16: False, lora_modules: all , bits: 4 , full_finetune: False, lora_dropout: 0.0 , warmup_steps: 100 , compress_statistics: NaN , dataset: NaN , gradient_accumulation_steps: NaN , learning_rate: NaN , quant_type: NaN , adam_beta2: NaN , update_freq: 8 | |
| eval_bert_f1 mean (SE): 90.4771 (0.0306). 95% CI (90.417, 90.537). Sample size: 2 | |
| eval_bert_f1 mean (SE): 90.4771 (0.0306). 95% CI (90.417, 90.537). Sample size: 2 | |
| eval_rougeL mean (SE): 64.8577 (0.0651). 95% CI (64.730, 64.985). Sample size: 2 | |
| ================================================================================ | |
| Config: max_steps: 4000, lora_r: 16 , lr: 0.0002, bf16: False, lora_modules: all , bits: 4 , full_finetune: False, lora_dropout: 0.0 , warmup_steps: 100 , compress_statistics: NaN , dataset: NaN , gradient_accumulation_steps: NaN , learning_rate: NaN , quant_type: NaN , adam_beta2: NaN , update_freq: 8 | |
| eval_bert_f1 mean (SE): 90.4778 (0.0201). 95% CI (90.438, 90.517). Sample size: 2 | |
| eval_bert_f1 mean (SE): 90.4778 (0.0201). 95% CI (90.438, 90.517). Sample size: 2 | |
| eval_rougeL mean (SE): 64.6057 (0.0300). 95% CI (64.547, 64.664). Sample size: 2 | |
| ================================================================================ | |
| Config: max_steps: 2500, lora_r: 64 , lr: 0.0003, bf16: False, lora_modules: all , bits: 4 , full_finetune: False, lora_dropout: 0.05, warmup_steps: NaN , compress_statistics: NaN , dataset: NaN , gradient_accumulation_steps: NaN , learning_rate: NaN , quant_type: NaN , adam_beta2: NaN , update_freq: 8 | |
| eval_bert_f1 mean (SE): 90.4781 (0.0000). 95% CI (90.478, 90.478). Sample size: 2 | |
| eval_bert_f1 mean (SE): 90.4781 (0.0000). 95% CI (90.478, 90.478). Sample size: 2 | |
| eval_rougeL mean (SE): 64.3865 (0.0000). 95% CI (64.386, 64.386). Sample size: 2 | |
| ================================================================================ | |
| Config: max_steps: 5000, lora_r: 64 , lr: 0.0003, bf16: False, lora_modules: all , bits: 4 , full_finetune: False, lora_dropout: 0.05, warmup_steps: NaN , compress_statistics: NaN , dataset: NaN , gradient_accumulation_steps: NaN , learning_rate: NaN , quant_type: NaN , adam_beta2: NaN , update_freq: 4 | |
| eval_bert_f1 mean (SE): 90.4827 (0.0000). 95% CI (90.483, 90.483). Sample size: 2 | |
| eval_bert_f1 mean (SE): 90.4827 (0.0000). 95% CI (90.483, 90.483). Sample size: 2 | |
| eval_rougeL mean (SE): 64.5578 (0.0000). 95% CI (64.558, 64.558). Sample size: 2 | |
| ================================================================================ | |
| Config: max_steps: 2500, lora_r: 128 , lr: 0.001, bf16: False, lora_modules: ffn , bits: 4 , full_finetune: False, lora_dropout: 0.0 , warmup_steps: NaN , compress_statistics: NaN , dataset: NaN , gradient_accumulation_steps: NaN , learning_rate: NaN , quant_type: NaN , adam_beta2: NaN , update_freq: 8 | |
| eval_bert_f1 mean (SE): 90.4867 (0.0000). 95% CI (90.487, 90.487). Sample size: 2 | |
| eval_bert_f1 mean (SE): 90.4867 (0.0000). 95% CI (90.487, 90.487). Sample size: 2 | |
| eval_rougeL mean (SE): 64.5968 (0.0000). 95% CI (64.597, 64.597). Sample size: 2 | |
| ================================================================================ | |
| Config: max_steps: 6000, lora_r: 64 , lr: 0.0002, bf16: True, lora_modules: all , bits: 4 , full_finetune: False, lora_dropout: 0.0 , warmup_steps: 100 , compress_statistics: False, dataset: NaN , gradient_accumulation_steps: NaN , learning_rate: NaN , quant_type: fp4 , adam_beta2: 0.999, update_freq: 6 | |
| eval_bert_f1 mean (SE): 90.4872 (0.0359). 95% CI (90.417, 90.558). Sample size: 6 | |
| eval_bert_f1 mean (SE): 90.4872 (0.0359). 95% CI (90.417, 90.558). Sample size: 6 | |
| eval_rougeL mean (SE): 63.6382 (0.5197). 95% CI (62.620, 64.657). Sample size: 6 | |
| ================================================================================ | |
| Config: max_steps: 6000, lora_r: 64 , lr: 0.0002, bf16: True, lora_modules: all , bits: 4 , full_finetune: False, lora_dropout: 0.0 , warmup_steps: 100 , compress_statistics: True, dataset: NaN , gradient_accumulation_steps: NaN , learning_rate: NaN , quant_type: fp4 , adam_beta2: 0.95, update_freq: 6 | |
| eval_bert_f1 mean (SE): 90.4912 (0.0526). 95% CI (90.388, 90.594). Sample size: 6 | |
| eval_bert_f1 mean (SE): 90.4912 (0.0526). 95% CI (90.388, 90.594). Sample size: 6 | |
| eval_rougeL mean (SE): 63.6371 (0.5744). 95% CI (62.511, 64.763). Sample size: 6 | |
| ================================================================================ | |
| Config: max_steps: 17500, lora_r: 64 , lr: 0.0002, bf16: True, lora_modules: all , bits: 4 , full_finetune: False, lora_dropout: 0.0 , warmup_steps: 100 , compress_statistics: False, dataset: NaN , gradient_accumulation_steps: NaN , learning_rate: NaN , quant_type: fp4 , adam_beta2: 0.95, update_freq: 6 | |
| eval_bert_f1 mean (SE): 90.4916 (0.0422). 95% CI (90.409, 90.574). Sample size: 4 | |
| eval_bert_f1 mean (SE): 90.4916 (0.0422). 95% CI (90.409, 90.574). Sample size: 4 | |
| eval_rougeL mean (SE): 65.0195 (0.4506). 95% CI (64.136, 65.903). Sample size: 4 | |
| ================================================================================ | |
| Config: max_steps: 2500, lora_r: 128 , lr: 0.0003, bf16: False, lora_modules: all , bits: 4 , full_finetune: False, lora_dropout: 0.0 , warmup_steps: NaN , compress_statistics: NaN , dataset: NaN , gradient_accumulation_steps: NaN , learning_rate: NaN , quant_type: NaN , adam_beta2: NaN , update_freq: 8 | |
| eval_bert_f1 mean (SE): 90.5034 (0.0000). 95% CI (90.503, 90.503). Sample size: 2 | |
| eval_bert_f1 mean (SE): 90.5034 (0.0000). 95% CI (90.503, 90.503). Sample size: 2 | |
| eval_rougeL mean (SE): 64.5204 (0.0000). 95% CI (64.520, 64.520). Sample size: 2 | |
| ================================================================================ | |
| Config: max_steps: 6667, lora_r: 64 , lr: 0.0001, bf16: False, lora_modules: all , bits: 4 , full_finetune: False, lora_dropout: 0.0 , warmup_steps: 100 , compress_statistics: NaN , dataset: NaN , gradient_accumulation_steps: NaN , learning_rate: NaN , quant_type: NaN , adam_beta2: NaN , update_freq: 6 | |
| eval_bert_f1 mean (SE): 90.5073 (nan). 95% CI (nan, nan). Sample size: 1 | |
| eval_bert_f1 mean (SE): 90.5073 (nan). 95% CI (nan, nan). Sample size: 1 | |
| eval_rougeL mean (SE): 64.4619 (nan). 95% CI (nan, nan). Sample size: 1 | |
| ================================================================================ | |
| Config: max_steps: 5000, lora_r: 64 , lr: 0.0003, bf16: False, lora_modules: all , bits: 4 , full_finetune: False, lora_dropout: 0.0 , warmup_steps: NaN , compress_statistics: NaN , dataset: NaN , gradient_accumulation_steps: NaN , learning_rate: NaN , quant_type: NaN , adam_beta2: NaN , update_freq: 4 | |
| eval_bert_f1 mean (SE): 90.5197 (0.0000). 95% CI (90.520, 90.520). Sample size: 2 | |
| eval_bert_f1 mean (SE): 90.5197 (0.0000). 95% CI (90.520, 90.520). Sample size: 2 | |
| eval_rougeL mean (SE): 64.6247 (0.0000). 95% CI (64.625, 64.625). Sample size: 2 | |
| ================================================================================ | |
| Config: max_steps: 5000, lora_r: 64 , lr: 0.0001, bf16: False, lora_modules: all , bits: 4 , full_finetune: False, lora_dropout: 0.0 , warmup_steps: 100 , compress_statistics: NaN , dataset: NaN , gradient_accumulation_steps: NaN , learning_rate: NaN , quant_type: NaN , adam_beta2: NaN , update_freq: 8 | |
| eval_bert_f1 mean (SE): 90.5214 (nan). 95% CI (nan, nan). Sample size: 1 | |
| eval_bert_f1 mean (SE): 90.5214 (nan). 95% CI (nan, nan). Sample size: 1 | |
| eval_rougeL mean (SE): 64.3744 (nan). 95% CI (nan, nan). Sample size: 1 | |
| ================================================================================ | |
| Config: max_steps: 5000, lora_r: 16 , lr: 0.0002, bf16: False, lora_modules: all , bits: 4 , full_finetune: False, lora_dropout: 0.0 , warmup_steps: 400 , compress_statistics: NaN , dataset: NaN , gradient_accumulation_steps: NaN , learning_rate: NaN , quant_type: NaN , adam_beta2: NaN , update_freq: 8 | |
| eval_bert_f1 mean (SE): 90.5233 (0.0255). 95% CI (90.473, 90.573). Sample size: 2 | |
| eval_bert_f1 mean (SE): 90.5233 (0.0255). 95% CI (90.473, 90.573). Sample size: 2 | |
| eval_rougeL mean (SE): 64.8376 (0.0045). 95% CI (64.829, 64.846). Sample size: 2 | |
| ================================================================================ | |
| Config: max_steps: 45550, lora_r: 64 , lr: 0.0002, bf16: False, lora_modules: all , bits: 4 , full_finetune: False, lora_dropout: 0.0 , warmup_steps: 100 , compress_statistics: NaN , dataset: NaN , gradient_accumulation_steps: NaN , learning_rate: NaN , quant_type: NaN , adam_beta2: NaN , update_freq: 6 | |
| eval_bert_f1 mean (SE): 90.5838 (nan). 95% CI (nan, nan). Sample size: 1 | |
| eval_bert_f1 mean (SE): 90.5838 (nan). 95% CI (nan, nan). Sample size: 1 | |
| eval_rougeL mean (SE): 64.8241 (nan). 95% CI (nan, nan). Sample size: 1 | |
| ================================================================================ | |
| Config: max_steps: 6000, lora_r: 64 , lr: 0.0002, bf16: False, lora_modules: all , bits: 4 , full_finetune: False, lora_dropout: 0.0 , warmup_steps: 100 , compress_statistics: NaN , dataset: NaN , gradient_accumulation_steps: NaN , learning_rate: NaN , quant_type: NaN , adam_beta2: NaN , update_freq: 6 | |
| eval_bert_f1 mean (SE): 90.6995 (nan). 95% CI (nan, nan). Sample size: 1 | |
| eval_bert_f1 mean (SE): 90.6995 (nan). 95% CI (nan, nan). Sample size: 1 | |
| eval_rougeL mean (SE): 65.2213 (nan). 95% CI (nan, nan). Sample size: 1 | |
| ================================================================================ | |
| Config: max_steps: 30375, lora_r: 64 , lr: 0.0002, bf16: False, lora_modules: all , bits: 4 , full_finetune: False, lora_dropout: 0.0 , warmup_steps: 100 , compress_statistics: NaN , dataset: NaN , gradient_accumulation_steps: NaN , learning_rate: NaN , quant_type: NaN , adam_beta2: NaN , update_freq: 6 | |
| eval_bert_f1 mean (SE): 90.7717 (nan). 95% CI (nan, nan). Sample size: 1 | |
| eval_bert_f1 mean (SE): 90.7717 (nan). 95% CI (nan, nan). Sample size: 1 | |
| eval_rougeL mean (SE): 65.4644 (nan). 95% CI (nan, nan). Sample size: 1 | |
| ================================================================================ | |
| Config: max_steps: 9000, lora_r: 64 , lr: 0.0002, bf16: False, lora_modules: all , bits: 4 , full_finetune: False, lora_dropout: 0.0 , warmup_steps: 100 , compress_statistics: NaN , dataset: NaN , gradient_accumulation_steps: NaN , learning_rate: NaN , quant_type: NaN , adam_beta2: NaN , update_freq: 6 | |
| eval_bert_f1 mean (SE): 90.7842 (nan). 95% CI (nan, nan). Sample size: 1 | |
| eval_bert_f1 mean (SE): 90.7842 (nan). 95% CI (nan, nan). Sample size: 1 | |
| eval_rougeL mean (SE): 65.6898 (nan). 95% CI (nan, nan). Sample size: 1 | |
| ================================================================================ | |
| Config: max_steps: 13500, lora_r: 64 , lr: 0.0002, bf16: False, lora_modules: all , bits: 4 , full_finetune: False, lora_dropout: 0.0 , warmup_steps: 100 , compress_statistics: NaN , dataset: NaN , gradient_accumulation_steps: NaN , learning_rate: NaN , quant_type: NaN , adam_beta2: NaN , update_freq: 6 | |
| eval_bert_f1 mean (SE): 90.8314 (nan). 95% CI (nan, nan). Sample size: 1 | |
| eval_bert_f1 mean (SE): 90.8314 (nan). 95% CI (nan, nan). Sample size: 1 | |
| eval_rougeL mean (SE): 66.1775 (nan). 95% CI (nan, nan). Sample size: 1 | |
| ================================================================================ | |
| Config: max_steps: 20250, lora_r: 64 , lr: 0.0002, bf16: False, lora_modules: all , bits: 4 , full_finetune: False, lora_dropout: 0.0 , warmup_steps: 100 , compress_statistics: NaN , dataset: NaN , gradient_accumulation_steps: NaN , learning_rate: NaN , quant_type: NaN , adam_beta2: NaN , update_freq: 6 | |
| eval_bert_f1 mean (SE): 90.8519 (nan). 95% CI (nan, nan). Sample size: 1 | |
| eval_bert_f1 mean (SE): 90.8519 (nan). 95% CI (nan, nan). Sample size: 1 | |
| eval_rougeL mean (SE): 66.4437 (nan). 95% CI (nan, nan). Sample size: 1 | |
| ================================================================================ |
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment