gptoss_20b_all_zhtw_lr5e-7_ep3_16_64_128_turn_v5data_500

This model is a fine-tuned version of openai/gpt-oss-20b on the multi_turn_miss_func_zh_tw_function_mix_v5_500_turn_oss_gpt_oss_20b_pretokenized, the multi_turn_miss_param_zh_tw_function_mix_v5_500_turn_oss_gpt_oss_20b_pretokenized, the multi_turn_zh_tw_function_mix_v5_500_turn_oss_gpt_oss_20b_pretokenized, the new_irrelevance_zh_tw_oss3000_oss_gpt_oss_20b_pretokenized and the apigen_zhtwV3_remove_sys_gpt_oss_20b_pretokenized datasets. It achieves the following results on the evaluation set:

• Loss: 1.0354

Model description

More information needed

Intended uses & limitations

More information needed

Training and evaluation data

More information needed

Training procedure

Training hyperparameters

The following hyperparameters were used during training:

• learning_rate: 5e-07
• train_batch_size: 2
• eval_batch_size: 4
• seed: 42
• gradient_accumulation_steps: 64
• total_train_batch_size: 128
• optimizer: Use OptimizerNames.ADAMW_TORCH_FUSED with betas=(0.9,0.999) and epsilon=1e-08 and optimizer_args=No additional optimizer arguments
• lr_scheduler_type: cosine
• lr_scheduler_warmup_steps: 15
• num_epochs: 3

Training results

Training Loss	Epoch	Step	Validation Loss
1.5138	0.0435	10	1.7372
1.8616	0.0870	20	1.7316
1.5077	0.1305	30	1.7197
1.5093	0.1739	40	1.6988
1.5221	0.2174	50	1.6783
1.4645	0.2609	60	1.6571
1.4865	0.3044	70	1.6276
1.24	0.3479	80	1.6035
1.482	0.3914	90	1.5790
1.3486	0.4349	100	1.5459
1.4871	0.4784	110	1.5275
1.3186	0.5218	120	1.5016
1.3892	0.5653	130	1.4813
1.4606	0.6088	140	1.4613
1.2964	0.6523	150	1.4345
1.2039	0.6958	160	1.4139
1.1981	0.7393	170	1.3928
1.2071	0.7828	180	1.3773
1.2505	0.8263	190	1.3566
1.241	0.8697	200	1.3361
1.4614	0.9132	210	1.3231
1.1201	0.9567	220	1.3060
1.1996	1.0	230	1.2896
1.1315	1.0435	240	1.2759
1.182	1.0870	250	1.2652
1.1442	1.1305	260	1.2473
1.1309	1.1739	270	1.2370
0.9868	1.2174	280	1.2237
1.1573	1.2609	290	1.2061
1.1479	1.3044	300	1.2005
1.1099	1.3479	310	1.1852
1.1088	1.3914	320	1.1753
1.0451	1.4349	330	1.1658
1.0777	1.4784	340	1.1505
1.0902	1.5218	350	1.1406
1.0681	1.5653	360	1.1321
1.0797	1.6088	370	1.1250
1.1503	1.6523	380	1.1155
1.1035	1.6958	390	1.1087
1.0596	1.7393	400	1.0997
1.0397	1.7828	410	1.0939
1.1203	1.8263	420	1.0878
0.9642	1.8697	430	1.0845
1.0697	1.9132	440	1.0780
0.9721	1.9567	450	1.0719
1.0128	2.0	460	1.0680
1.0888	2.0435	470	1.0658
1.0267	2.0870	480	1.0626
1.0126	2.1305	490	1.0589
0.9058	2.1739	500	1.0564
1.031	2.2174	510	1.0544
1.0226	2.2609	520	1.0503
1.1264	2.3044	530	1.0485
1.0121	2.3479	540	1.0477
1.0678	2.3914	550	1.0450
1.0438	2.4349	560	1.0406
1.1264	2.4784	570	1.0405
1.0563	2.5218	580	1.0410
1.0601	2.5653	590	1.0386
0.9677	2.6088	600	1.0440
1.1078	2.6523	610	1.0392
1.0677	2.6958	620	1.0387
1.0246	2.7393	630	1.0376
1.0707	2.7828	640	1.0387
0.9897	2.8263	650	1.0386
0.9436	2.8697	660	1.0377
1.1909	2.9132	670	1.0375
0.9286	2.9567	680	1.0350
1.0323	3.0	690	1.0354

Framework versions

• PEFT 0.18.1
• Transformers 4.57.6
• Pytorch 2.8.0+cu128
• Datasets 4.0.0
• Tokenizers 0.22.2