Qwen3-30B-A3B-Base-alpaca-th-52k-dolly-th-15k-wangchan-instruct-seed-2405

This model is a fine-tuned version of Qwen/Qwen3-30B-A3B-Base on the alpaca-th-52k, the dolly-th-15k and the wangchan-instruct datasets. It achieves the following results on the evaluation set:

• Loss: 0.6619

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: 0.0002
• train_batch_size: 2
• eval_batch_size: 2
• seed: 2405
• distributed_type: multi-GPU
• num_devices: 128
• gradient_accumulation_steps: 8
• total_train_batch_size: 2048
• total_eval_batch_size: 256
• optimizer: Use OptimizerNames.ADAMW_TORCH with betas=(0.9,0.999) and epsilon=1e-08 and optimizer_args=No additional optimizer arguments
• lr_scheduler_type: cosine
• lr_scheduler_warmup_ratio: 0.1
• num_epochs: 3.0

Training results

Training Loss	Epoch	Step	Validation Loss
0.8866	0.2299	10	1.0011
0.7869	0.4598	20	0.8175
0.724	0.6897	30	0.7251
0.702	0.9195	40	0.7024
0.6926	1.1379	50	0.6904
0.6795	1.3678	60	0.6821
0.6694	1.5977	70	0.6757
0.674	1.8276	80	0.6708
0.6543	2.0460	90	0.6660
0.6492	2.2759	100	0.6639
0.6508	2.5057	110	0.6626
0.656	2.7356	120	0.6621
0.6492	2.9655	130	0.6619

Framework versions

• PEFT 0.15.2
• Transformers 4.52.3
• Pytorch 2.7.0+cu126
• Datasets 3.6.0
• Tokenizers 0.21.1