oliverchang/Affine-FineTuneKimi-5GNi9WnuGCnrpssb1BT5VgzseXSdijWwgS9MZ7PHGTr3TBDS

Kimi-K2.5-Affine

This model is a post-trained version of moonshotai/Kimi-K2.5. It achieves the following results on the evaluation set:

• Loss: 0.4547

Training results

Step	Training Loss	Eval Loss
0	0.8234	0.7000
500	0.6187	0.5842
1000	0.5523	0.5301
1500	0.5194	0.5087
2000	0.4981	0.4923
2500	0.4836	0.4814
3000	0.4752	0.4761
3500	0.4698	0.4703
4000	0.4610	0.4652
4500	0.4573	0.4641
5000	0.4491	0.4598
5500	0.4432	0.4587
6000	0.4389	0.4572
6500	0.4351	0.4563
7000	0.4318	0.4558
7500	0.4297	0.4551
8000	0.4280	0.4547

Framework versions

• Transformers 4.57.1
• PyTorch 2.6.0+cu124
• Safetensors 0.5.3