tsavage68/Hand_off_DS_Llama8B_1000steps_1e6rate_SFT

final_checkpoint

This model is a fine-tuned version of deepseek-ai/DeepSeek-R1-Distill-Llama-8B on an unknown dataset. It achieves the following results on the evaluation set:

• Loss: 0.4805

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: 1e-06
• train_batch_size: 2
• eval_batch_size: 2
• seed: 42
• gradient_accumulation_steps: 2
• total_train_batch_size: 4
• optimizer: Use adafactor and the args are: No additional optimizer arguments
• lr_scheduler_type: cosine
• lr_scheduler_warmup_steps: 100
• training_steps: 1000

Training results

Training Loss	Epoch	Step	Validation Loss
2.6773	0.0464	50	2.4460
0.7309	0.0929	100	0.7180
0.5965	0.1393	150	0.6000
0.5401	0.1857	200	0.5535
0.5055	0.2321	250	0.5285
0.4901	0.2786	300	0.5148
0.5266	0.3250	350	0.5047
0.4829	0.3714	400	0.4973
0.4825	0.4178	450	0.4922
0.508	0.4643	500	0.4886
0.503	0.5107	550	0.4858
0.514	0.5571	600	0.4835
0.492	0.6035	650	0.4822
0.4743	0.6500	700	0.4814
0.4942	0.6964	750	0.4809
0.4811	0.7428	800	0.4806
0.4645	0.7892	850	0.4805
0.4673	0.8357	900	0.4805
0.4933	0.8821	950	0.4805
0.5026	0.9285	1000	0.4805

Framework versions

• Transformers 4.49.0
• Pytorch 2.6.0+cu124
• Datasets 3.3.1
• Tokenizers 0.21.0