tinyllama_moe_sft_routeraux_ep3

This model is a fine-tuned version of ondevicellm/tinyllama_moe_v2 on the HuggingFaceH4/ultrachat_200k dataset. It achieves the following results on the evaluation set:

• Loss: 1.2954

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: 2e-05
• train_batch_size: 16
• eval_batch_size: 8
• seed: 42
• distributed_type: multi-GPU
• num_devices: 4
• gradient_accumulation_steps: 2
• total_train_batch_size: 128
• total_eval_batch_size: 32
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: cosine
• lr_scheduler_warmup_steps: 120
• num_epochs: 3

Training results

Training Loss	Epoch	Step	Validation Loss
1.72	0.09	100	1.6775
1.4985	0.18	200	1.4900
1.4482	0.26	300	1.4473
1.4152	0.35	400	1.4215
1.3777	0.44	500	1.4031
1.3932	0.53	600	1.3886
1.375	0.61	700	1.3762
1.3574	0.7	800	1.3657
1.349	0.79	900	1.3563
1.3276	0.88	1000	1.3481
1.3491	0.96	1100	1.3409
1.2812	1.05	1200	1.3358
1.2831	1.14	1300	1.3308
1.2917	1.23	1400	1.3258
1.2812	1.31	1500	1.3219
1.2819	1.4	1600	1.3178
1.2756	1.49	1700	1.3145
1.2584	1.58	1800	1.3107
1.2806	1.66	1900	1.3083
1.2815	1.75	2000	1.3054
1.2676	1.84	2100	1.3031
1.2388	1.93	2200	1.3011
1.2385	2.01	2300	1.3015
1.2459	2.1	2400	1.3000
1.2349	2.19	2500	1.2989
1.2277	2.28	2600	1.2981
1.2243	2.37	2700	1.2973
1.2298	2.45	2800	1.2967
1.2362	2.54	2900	1.2961
1.216	2.63	3000	1.2958
1.2381	2.72	3100	1.2957
1.2274	2.8	3200	1.2955
1.2235	2.89	3300	1.2954
1.2438	2.98	3400	1.2954

Framework versions

• Transformers 4.37.0
• Pytorch 2.1.2+cu118
• Datasets 2.16.1
• Tokenizers 0.15.0