RefalMachine/ruadapt_solar_10.7_darulm_unigram_proj_init_part2_v5

solar_10.7_darulm_unigram_proj_init_darulm_part2_r128_a512_v3_1.0_512_28_02_24

This model is a fine-tuned version of outputs/solar_10.7_darulm_unigram_proj_init_8node_darulm_part1_v3_1.0_512_12_02_24 on the None dataset. It achieves the following results on the evaluation set:

• Loss: 2.2309
• Accuracy: 0.5309

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-05
• train_batch_size: 1
• eval_batch_size: 1
• seed: 42
• distributed_type: multi-GPU
• num_devices: 24
• gradient_accumulation_steps: 5
• total_train_batch_size: 120
• total_eval_batch_size: 24
• optimizer: Adam with betas=(0.9,0.95) and epsilon=1e-05
• lr_scheduler_type: linear
• num_epochs: 1.0
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Accuracy
No log	0.0	1	2.3534	0.5148
2.4427	0.01	500	2.3338	0.5155
2.4399	0.02	1000	2.3276	0.5164
2.4244	0.03	1500	2.3231	0.5169
2.4336	0.04	2000	2.3194	0.5177
2.4201	0.06	2500	2.3156	0.5180
2.4245	0.07	3000	2.3128	0.5185
2.4157	0.08	3500	2.3097	0.5187
2.4054	0.09	4000	2.3070	0.5194
2.4161	0.1	4500	2.3033	0.5197
2.395	0.11	5000	2.3020	0.5201
2.4037	0.12	5500	2.3001	0.5204
2.4188	0.13	6000	2.2977	0.5206
2.406	0.15	6500	2.2961	0.5208
2.4022	0.16	7000	2.2943	0.5210
2.3952	0.17	7500	2.2926	0.5217
2.394	0.18	8000	2.2909	0.5217
2.3828	0.19	8500	2.2891	0.5218
2.3903	0.2	9000	2.2882	0.5223
2.3943	0.21	9500	2.2861	0.5224
2.3944	0.22	10000	2.2851	0.5224
2.3872	0.23	10500	2.2841	0.5227
2.381	0.25	11000	2.2820	0.5228
2.3832	0.26	11500	2.2798	0.5232
2.3813	0.27	12000	2.2793	0.5237
2.3715	0.28	12500	2.2779	0.5241
2.3898	0.29	13000	2.2764	0.5240
2.3717	0.3	13500	2.2757	0.5240
2.3745	0.31	14000	2.2742	0.5244
2.3657	0.32	14500	2.2732	0.5244
2.3782	0.34	15000	2.2715	0.5247
2.3761	0.35	15500	2.2706	0.5247
2.3827	0.36	16000	2.2692	0.5249
2.3659	0.37	16500	2.2678	0.5251
2.3551	0.38	17000	2.2674	0.5252
2.3605	0.39	17500	2.2662	0.5255
2.3579	0.4	18000	2.2654	0.5256
2.361	0.41	18500	2.2642	0.5257
2.3632	0.42	19000	2.2652	0.5254
2.3409	0.44	19500	2.2625	0.5261
2.3546	0.45	20000	2.2631	0.5259
2.361	0.46	20500	2.2611	0.5264
2.355	0.47	21000	2.2598	0.5264
2.3599	0.48	21500	2.2588	0.5265
2.3554	0.49	22000	2.2583	0.5265
2.3552	0.5	22500	2.2571	0.5268
2.3574	0.51	23000	2.2565	0.5268
2.3527	0.53	23500	2.2557	0.5272
2.3574	0.54	24000	2.2548	0.5272
2.3395	0.55	24500	2.2534	0.5274
2.3517	0.56	25000	2.2531	0.5272
2.346	0.57	25500	2.2521	0.5275
2.3469	0.58	26000	2.2515	0.5275
2.3451	0.59	26500	2.2509	0.5278
2.3373	0.6	27000	2.2501	0.5277
2.3512	0.61	27500	2.2493	0.5281
2.3351	0.63	28000	2.2485	0.5282
2.3431	0.64	28500	2.2476	0.5282
2.3399	0.65	29000	2.2463	0.5283
2.3376	0.66	29500	2.2463	0.5284
2.3574	0.67	30000	2.2456	0.5285
2.3312	0.68	30500	2.2447	0.5289
2.3442	0.69	31000	2.2442	0.5288
2.338	0.7	31500	2.2434	0.5289
2.3345	0.72	32000	2.2433	0.5291
2.3314	0.73	32500	2.2420	0.5292
2.326	0.74	33000	2.2414	0.5293
2.3247	0.75	33500	2.2409	0.5295
2.3363	0.76	34000	2.2403	0.5296
2.3409	0.77	34500	2.2395	0.5297
2.335	0.78	35000	2.2391	0.5295
2.3194	0.79	35500	2.2383	0.5298
2.3367	0.8	36000	2.2379	0.5301
2.3286	0.82	36500	2.2372	0.5301
2.3225	0.83	37000	2.2366	0.5302
2.3198	0.84	37500	2.2363	0.5301
2.3274	0.85	38000	2.2355	0.5301
2.3195	0.86	38500	2.2349	0.5303
2.3418	0.87	39000	2.2344	0.5303
2.323	0.88	39500	2.2340	0.5304
2.3211	0.89	40000	2.2336	0.5304
2.3332	0.91	40500	2.2334	0.5306
2.3226	0.92	41000	2.2329	0.5307
2.3329	0.93	41500	2.2325	0.5308
2.3172	0.94	42000	2.2321	0.5307
2.3231	0.95	42500	2.2319	0.5308
2.314	0.96	43000	2.2316	0.5309
2.3205	0.97	43500	2.2315	0.5308
2.3208	0.98	44000	2.2312	0.5309
2.3228	0.99	44500	2.2310	0.5309

Framework versions

• Transformers 4.37.2
• Pytorch 2.1.2
• Datasets 2.16.1
• Tokenizers 0.15.2

Training procedure

Framework versions

• PEFT 0.6.0