1255 lines
34 KiB
JSON
1255 lines
34 KiB
JSON
{
|
|
"best_global_step": null,
|
|
"best_metric": null,
|
|
"best_model_checkpoint": null,
|
|
"epoch": 1.6,
|
|
"eval_steps": 100,
|
|
"global_step": 1000,
|
|
"is_hyper_param_search": false,
|
|
"is_local_process_zero": true,
|
|
"is_world_process_zero": true,
|
|
"log_history": [
|
|
{
|
|
"entropy": 0.6421742392703891,
|
|
"epoch": 0.016,
|
|
"grad_norm": 4.4375,
|
|
"learning_rate": 1.8e-07,
|
|
"loss": 0.953,
|
|
"mean_token_accuracy": 0.792544937506318,
|
|
"num_tokens": 282849.0,
|
|
"step": 10
|
|
},
|
|
{
|
|
"entropy": 0.6719903893768787,
|
|
"epoch": 0.032,
|
|
"grad_norm": 4.25,
|
|
"learning_rate": 3.8e-07,
|
|
"loss": 0.9758,
|
|
"mean_token_accuracy": 0.7857426963746548,
|
|
"num_tokens": 561726.0,
|
|
"step": 20
|
|
},
|
|
{
|
|
"entropy": 0.6603780068457127,
|
|
"epoch": 0.048,
|
|
"grad_norm": 4.40625,
|
|
"learning_rate": 5.800000000000001e-07,
|
|
"loss": 0.9582,
|
|
"mean_token_accuracy": 0.7894285101443529,
|
|
"num_tokens": 842092.0,
|
|
"step": 30
|
|
},
|
|
{
|
|
"entropy": 0.6775440879166126,
|
|
"epoch": 0.064,
|
|
"grad_norm": 3.890625,
|
|
"learning_rate": 7.8e-07,
|
|
"loss": 0.9776,
|
|
"mean_token_accuracy": 0.785004175081849,
|
|
"num_tokens": 1133349.0,
|
|
"step": 40
|
|
},
|
|
{
|
|
"entropy": 0.6926743468269706,
|
|
"epoch": 0.08,
|
|
"grad_norm": 4.6875,
|
|
"learning_rate": 9.800000000000001e-07,
|
|
"loss": 0.9497,
|
|
"mean_token_accuracy": 0.7896979171782732,
|
|
"num_tokens": 1422340.0,
|
|
"step": 50
|
|
},
|
|
{
|
|
"entropy": 0.7043748194351792,
|
|
"epoch": 0.096,
|
|
"grad_norm": 3.484375,
|
|
"learning_rate": 1.1800000000000001e-06,
|
|
"loss": 0.9217,
|
|
"mean_token_accuracy": 0.7949878334999084,
|
|
"num_tokens": 1711821.0,
|
|
"step": 60
|
|
},
|
|
{
|
|
"entropy": 0.7551965553313493,
|
|
"epoch": 0.112,
|
|
"grad_norm": 2.71875,
|
|
"learning_rate": 1.3800000000000001e-06,
|
|
"loss": 0.931,
|
|
"mean_token_accuracy": 0.7912098452448845,
|
|
"num_tokens": 1991855.0,
|
|
"step": 70
|
|
},
|
|
{
|
|
"entropy": 0.7913430579006672,
|
|
"epoch": 0.128,
|
|
"grad_norm": 2.359375,
|
|
"learning_rate": 1.5800000000000001e-06,
|
|
"loss": 0.9172,
|
|
"mean_token_accuracy": 0.7935858864337206,
|
|
"num_tokens": 2288720.0,
|
|
"step": 80
|
|
},
|
|
{
|
|
"entropy": 0.7881425807252527,
|
|
"epoch": 0.144,
|
|
"grad_norm": 2.125,
|
|
"learning_rate": 1.7800000000000001e-06,
|
|
"loss": 0.917,
|
|
"mean_token_accuracy": 0.7932404045015573,
|
|
"num_tokens": 2564352.0,
|
|
"step": 90
|
|
},
|
|
{
|
|
"entropy": 0.7627748234197498,
|
|
"epoch": 0.16,
|
|
"grad_norm": 2.125,
|
|
"learning_rate": 1.98e-06,
|
|
"loss": 0.9014,
|
|
"mean_token_accuracy": 0.7970093633979559,
|
|
"num_tokens": 2841876.0,
|
|
"step": 100
|
|
},
|
|
{
|
|
"epoch": 0.16,
|
|
"eval_math_entropy": 0.7972113332748413,
|
|
"eval_math_loss": 0.9417253136634827,
|
|
"eval_math_mean_token_accuracy": 0.7907100429534912,
|
|
"eval_math_num_tokens": 2841876.0,
|
|
"eval_math_runtime": 35.8728,
|
|
"eval_math_samples_per_second": 13.938,
|
|
"eval_math_steps_per_second": 3.485,
|
|
"step": 100
|
|
},
|
|
{
|
|
"epoch": 0.16,
|
|
"eval_physics_entropy": 0.7183168392181396,
|
|
"eval_physics_loss": 0.7405859231948853,
|
|
"eval_physics_mean_token_accuracy": 0.822186062335968,
|
|
"eval_physics_num_tokens": 2841876.0,
|
|
"eval_physics_runtime": 41.4814,
|
|
"eval_physics_samples_per_second": 12.054,
|
|
"eval_physics_steps_per_second": 3.013,
|
|
"step": 100
|
|
},
|
|
{
|
|
"entropy": 0.7770459342747926,
|
|
"epoch": 0.176,
|
|
"grad_norm": 1.9765625,
|
|
"learning_rate": 2.1800000000000003e-06,
|
|
"loss": 0.8952,
|
|
"mean_token_accuracy": 0.7969978783279658,
|
|
"num_tokens": 3129237.0,
|
|
"step": 110
|
|
},
|
|
{
|
|
"entropy": 0.7958783155307174,
|
|
"epoch": 0.192,
|
|
"grad_norm": 2.3125,
|
|
"learning_rate": 2.38e-06,
|
|
"loss": 0.9119,
|
|
"mean_token_accuracy": 0.793396720290184,
|
|
"num_tokens": 3414875.0,
|
|
"step": 120
|
|
},
|
|
{
|
|
"entropy": 0.7911858700215817,
|
|
"epoch": 0.208,
|
|
"grad_norm": 2.4375,
|
|
"learning_rate": 2.5800000000000003e-06,
|
|
"loss": 0.8979,
|
|
"mean_token_accuracy": 0.7956417176872492,
|
|
"num_tokens": 3695623.0,
|
|
"step": 130
|
|
},
|
|
{
|
|
"entropy": 0.7871318949386478,
|
|
"epoch": 0.224,
|
|
"grad_norm": 2.09375,
|
|
"learning_rate": 2.7800000000000005e-06,
|
|
"loss": 0.8882,
|
|
"mean_token_accuracy": 0.7958744909614325,
|
|
"num_tokens": 3977666.0,
|
|
"step": 140
|
|
},
|
|
{
|
|
"entropy": 0.7629930060356855,
|
|
"epoch": 0.24,
|
|
"grad_norm": 1.9453125,
|
|
"learning_rate": 2.9800000000000003e-06,
|
|
"loss": 0.8654,
|
|
"mean_token_accuracy": 0.8001766365021468,
|
|
"num_tokens": 4269287.0,
|
|
"step": 150
|
|
},
|
|
{
|
|
"entropy": 0.7563591374084353,
|
|
"epoch": 0.256,
|
|
"grad_norm": 2.015625,
|
|
"learning_rate": 3.1800000000000005e-06,
|
|
"loss": 0.846,
|
|
"mean_token_accuracy": 0.8012917254120111,
|
|
"num_tokens": 4558632.0,
|
|
"step": 160
|
|
},
|
|
{
|
|
"entropy": 0.7706880155950785,
|
|
"epoch": 0.272,
|
|
"grad_norm": 1.859375,
|
|
"learning_rate": 3.3800000000000007e-06,
|
|
"loss": 0.8589,
|
|
"mean_token_accuracy": 0.8001386649906636,
|
|
"num_tokens": 4839270.0,
|
|
"step": 170
|
|
},
|
|
{
|
|
"entropy": 0.7551738245412707,
|
|
"epoch": 0.288,
|
|
"grad_norm": 1.609375,
|
|
"learning_rate": 3.58e-06,
|
|
"loss": 0.8468,
|
|
"mean_token_accuracy": 0.8015156831592322,
|
|
"num_tokens": 5125447.0,
|
|
"step": 180
|
|
},
|
|
{
|
|
"entropy": 0.7806578669697046,
|
|
"epoch": 0.304,
|
|
"grad_norm": 1.7734375,
|
|
"learning_rate": 3.7800000000000002e-06,
|
|
"loss": 0.8405,
|
|
"mean_token_accuracy": 0.7998461078852415,
|
|
"num_tokens": 5415925.0,
|
|
"step": 190
|
|
},
|
|
{
|
|
"entropy": 0.7316340124234557,
|
|
"epoch": 0.32,
|
|
"grad_norm": 1.6953125,
|
|
"learning_rate": 3.980000000000001e-06,
|
|
"loss": 0.7942,
|
|
"mean_token_accuracy": 0.8076768793165684,
|
|
"num_tokens": 5710178.0,
|
|
"step": 200
|
|
},
|
|
{
|
|
"epoch": 0.32,
|
|
"eval_math_entropy": 0.7726770024299622,
|
|
"eval_math_loss": 0.84871506690979,
|
|
"eval_math_mean_token_accuracy": 0.7979933972358704,
|
|
"eval_math_num_tokens": 5710178.0,
|
|
"eval_math_runtime": 35.765,
|
|
"eval_math_samples_per_second": 13.98,
|
|
"eval_math_steps_per_second": 3.495,
|
|
"step": 200
|
|
},
|
|
{
|
|
"epoch": 0.32,
|
|
"eval_physics_entropy": 0.6965276994705201,
|
|
"eval_physics_loss": 0.7036627531051636,
|
|
"eval_physics_mean_token_accuracy": 0.820941798210144,
|
|
"eval_physics_num_tokens": 5710178.0,
|
|
"eval_physics_runtime": 41.4597,
|
|
"eval_physics_samples_per_second": 12.06,
|
|
"eval_physics_steps_per_second": 3.015,
|
|
"step": 200
|
|
},
|
|
{
|
|
"entropy": 0.7737893080338836,
|
|
"epoch": 0.336,
|
|
"grad_norm": 1.828125,
|
|
"learning_rate": 4.18e-06,
|
|
"loss": 0.8126,
|
|
"mean_token_accuracy": 0.800426610186696,
|
|
"num_tokens": 5993843.0,
|
|
"step": 210
|
|
},
|
|
{
|
|
"entropy": 0.7482354704290628,
|
|
"epoch": 0.352,
|
|
"grad_norm": 1.9609375,
|
|
"learning_rate": 4.38e-06,
|
|
"loss": 0.7874,
|
|
"mean_token_accuracy": 0.8061643201857805,
|
|
"num_tokens": 6279137.0,
|
|
"step": 220
|
|
},
|
|
{
|
|
"entropy": 0.7608217550441623,
|
|
"epoch": 0.368,
|
|
"grad_norm": 1.65625,
|
|
"learning_rate": 4.58e-06,
|
|
"loss": 0.7742,
|
|
"mean_token_accuracy": 0.803000221401453,
|
|
"num_tokens": 6562387.0,
|
|
"step": 230
|
|
},
|
|
{
|
|
"entropy": 0.748550821095705,
|
|
"epoch": 0.384,
|
|
"grad_norm": 1.84375,
|
|
"learning_rate": 4.78e-06,
|
|
"loss": 0.7565,
|
|
"mean_token_accuracy": 0.8081529751420021,
|
|
"num_tokens": 6851520.0,
|
|
"step": 240
|
|
},
|
|
{
|
|
"entropy": 0.7808488950133323,
|
|
"epoch": 0.4,
|
|
"grad_norm": 1.75,
|
|
"learning_rate": 4.980000000000001e-06,
|
|
"loss": 0.7953,
|
|
"mean_token_accuracy": 0.7999043371528387,
|
|
"num_tokens": 7138932.0,
|
|
"step": 250
|
|
},
|
|
{
|
|
"entropy": 0.7674930268898607,
|
|
"epoch": 0.416,
|
|
"grad_norm": 1.6796875,
|
|
"learning_rate": 5.18e-06,
|
|
"loss": 0.7597,
|
|
"mean_token_accuracy": 0.8044768057763576,
|
|
"num_tokens": 7420720.0,
|
|
"step": 260
|
|
},
|
|
{
|
|
"entropy": 0.7496220650151372,
|
|
"epoch": 0.432,
|
|
"grad_norm": 1.5390625,
|
|
"learning_rate": 5.380000000000001e-06,
|
|
"loss": 0.7468,
|
|
"mean_token_accuracy": 0.8073872365057468,
|
|
"num_tokens": 7695518.0,
|
|
"step": 270
|
|
},
|
|
{
|
|
"entropy": 0.7439631385728717,
|
|
"epoch": 0.448,
|
|
"grad_norm": 2.21875,
|
|
"learning_rate": 5.580000000000001e-06,
|
|
"loss": 0.7339,
|
|
"mean_token_accuracy": 0.8076559219509363,
|
|
"num_tokens": 7983543.0,
|
|
"step": 280
|
|
},
|
|
{
|
|
"entropy": 0.7457249015569687,
|
|
"epoch": 0.464,
|
|
"grad_norm": 2.65625,
|
|
"learning_rate": 5.78e-06,
|
|
"loss": 0.741,
|
|
"mean_token_accuracy": 0.8087894931435585,
|
|
"num_tokens": 8271787.0,
|
|
"step": 290
|
|
},
|
|
{
|
|
"entropy": 0.7761816097423434,
|
|
"epoch": 0.48,
|
|
"grad_norm": 1.78125,
|
|
"learning_rate": 5.98e-06,
|
|
"loss": 0.7548,
|
|
"mean_token_accuracy": 0.8063571337610483,
|
|
"num_tokens": 8556170.0,
|
|
"step": 300
|
|
},
|
|
{
|
|
"epoch": 0.48,
|
|
"eval_math_entropy": 0.7653279647827148,
|
|
"eval_math_loss": 0.7636919021606445,
|
|
"eval_math_mean_token_accuracy": 0.8056883907318115,
|
|
"eval_math_num_tokens": 8556170.0,
|
|
"eval_math_runtime": 35.8013,
|
|
"eval_math_samples_per_second": 13.966,
|
|
"eval_math_steps_per_second": 3.491,
|
|
"step": 300
|
|
},
|
|
{
|
|
"epoch": 0.48,
|
|
"eval_physics_entropy": 0.6890090732574463,
|
|
"eval_physics_loss": 0.6574941277503967,
|
|
"eval_physics_mean_token_accuracy": 0.8227382025718689,
|
|
"eval_physics_num_tokens": 8556170.0,
|
|
"eval_physics_runtime": 41.4883,
|
|
"eval_physics_samples_per_second": 12.052,
|
|
"eval_physics_steps_per_second": 3.013,
|
|
"step": 300
|
|
},
|
|
{
|
|
"entropy": 0.7349074123427272,
|
|
"epoch": 0.496,
|
|
"grad_norm": 1.828125,
|
|
"learning_rate": 6.18e-06,
|
|
"loss": 0.7172,
|
|
"mean_token_accuracy": 0.8132212206721305,
|
|
"num_tokens": 8840590.0,
|
|
"step": 310
|
|
},
|
|
{
|
|
"entropy": 0.7443849379196763,
|
|
"epoch": 0.512,
|
|
"grad_norm": 1.6640625,
|
|
"learning_rate": 6.380000000000001e-06,
|
|
"loss": 0.7181,
|
|
"mean_token_accuracy": 0.812743404135108,
|
|
"num_tokens": 9131903.0,
|
|
"step": 320
|
|
},
|
|
{
|
|
"entropy": 0.7170431386679411,
|
|
"epoch": 0.528,
|
|
"grad_norm": 1.7109375,
|
|
"learning_rate": 6.5800000000000005e-06,
|
|
"loss": 0.7045,
|
|
"mean_token_accuracy": 0.8160003747791051,
|
|
"num_tokens": 9429519.0,
|
|
"step": 330
|
|
},
|
|
{
|
|
"entropy": 0.7328394843265414,
|
|
"epoch": 0.544,
|
|
"grad_norm": 2.234375,
|
|
"learning_rate": 6.780000000000001e-06,
|
|
"loss": 0.722,
|
|
"mean_token_accuracy": 0.8143766384571791,
|
|
"num_tokens": 9718247.0,
|
|
"step": 340
|
|
},
|
|
{
|
|
"entropy": 0.7620531113818287,
|
|
"epoch": 0.56,
|
|
"grad_norm": 2.453125,
|
|
"learning_rate": 6.98e-06,
|
|
"loss": 0.7395,
|
|
"mean_token_accuracy": 0.8070962373167276,
|
|
"num_tokens": 9991100.0,
|
|
"step": 350
|
|
},
|
|
{
|
|
"entropy": 0.7331368897110224,
|
|
"epoch": 0.576,
|
|
"grad_norm": 1.8515625,
|
|
"learning_rate": 7.180000000000001e-06,
|
|
"loss": 0.7173,
|
|
"mean_token_accuracy": 0.8115411166101694,
|
|
"num_tokens": 10276090.0,
|
|
"step": 360
|
|
},
|
|
{
|
|
"entropy": 0.7342592483386398,
|
|
"epoch": 0.592,
|
|
"grad_norm": 1.546875,
|
|
"learning_rate": 7.3800000000000005e-06,
|
|
"loss": 0.7141,
|
|
"mean_token_accuracy": 0.8124774299561978,
|
|
"num_tokens": 10566762.0,
|
|
"step": 370
|
|
},
|
|
{
|
|
"entropy": 0.7440765531733632,
|
|
"epoch": 0.608,
|
|
"grad_norm": 1.59375,
|
|
"learning_rate": 7.58e-06,
|
|
"loss": 0.7237,
|
|
"mean_token_accuracy": 0.8122259072959424,
|
|
"num_tokens": 10857002.0,
|
|
"step": 380
|
|
},
|
|
{
|
|
"entropy": 0.7216648861765862,
|
|
"epoch": 0.624,
|
|
"grad_norm": 2.25,
|
|
"learning_rate": 7.78e-06,
|
|
"loss": 0.7057,
|
|
"mean_token_accuracy": 0.8139940846711398,
|
|
"num_tokens": 11147326.0,
|
|
"step": 390
|
|
},
|
|
{
|
|
"entropy": 0.7119627617299557,
|
|
"epoch": 0.64,
|
|
"grad_norm": 1.5234375,
|
|
"learning_rate": 7.980000000000002e-06,
|
|
"loss": 0.6947,
|
|
"mean_token_accuracy": 0.8168974690139293,
|
|
"num_tokens": 11437710.0,
|
|
"step": 400
|
|
},
|
|
{
|
|
"epoch": 0.64,
|
|
"eval_math_entropy": 0.7429183626174927,
|
|
"eval_math_loss": 0.7330864071846008,
|
|
"eval_math_mean_token_accuracy": 0.8096566972732544,
|
|
"eval_math_num_tokens": 11437710.0,
|
|
"eval_math_runtime": 35.7944,
|
|
"eval_math_samples_per_second": 13.969,
|
|
"eval_math_steps_per_second": 3.492,
|
|
"step": 400
|
|
},
|
|
{
|
|
"epoch": 0.64,
|
|
"eval_physics_entropy": 0.6820782516002655,
|
|
"eval_physics_loss": 0.6511842012405396,
|
|
"eval_physics_mean_token_accuracy": 0.8221303143501282,
|
|
"eval_physics_num_tokens": 11437710.0,
|
|
"eval_physics_runtime": 41.4584,
|
|
"eval_physics_samples_per_second": 12.06,
|
|
"eval_physics_steps_per_second": 3.015,
|
|
"step": 400
|
|
},
|
|
{
|
|
"entropy": 0.712121813185513,
|
|
"epoch": 0.656,
|
|
"grad_norm": 1.40625,
|
|
"learning_rate": 8.18e-06,
|
|
"loss": 0.6914,
|
|
"mean_token_accuracy": 0.8179911982268095,
|
|
"num_tokens": 11727268.0,
|
|
"step": 410
|
|
},
|
|
{
|
|
"entropy": 0.7159143410623073,
|
|
"epoch": 0.672,
|
|
"grad_norm": 1.65625,
|
|
"learning_rate": 8.380000000000001e-06,
|
|
"loss": 0.7016,
|
|
"mean_token_accuracy": 0.813546035438776,
|
|
"num_tokens": 12020478.0,
|
|
"step": 420
|
|
},
|
|
{
|
|
"entropy": 0.7149255678057671,
|
|
"epoch": 0.688,
|
|
"grad_norm": 1.5078125,
|
|
"learning_rate": 8.580000000000001e-06,
|
|
"loss": 0.6908,
|
|
"mean_token_accuracy": 0.8162215895950794,
|
|
"num_tokens": 12300380.0,
|
|
"step": 430
|
|
},
|
|
{
|
|
"entropy": 0.7053035136312247,
|
|
"epoch": 0.704,
|
|
"grad_norm": 1.78125,
|
|
"learning_rate": 8.78e-06,
|
|
"loss": 0.6839,
|
|
"mean_token_accuracy": 0.8172036681324244,
|
|
"num_tokens": 12591366.0,
|
|
"step": 440
|
|
},
|
|
{
|
|
"entropy": 0.7228720126673579,
|
|
"epoch": 0.72,
|
|
"grad_norm": 1.7421875,
|
|
"learning_rate": 8.98e-06,
|
|
"loss": 0.6945,
|
|
"mean_token_accuracy": 0.8145164046436548,
|
|
"num_tokens": 12869618.0,
|
|
"step": 450
|
|
},
|
|
{
|
|
"entropy": 0.7052439017221331,
|
|
"epoch": 0.736,
|
|
"grad_norm": 1.75,
|
|
"learning_rate": 9.180000000000002e-06,
|
|
"loss": 0.6866,
|
|
"mean_token_accuracy": 0.8176059562712907,
|
|
"num_tokens": 13156311.0,
|
|
"step": 460
|
|
},
|
|
{
|
|
"entropy": 0.692553037032485,
|
|
"epoch": 0.752,
|
|
"grad_norm": 1.5625,
|
|
"learning_rate": 9.38e-06,
|
|
"loss": 0.6644,
|
|
"mean_token_accuracy": 0.821098442748189,
|
|
"num_tokens": 13438911.0,
|
|
"step": 470
|
|
},
|
|
{
|
|
"entropy": 0.6714967370033265,
|
|
"epoch": 0.768,
|
|
"grad_norm": 1.3203125,
|
|
"learning_rate": 9.58e-06,
|
|
"loss": 0.6595,
|
|
"mean_token_accuracy": 0.825907539948821,
|
|
"num_tokens": 13735397.0,
|
|
"step": 480
|
|
},
|
|
{
|
|
"entropy": 0.7148594724014401,
|
|
"epoch": 0.784,
|
|
"grad_norm": 1.7890625,
|
|
"learning_rate": 9.780000000000001e-06,
|
|
"loss": 0.6954,
|
|
"mean_token_accuracy": 0.8156244982033968,
|
|
"num_tokens": 14024497.0,
|
|
"step": 490
|
|
},
|
|
{
|
|
"entropy": 0.7110015576705336,
|
|
"epoch": 0.8,
|
|
"grad_norm": 1.6328125,
|
|
"learning_rate": 9.980000000000001e-06,
|
|
"loss": 0.69,
|
|
"mean_token_accuracy": 0.8156023971736431,
|
|
"num_tokens": 14316427.0,
|
|
"step": 500
|
|
},
|
|
{
|
|
"epoch": 0.8,
|
|
"eval_math_entropy": 0.7275117125511169,
|
|
"eval_math_loss": 0.7106672525405884,
|
|
"eval_math_mean_token_accuracy": 0.8132652158737183,
|
|
"eval_math_num_tokens": 14316427.0,
|
|
"eval_math_runtime": 35.7797,
|
|
"eval_math_samples_per_second": 13.974,
|
|
"eval_math_steps_per_second": 3.494,
|
|
"step": 500
|
|
},
|
|
{
|
|
"epoch": 0.8,
|
|
"eval_physics_entropy": 0.6752480027675629,
|
|
"eval_physics_loss": 0.6474179029464722,
|
|
"eval_physics_mean_token_accuracy": 0.8222041974067688,
|
|
"eval_physics_num_tokens": 14316427.0,
|
|
"eval_physics_runtime": 41.1592,
|
|
"eval_physics_samples_per_second": 12.148,
|
|
"eval_physics_steps_per_second": 3.037,
|
|
"step": 500
|
|
},
|
|
{
|
|
"entropy": 0.7031571570783853,
|
|
"epoch": 0.816,
|
|
"grad_norm": 1.7421875,
|
|
"learning_rate": 1.018e-05,
|
|
"loss": 0.6756,
|
|
"mean_token_accuracy": 0.8171336345374585,
|
|
"num_tokens": 14607136.0,
|
|
"step": 510
|
|
},
|
|
{
|
|
"entropy": 0.693414007127285,
|
|
"epoch": 0.832,
|
|
"grad_norm": 1.5859375,
|
|
"learning_rate": 1.038e-05,
|
|
"loss": 0.6706,
|
|
"mean_token_accuracy": 0.8211553748697042,
|
|
"num_tokens": 14894084.0,
|
|
"step": 520
|
|
},
|
|
{
|
|
"entropy": 0.7212638273835182,
|
|
"epoch": 0.848,
|
|
"grad_norm": 1.59375,
|
|
"learning_rate": 1.0580000000000002e-05,
|
|
"loss": 0.6965,
|
|
"mean_token_accuracy": 0.8145442601293326,
|
|
"num_tokens": 15177039.0,
|
|
"step": 530
|
|
},
|
|
{
|
|
"entropy": 0.7037423193454743,
|
|
"epoch": 0.864,
|
|
"grad_norm": 1.4296875,
|
|
"learning_rate": 1.0780000000000002e-05,
|
|
"loss": 0.6824,
|
|
"mean_token_accuracy": 0.816459146887064,
|
|
"num_tokens": 15463393.0,
|
|
"step": 540
|
|
},
|
|
{
|
|
"entropy": 0.7084541182965041,
|
|
"epoch": 0.88,
|
|
"grad_norm": 1.5546875,
|
|
"learning_rate": 1.0980000000000002e-05,
|
|
"loss": 0.6861,
|
|
"mean_token_accuracy": 0.8153413504362106,
|
|
"num_tokens": 15747943.0,
|
|
"step": 550
|
|
},
|
|
{
|
|
"entropy": 0.7014794830232859,
|
|
"epoch": 0.896,
|
|
"grad_norm": 1.5078125,
|
|
"learning_rate": 1.1180000000000001e-05,
|
|
"loss": 0.6844,
|
|
"mean_token_accuracy": 0.8173291139304638,
|
|
"num_tokens": 16027931.0,
|
|
"step": 560
|
|
},
|
|
{
|
|
"entropy": 0.6914853570982814,
|
|
"epoch": 0.912,
|
|
"grad_norm": 1.546875,
|
|
"learning_rate": 1.138e-05,
|
|
"loss": 0.6643,
|
|
"mean_token_accuracy": 0.8196637153625488,
|
|
"num_tokens": 16317127.0,
|
|
"step": 570
|
|
},
|
|
{
|
|
"entropy": 0.6866581376641989,
|
|
"epoch": 0.928,
|
|
"grad_norm": 1.46875,
|
|
"learning_rate": 1.1580000000000001e-05,
|
|
"loss": 0.661,
|
|
"mean_token_accuracy": 0.8199340119957924,
|
|
"num_tokens": 16598757.0,
|
|
"step": 580
|
|
},
|
|
{
|
|
"entropy": 0.7026519874110818,
|
|
"epoch": 0.944,
|
|
"grad_norm": 7.0625,
|
|
"learning_rate": 1.178e-05,
|
|
"loss": 0.6838,
|
|
"mean_token_accuracy": 0.8174264524132013,
|
|
"num_tokens": 16882236.0,
|
|
"step": 590
|
|
},
|
|
{
|
|
"entropy": 0.7053765447810292,
|
|
"epoch": 0.96,
|
|
"grad_norm": 1.515625,
|
|
"learning_rate": 1.198e-05,
|
|
"loss": 0.6832,
|
|
"mean_token_accuracy": 0.8171489015221596,
|
|
"num_tokens": 17169495.0,
|
|
"step": 600
|
|
},
|
|
{
|
|
"epoch": 0.96,
|
|
"eval_math_entropy": 0.7025088634490967,
|
|
"eval_math_loss": 0.6926642656326294,
|
|
"eval_math_mean_token_accuracy": 0.8157670006752015,
|
|
"eval_math_num_tokens": 17169495.0,
|
|
"eval_math_runtime": 35.7558,
|
|
"eval_math_samples_per_second": 13.984,
|
|
"eval_math_steps_per_second": 3.496,
|
|
"step": 600
|
|
},
|
|
{
|
|
"epoch": 0.96,
|
|
"eval_physics_entropy": 0.6707950072288513,
|
|
"eval_physics_loss": 0.6470632553100586,
|
|
"eval_physics_mean_token_accuracy": 0.8212249627113343,
|
|
"eval_physics_num_tokens": 17169495.0,
|
|
"eval_physics_runtime": 41.7459,
|
|
"eval_physics_samples_per_second": 11.977,
|
|
"eval_physics_steps_per_second": 2.994,
|
|
"step": 600
|
|
},
|
|
{
|
|
"entropy": 0.684873866662383,
|
|
"epoch": 0.976,
|
|
"grad_norm": 1.8515625,
|
|
"learning_rate": 1.218e-05,
|
|
"loss": 0.6671,
|
|
"mean_token_accuracy": 0.8196386702358722,
|
|
"num_tokens": 17454147.0,
|
|
"step": 610
|
|
},
|
|
{
|
|
"entropy": 0.7006405493244529,
|
|
"epoch": 0.992,
|
|
"grad_norm": 1.5390625,
|
|
"learning_rate": 1.2380000000000002e-05,
|
|
"loss": 0.672,
|
|
"mean_token_accuracy": 0.8165230687707663,
|
|
"num_tokens": 17742596.0,
|
|
"step": 620
|
|
},
|
|
{
|
|
"entropy": 0.6851861752569676,
|
|
"epoch": 1.008,
|
|
"grad_norm": 1.3828125,
|
|
"learning_rate": 1.2580000000000002e-05,
|
|
"loss": 0.6635,
|
|
"mean_token_accuracy": 0.8197994366288185,
|
|
"num_tokens": 18033540.0,
|
|
"step": 630
|
|
},
|
|
{
|
|
"entropy": 0.6624577900394797,
|
|
"epoch": 1.024,
|
|
"grad_norm": 1.5,
|
|
"learning_rate": 1.2780000000000001e-05,
|
|
"loss": 0.6405,
|
|
"mean_token_accuracy": 0.823651035502553,
|
|
"num_tokens": 18316346.0,
|
|
"step": 640
|
|
},
|
|
{
|
|
"entropy": 0.6590202970430254,
|
|
"epoch": 1.04,
|
|
"grad_norm": 1.6484375,
|
|
"learning_rate": 1.2980000000000001e-05,
|
|
"loss": 0.6392,
|
|
"mean_token_accuracy": 0.8249083694070578,
|
|
"num_tokens": 18595230.0,
|
|
"step": 650
|
|
},
|
|
{
|
|
"entropy": 0.6720681777223945,
|
|
"epoch": 1.056,
|
|
"grad_norm": 1.5625,
|
|
"learning_rate": 1.3180000000000001e-05,
|
|
"loss": 0.6481,
|
|
"mean_token_accuracy": 0.8219482194632292,
|
|
"num_tokens": 18885547.0,
|
|
"step": 660
|
|
},
|
|
{
|
|
"entropy": 0.666364430822432,
|
|
"epoch": 1.072,
|
|
"grad_norm": 1.5625,
|
|
"learning_rate": 1.3380000000000002e-05,
|
|
"loss": 0.6453,
|
|
"mean_token_accuracy": 0.821395493671298,
|
|
"num_tokens": 19169083.0,
|
|
"step": 670
|
|
},
|
|
{
|
|
"entropy": 0.6612174535170198,
|
|
"epoch": 1.088,
|
|
"grad_norm": 1.515625,
|
|
"learning_rate": 1.3580000000000002e-05,
|
|
"loss": 0.6328,
|
|
"mean_token_accuracy": 0.823925781622529,
|
|
"num_tokens": 19461465.0,
|
|
"step": 680
|
|
},
|
|
{
|
|
"entropy": 0.6591472556814552,
|
|
"epoch": 1.104,
|
|
"grad_norm": 1.3125,
|
|
"learning_rate": 1.378e-05,
|
|
"loss": 0.6414,
|
|
"mean_token_accuracy": 0.8235542017966508,
|
|
"num_tokens": 19755027.0,
|
|
"step": 690
|
|
},
|
|
{
|
|
"entropy": 0.6446556953713298,
|
|
"epoch": 1.12,
|
|
"grad_norm": 1.5390625,
|
|
"learning_rate": 1.398e-05,
|
|
"loss": 0.6235,
|
|
"mean_token_accuracy": 0.8280277159065008,
|
|
"num_tokens": 20043293.0,
|
|
"step": 700
|
|
},
|
|
{
|
|
"epoch": 1.12,
|
|
"eval_math_entropy": 0.6619428229331971,
|
|
"eval_math_loss": 0.6814246773719788,
|
|
"eval_math_mean_token_accuracy": 0.8176272907257081,
|
|
"eval_math_num_tokens": 20043293.0,
|
|
"eval_math_runtime": 35.7663,
|
|
"eval_math_samples_per_second": 13.98,
|
|
"eval_math_steps_per_second": 3.495,
|
|
"step": 700
|
|
},
|
|
{
|
|
"epoch": 1.12,
|
|
"eval_physics_entropy": 0.6304002203941346,
|
|
"eval_physics_loss": 0.6498542428016663,
|
|
"eval_physics_mean_token_accuracy": 0.8209347081184387,
|
|
"eval_physics_num_tokens": 20043293.0,
|
|
"eval_physics_runtime": 41.4297,
|
|
"eval_physics_samples_per_second": 12.069,
|
|
"eval_physics_steps_per_second": 3.017,
|
|
"step": 700
|
|
},
|
|
{
|
|
"entropy": 0.6416833782568574,
|
|
"epoch": 1.1360000000000001,
|
|
"grad_norm": 1.46875,
|
|
"learning_rate": 1.418e-05,
|
|
"loss": 0.6298,
|
|
"mean_token_accuracy": 0.8263315539807081,
|
|
"num_tokens": 20323010.0,
|
|
"step": 710
|
|
},
|
|
{
|
|
"entropy": 0.6587465461343527,
|
|
"epoch": 1.152,
|
|
"grad_norm": 1.578125,
|
|
"learning_rate": 1.4380000000000001e-05,
|
|
"loss": 0.6376,
|
|
"mean_token_accuracy": 0.8255084995180368,
|
|
"num_tokens": 20613266.0,
|
|
"step": 720
|
|
},
|
|
{
|
|
"entropy": 0.6476276513189078,
|
|
"epoch": 1.168,
|
|
"grad_norm": 1.40625,
|
|
"learning_rate": 1.4580000000000001e-05,
|
|
"loss": 0.6316,
|
|
"mean_token_accuracy": 0.8264493472874165,
|
|
"num_tokens": 20910404.0,
|
|
"step": 730
|
|
},
|
|
{
|
|
"entropy": 0.6569468161091209,
|
|
"epoch": 1.184,
|
|
"grad_norm": 1.546875,
|
|
"learning_rate": 1.478e-05,
|
|
"loss": 0.6336,
|
|
"mean_token_accuracy": 0.8247423082590103,
|
|
"num_tokens": 21200197.0,
|
|
"step": 740
|
|
},
|
|
{
|
|
"entropy": 0.6399336410686374,
|
|
"epoch": 1.2,
|
|
"grad_norm": 1.40625,
|
|
"learning_rate": 1.498e-05,
|
|
"loss": 0.622,
|
|
"mean_token_accuracy": 0.8271243523806333,
|
|
"num_tokens": 21496441.0,
|
|
"step": 750
|
|
},
|
|
{
|
|
"entropy": 0.6582916142418981,
|
|
"epoch": 1.216,
|
|
"grad_norm": 1.609375,
|
|
"learning_rate": 1.5180000000000002e-05,
|
|
"loss": 0.6353,
|
|
"mean_token_accuracy": 0.8234689589589834,
|
|
"num_tokens": 21778030.0,
|
|
"step": 760
|
|
},
|
|
{
|
|
"entropy": 0.637734330072999,
|
|
"epoch": 1.232,
|
|
"grad_norm": 1.5703125,
|
|
"learning_rate": 1.5380000000000002e-05,
|
|
"loss": 0.6159,
|
|
"mean_token_accuracy": 0.8288460947573185,
|
|
"num_tokens": 22060851.0,
|
|
"step": 770
|
|
},
|
|
{
|
|
"entropy": 0.636210915632546,
|
|
"epoch": 1.248,
|
|
"grad_norm": 1.5078125,
|
|
"learning_rate": 1.5580000000000003e-05,
|
|
"loss": 0.6165,
|
|
"mean_token_accuracy": 0.82775485701859,
|
|
"num_tokens": 22348738.0,
|
|
"step": 780
|
|
},
|
|
{
|
|
"entropy": 0.6564677763730288,
|
|
"epoch": 1.264,
|
|
"grad_norm": 1.3671875,
|
|
"learning_rate": 1.578e-05,
|
|
"loss": 0.6349,
|
|
"mean_token_accuracy": 0.8237839676439762,
|
|
"num_tokens": 22635039.0,
|
|
"step": 790
|
|
},
|
|
{
|
|
"entropy": 0.6571361353620887,
|
|
"epoch": 1.28,
|
|
"grad_norm": 1.421875,
|
|
"learning_rate": 1.5980000000000003e-05,
|
|
"loss": 0.6366,
|
|
"mean_token_accuracy": 0.8244204126298428,
|
|
"num_tokens": 22915190.0,
|
|
"step": 800
|
|
},
|
|
{
|
|
"epoch": 1.28,
|
|
"eval_math_entropy": 0.6791466112136841,
|
|
"eval_math_loss": 0.669770359992981,
|
|
"eval_math_mean_token_accuracy": 0.8196616005897522,
|
|
"eval_math_num_tokens": 22915190.0,
|
|
"eval_math_runtime": 35.7574,
|
|
"eval_math_samples_per_second": 13.983,
|
|
"eval_math_steps_per_second": 3.496,
|
|
"step": 800
|
|
},
|
|
{
|
|
"epoch": 1.28,
|
|
"eval_physics_entropy": 0.6547831742763519,
|
|
"eval_physics_loss": 0.650165319442749,
|
|
"eval_physics_mean_token_accuracy": 0.82009166097641,
|
|
"eval_physics_num_tokens": 22915190.0,
|
|
"eval_physics_runtime": 41.4354,
|
|
"eval_physics_samples_per_second": 12.067,
|
|
"eval_physics_steps_per_second": 3.017,
|
|
"step": 800
|
|
},
|
|
{
|
|
"entropy": 0.6429510954767466,
|
|
"epoch": 1.296,
|
|
"grad_norm": 1.40625,
|
|
"learning_rate": 1.618e-05,
|
|
"loss": 0.6172,
|
|
"mean_token_accuracy": 0.8271825641393662,
|
|
"num_tokens": 23194947.0,
|
|
"step": 810
|
|
},
|
|
{
|
|
"entropy": 0.6526322312653065,
|
|
"epoch": 1.312,
|
|
"grad_norm": 1.5390625,
|
|
"learning_rate": 1.638e-05,
|
|
"loss": 0.6414,
|
|
"mean_token_accuracy": 0.8243110898882151,
|
|
"num_tokens": 23483496.0,
|
|
"step": 820
|
|
},
|
|
{
|
|
"entropy": 0.6679502466693521,
|
|
"epoch": 1.328,
|
|
"grad_norm": 1.6171875,
|
|
"learning_rate": 1.658e-05,
|
|
"loss": 0.6441,
|
|
"mean_token_accuracy": 0.8223706830292941,
|
|
"num_tokens": 23762589.0,
|
|
"step": 830
|
|
},
|
|
{
|
|
"entropy": 0.6403571154922247,
|
|
"epoch": 1.3439999999999999,
|
|
"grad_norm": 1.4921875,
|
|
"learning_rate": 1.6780000000000002e-05,
|
|
"loss": 0.6222,
|
|
"mean_token_accuracy": 0.8268204662948847,
|
|
"num_tokens": 24041497.0,
|
|
"step": 840
|
|
},
|
|
{
|
|
"entropy": 0.6487369466573,
|
|
"epoch": 1.3599999999999999,
|
|
"grad_norm": 1.203125,
|
|
"learning_rate": 1.698e-05,
|
|
"loss": 0.6282,
|
|
"mean_token_accuracy": 0.8273422639817,
|
|
"num_tokens": 24330591.0,
|
|
"step": 850
|
|
},
|
|
{
|
|
"entropy": 0.688577077165246,
|
|
"epoch": 1.376,
|
|
"grad_norm": 1.4453125,
|
|
"learning_rate": 1.718e-05,
|
|
"loss": 0.6651,
|
|
"mean_token_accuracy": 0.8164406325668097,
|
|
"num_tokens": 24611420.0,
|
|
"step": 860
|
|
},
|
|
{
|
|
"entropy": 0.643337251059711,
|
|
"epoch": 1.392,
|
|
"grad_norm": 1.4375,
|
|
"learning_rate": 1.7380000000000003e-05,
|
|
"loss": 0.6209,
|
|
"mean_token_accuracy": 0.825714922696352,
|
|
"num_tokens": 24894694.0,
|
|
"step": 870
|
|
},
|
|
{
|
|
"entropy": 0.6371515756472945,
|
|
"epoch": 1.408,
|
|
"grad_norm": 1.453125,
|
|
"learning_rate": 1.758e-05,
|
|
"loss": 0.6156,
|
|
"mean_token_accuracy": 0.8278428696095943,
|
|
"num_tokens": 25184946.0,
|
|
"step": 880
|
|
},
|
|
{
|
|
"entropy": 0.6312845427542925,
|
|
"epoch": 1.424,
|
|
"grad_norm": 1.25,
|
|
"learning_rate": 1.7780000000000003e-05,
|
|
"loss": 0.615,
|
|
"mean_token_accuracy": 0.8290112767368555,
|
|
"num_tokens": 25474746.0,
|
|
"step": 890
|
|
},
|
|
{
|
|
"entropy": 0.6522130774334073,
|
|
"epoch": 1.44,
|
|
"grad_norm": 1.3671875,
|
|
"learning_rate": 1.798e-05,
|
|
"loss": 0.6295,
|
|
"mean_token_accuracy": 0.8248222030699253,
|
|
"num_tokens": 25758550.0,
|
|
"step": 900
|
|
},
|
|
{
|
|
"epoch": 1.44,
|
|
"eval_math_entropy": 0.6579540917873382,
|
|
"eval_math_loss": 0.6607276797294617,
|
|
"eval_math_mean_token_accuracy": 0.8207882242202759,
|
|
"eval_math_num_tokens": 25758550.0,
|
|
"eval_math_runtime": 35.7442,
|
|
"eval_math_samples_per_second": 13.988,
|
|
"eval_math_steps_per_second": 3.497,
|
|
"step": 900
|
|
},
|
|
{
|
|
"epoch": 1.44,
|
|
"eval_physics_entropy": 0.6424285123348236,
|
|
"eval_physics_loss": 0.651524543762207,
|
|
"eval_physics_mean_token_accuracy": 0.8197560949325562,
|
|
"eval_physics_num_tokens": 25758550.0,
|
|
"eval_physics_runtime": 41.435,
|
|
"eval_physics_samples_per_second": 12.067,
|
|
"eval_physics_steps_per_second": 3.017,
|
|
"step": 900
|
|
},
|
|
{
|
|
"entropy": 0.6359886834397912,
|
|
"epoch": 1.456,
|
|
"grad_norm": 1.3203125,
|
|
"learning_rate": 1.8180000000000002e-05,
|
|
"loss": 0.6177,
|
|
"mean_token_accuracy": 0.8268891926854849,
|
|
"num_tokens": 26039862.0,
|
|
"step": 910
|
|
},
|
|
{
|
|
"entropy": 0.6429944572970271,
|
|
"epoch": 1.472,
|
|
"grad_norm": 1.28125,
|
|
"learning_rate": 1.8380000000000004e-05,
|
|
"loss": 0.6266,
|
|
"mean_token_accuracy": 0.8256016962230206,
|
|
"num_tokens": 26318666.0,
|
|
"step": 920
|
|
},
|
|
{
|
|
"entropy": 0.666933044232428,
|
|
"epoch": 1.488,
|
|
"grad_norm": 1.265625,
|
|
"learning_rate": 1.858e-05,
|
|
"loss": 0.6455,
|
|
"mean_token_accuracy": 0.8214054699987173,
|
|
"num_tokens": 26594867.0,
|
|
"step": 930
|
|
},
|
|
{
|
|
"entropy": 0.6266062622889876,
|
|
"epoch": 1.504,
|
|
"grad_norm": 1.5,
|
|
"learning_rate": 1.878e-05,
|
|
"loss": 0.6092,
|
|
"mean_token_accuracy": 0.8303067740052938,
|
|
"num_tokens": 26887233.0,
|
|
"step": 940
|
|
},
|
|
{
|
|
"entropy": 0.659474827721715,
|
|
"epoch": 1.52,
|
|
"grad_norm": 1.140625,
|
|
"learning_rate": 1.898e-05,
|
|
"loss": 0.6344,
|
|
"mean_token_accuracy": 0.8236033659428358,
|
|
"num_tokens": 27177448.0,
|
|
"step": 950
|
|
},
|
|
{
|
|
"entropy": 0.6371477670967579,
|
|
"epoch": 1.536,
|
|
"grad_norm": 1.1484375,
|
|
"learning_rate": 1.918e-05,
|
|
"loss": 0.6245,
|
|
"mean_token_accuracy": 0.8263927459716797,
|
|
"num_tokens": 27467857.0,
|
|
"step": 960
|
|
},
|
|
{
|
|
"entropy": 0.6363119596615434,
|
|
"epoch": 1.552,
|
|
"grad_norm": 1.265625,
|
|
"learning_rate": 1.938e-05,
|
|
"loss": 0.6172,
|
|
"mean_token_accuracy": 0.8277775753289461,
|
|
"num_tokens": 27755138.0,
|
|
"step": 970
|
|
},
|
|
{
|
|
"entropy": 0.625205036252737,
|
|
"epoch": 1.568,
|
|
"grad_norm": 1.15625,
|
|
"learning_rate": 1.9580000000000002e-05,
|
|
"loss": 0.6047,
|
|
"mean_token_accuracy": 0.8324479281902313,
|
|
"num_tokens": 28049363.0,
|
|
"step": 980
|
|
},
|
|
{
|
|
"entropy": 0.6341174798086285,
|
|
"epoch": 1.584,
|
|
"grad_norm": 1.375,
|
|
"learning_rate": 1.978e-05,
|
|
"loss": 0.6153,
|
|
"mean_token_accuracy": 0.8282931160181761,
|
|
"num_tokens": 28337890.0,
|
|
"step": 990
|
|
},
|
|
{
|
|
"entropy": 0.6248919799923897,
|
|
"epoch": 1.6,
|
|
"grad_norm": 1.3359375,
|
|
"learning_rate": 1.9980000000000002e-05,
|
|
"loss": 0.6112,
|
|
"mean_token_accuracy": 0.827129240334034,
|
|
"num_tokens": 28626068.0,
|
|
"step": 1000
|
|
},
|
|
{
|
|
"epoch": 1.6,
|
|
"eval_math_entropy": 0.6310200927257538,
|
|
"eval_math_loss": 0.651292085647583,
|
|
"eval_math_mean_token_accuracy": 0.8225455293655396,
|
|
"eval_math_num_tokens": 28626068.0,
|
|
"eval_math_runtime": 35.7544,
|
|
"eval_math_samples_per_second": 13.984,
|
|
"eval_math_steps_per_second": 3.496,
|
|
"step": 1000
|
|
},
|
|
{
|
|
"epoch": 1.6,
|
|
"eval_physics_entropy": 0.6265229551792145,
|
|
"eval_physics_loss": 0.6552274227142334,
|
|
"eval_physics_mean_token_accuracy": 0.8185877275466918,
|
|
"eval_physics_num_tokens": 28626068.0,
|
|
"eval_physics_runtime": 41.1356,
|
|
"eval_physics_samples_per_second": 12.155,
|
|
"eval_physics_steps_per_second": 3.039,
|
|
"step": 1000
|
|
}
|
|
],
|
|
"logging_steps": 10,
|
|
"max_steps": 10000,
|
|
"num_input_tokens_seen": 0,
|
|
"num_train_epochs": 16,
|
|
"save_steps": 500,
|
|
"stateful_callbacks": {
|
|
"TrainerControl": {
|
|
"args": {
|
|
"should_epoch_stop": false,
|
|
"should_evaluate": false,
|
|
"should_log": false,
|
|
"should_save": true,
|
|
"should_training_stop": false
|
|
},
|
|
"attributes": {}
|
|
}
|
|
},
|
|
"total_flos": 2.1731341502193254e+18,
|
|
"train_batch_size": 4,
|
|
"trial_name": null,
|
|
"trial_params": null
|
|
}
|