add additional 2-epoch checkpoint, better regularization

config.json

@@ -1,5 +1,5 @@
 {
-  "_name_or_path": "pszemraj/led-large-book-summary-11E",
+  "_name_or_path": "pszemraj/led-large-book-summary",
   "_num_labels": 3,
   "activation_dropout": 0.0,
   "activation_function": "gelu",
@@ -52,8 +52,8 @@
   "length_penalty": 0.8,
   "max_decoder_position_embeddings": 1024,
   "max_encoder_position_embeddings": 16384,
-  "max_length": 512,
-  "min_length": 32,
+  "max_length": 1024,
+  "min_length": 8,
   "model_type": "led",
   "no_repeat_ngram_size": 3,
   "num_beams": 4,
@@ -61,6 +61,7 @@
   "output_past": false,
   "pad_token_id": 1,
   "prefix": " ",
+  "repetition_penalty": 3.5,
   "torch_dtype": "float32",
   "transformers_version": "4.19.2",
   "use_cache": false,

latest

@@ -0,0 +1 @@
+global_step296

pytorch_model.bin

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:cae26c66ffc396d3f1b515359fef3cc2b4a080043dbfa56f7a968450c8ad1b27
+oid sha256:0603a70f15308ebccb9b66369464a83efb920ed266f736bdca0279bc066eecf7
 size 1839482407

rng_state_0.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:97f7901e7b6ddfc02769b63a371fd87e014fae5a4c2ef46f9e10298a4e62e643
+size 14439

tokenizer_config.json

@@ -1 +1 @@
-{"errors": "replace", "bos_token": {"content": "<s>", "single_word": false, "lstrip": false, "rstrip": false, "normalized": true, "__type": "AddedToken"}, "eos_token": {"content": "</s>", "single_word": false, "lstrip": false, "rstrip": false, "normalized": true, "__type": "AddedToken"}, "sep_token": {"content": "</s>", "single_word": false, "lstrip": false, "rstrip": false, "normalized": true, "__type": "AddedToken"}, "cls_token": {"content": "<s>", "single_word": false, "lstrip": false, "rstrip": false, "normalized": true, "__type": "AddedToken"}, "unk_token": {"content": "<unk>", "single_word": false, "lstrip": false, "rstrip": false, "normalized": true, "__type": "AddedToken"}, "pad_token": {"content": "<pad>", "single_word": false, "lstrip": false, "rstrip": false, "normalized": true, "__type": "AddedToken"}, "mask_token": {"content": "<mask>", "single_word": false, "lstrip": true, "rstrip": false, "normalized": true, "__type": "AddedToken"}, "add_prefix_space": false, "trim_offsets": true, "model_max_length": 16384, "special_tokens_map_file": "/root/.cache/huggingface/transformers/2ad921573d53ebf0c0450d63a211e61d8e328324e84830c365abff01f2d115f1.cb2244924ab24d706b02fd7fcedaea4531566537687a539ebb94db511fd122a0", "name_or_path": "pszemraj/led-large-book-summary-11E", "tokenizer_class": "LEDTokenizer"}
+{"errors": "replace", "bos_token": {"content": "<s>", "single_word": false, "lstrip": false, "rstrip": false, "normalized": true, "__type": "AddedToken"}, "eos_token": {"content": "</s>", "single_word": false, "lstrip": false, "rstrip": false, "normalized": true, "__type": "AddedToken"}, "sep_token": {"content": "</s>", "single_word": false, "lstrip": false, "rstrip": false, "normalized": true, "__type": "AddedToken"}, "cls_token": {"content": "<s>", "single_word": false, "lstrip": false, "rstrip": false, "normalized": true, "__type": "AddedToken"}, "unk_token": {"content": "<unk>", "single_word": false, "lstrip": false, "rstrip": false, "normalized": true, "__type": "AddedToken"}, "pad_token": {"content": "<pad>", "single_word": false, "lstrip": false, "rstrip": false, "normalized": true, "__type": "AddedToken"}, "mask_token": {"content": "<mask>", "single_word": false, "lstrip": true, "rstrip": false, "normalized": true, "__type": "AddedToken"}, "add_prefix_space": false, "trim_offsets": true, "model_max_length": 16384, "special_tokens_map_file": "/root/.cache/huggingface/transformers/2ad921573d53ebf0c0450d63a211e61d8e328324e84830c365abff01f2d115f1.cb2244924ab24d706b02fd7fcedaea4531566537687a539ebb94db511fd122a0", "name_or_path": "pszemraj/led-large-book-summary", "tokenizer_class": "LEDTokenizer"}

trainer_state.json

@@ -1,1537 +1,364 @@
 {
   "best_metric": null,
   "best_model_checkpoint": null,
-  "epoch": 1.9994069974303224,
-  "global_step": 1264,
+  "epoch": 2.9914638001896936,
+  "global_step": 294,
   "is_hyper_param_search": false,
   "is_local_process_zero": true,
   "is_world_process_zero": true,
   "log_history": [
     {
-      "epoch": 0.01,
-      "learning_rate": 2.631578947368421e-06,
-      "loss": 1.0727,
+      "epoch": 0.05,
+      "learning_rate": 7.5e-06,
+      "loss": 0.481,
       "step": 5
     },
     {
-      "epoch": 0.02,
-      "learning_rate": 5.263157894736842e-06,
-      "loss": 0.8573,
+      "epoch": 0.1,
+      "learning_rate": 1.5e-05,
+      "loss": 0.5091,
       "step": 10
     },
     {
-      "epoch": 0.02,
-      "learning_rate": 7.894736842105265e-06,
-      "loss": 1.0991,
+      "epoch": 0.15,
+      "learning_rate": 2.25e-05,
+      "loss": 0.4303,
       "step": 15
     },
     {
-      "epoch": 0.03,
-      "learning_rate": 1.0526315789473684e-05,
-      "loss": 0.8893,
+      "epoch": 0.2,
+      "learning_rate": 3e-05,
+      "loss": 0.4055,
       "step": 20
     },
     {
-      "epoch": 0.04,
-      "learning_rate": 1.3157894736842108e-05,
-      "loss": 0.9549,
+      "epoch": 0.25,
+      "learning_rate": 2.998662940889891e-05,
+      "loss": 0.4338,
       "step": 25
     },
     {
-      "epoch": 0.05,
-      "learning_rate": 1.578947368421053e-05,
-      "loss": 0.9953,
+      "epoch": 0.3,
+      "learning_rate": 2.9946541471956496e-05,
+      "loss": 0.4793,
       "step": 30
     },
     {
-      "epoch": 0.06,
-      "learning_rate": 1.8421052631578947e-05,
-      "loss": 1.0163,
+      "epoch": 0.35,
+      "learning_rate": 2.9879807655761145e-05,
+      "loss": 0.4291,
       "step": 35
     },
     {
-      "epoch": 0.06,
-      "learning_rate": 1.9999868674866755e-05,
-      "loss": 0.9562,
+      "epoch": 0.4,
+      "learning_rate": 2.9786546929722055e-05,
+      "loss": 0.4908,
       "step": 40
     },
     {
-      "epoch": 0.07,
-      "learning_rate": 1.9998391306730024e-05,
-      "loss": 1.0517,
+      "epoch": 0.46,
+      "learning_rate": 2.966692555397705e-05,
+      "loss": 0.463,
       "step": 45
     },
     {
-      "epoch": 0.08,
-      "learning_rate": 1.9995272657365104e-05,
-      "loss": 0.9458,
+      "epoch": 0.51,
+      "learning_rate": 2.9521156782993066e-05,
+      "loss": 0.528,
       "step": 50
     },
     {
-      "epoch": 0.09,
-      "learning_rate": 1.9990513238712407e-05,
-      "loss": 0.9163,
+      "epoch": 0.56,
+      "learning_rate": 2.9349500485387718e-05,
+      "loss": 0.5178,
       "step": 55
     },
     {
-      "epoch": 0.09,
-      "learning_rate": 1.998411383205207e-05,
-      "loss": 0.8653,
+      "epoch": 0.61,
+      "learning_rate": 2.9152262680649704e-05,
+      "loss": 0.4602,
       "step": 60
     },
     {
-      "epoch": 0.1,
-      "learning_rate": 1.9976075487875692e-05,
-      "loss": 1.063,
+      "epoch": 0.66,
+      "learning_rate": 2.8929794993583937e-05,
+      "loss": 0.5044,
       "step": 65
     },
     {
-      "epoch": 0.11,
-      "learning_rate": 1.9966399525713882e-05,
-      "loss": 0.8901,
+      "epoch": 0.71,
+      "learning_rate": 2.8682494027454e-05,
+      "loss": 0.4217,
       "step": 70
     },
     {
-      "epoch": 0.12,
-      "learning_rate": 1.9955087533919662e-05,
-      "loss": 1.0556,
+      "epoch": 0.76,
+      "learning_rate": 2.8410800656939512e-05,
+      "loss": 0.502,
       "step": 75
     },
     {
-      "epoch": 0.13,
-      "learning_rate": 1.994214136940773e-05,
-      "loss": 0.9434,
+      "epoch": 0.81,
+      "learning_rate": 2.811519924216873e-05,
+      "loss": 0.4549,
       "step": 80
     },
     {
-      "epoch": 0.13,
-      "learning_rate": 1.9927563157349646e-05,
-      "loss": 0.9989,
+      "epoch": 0.86,
+      "learning_rate": 2.779621676522777e-05,
+      "loss": 0.5692,
       "step": 85
     },
     {
-      "epoch": 0.14,
-      "learning_rate": 1.9911355290824955e-05,
-      "loss": 0.9829,
+      "epoch": 0.91,
+      "learning_rate": 2.7454421890685647e-05,
+      "loss": 0.4312,
       "step": 90
     },
     {
-      "epoch": 0.15,
-      "learning_rate": 1.9893520430428378e-05,
-      "loss": 0.9988,
+      "epoch": 0.96,
+      "learning_rate": 2.709042395181008e-05,
+      "loss": 0.4938,
       "step": 95
     },
     {
-      "epoch": 0.16,
-      "learning_rate": 1.987406150383304e-05,
-      "loss": 1.0073,
+      "epoch": 1.02,
+      "learning_rate": 2.6704871864281377e-05,
+      "loss": 0.5433,
       "step": 100
     },
     {
-      "epoch": 0.17,
-      "learning_rate": 1.98529817053099e-05,
-      "loss": 0.8863,
+      "epoch": 1.07,
+      "learning_rate": 2.6298452969340952e-05,
+      "loss": 0.3459,
       "step": 105
     },
     {
-      "epoch": 0.17,
-      "learning_rate": 1.983028449520338e-05,
-      "loss": 0.8015,
+      "epoch": 1.12,
+      "learning_rate": 2.58718918084368e-05,
+      "loss": 0.3739,
       "step": 110
     },
     {
-      "epoch": 0.18,
-      "learning_rate": 1.980597359936335e-05,
-      "loss": 1.0477,
+      "epoch": 1.17,
+      "learning_rate": 2.5425948831550528e-05,
+      "loss": 0.375,
       "step": 115
     },
     {
-      "epoch": 0.19,
-      "learning_rate": 1.9780053008533486e-05,
-      "loss": 0.961,
+      "epoch": 1.22,
+      "learning_rate": 2.496141904150859e-05,
+      "loss": 0.3809,
       "step": 120
     },
     {
-      "epoch": 0.2,
-      "learning_rate": 1.97525269776962e-05,
-      "loss": 0.9476,
+      "epoch": 1.27,
+      "learning_rate": 2.447913057669456e-05,
+      "loss": 0.4183,
       "step": 125
     },
     {
-      "epoch": 0.21,
-      "learning_rate": 1.9723400025374168e-05,
-      "loss": 1.1,
+      "epoch": 1.32,
+      "learning_rate": 2.3979943234689226e-05,
+      "loss": 0.4207,
       "step": 130
     },
     {
-      "epoch": 0.21,
-      "learning_rate": 1.969267693288855e-05,
-      "loss": 0.946,
+      "epoch": 1.37,
+      "learning_rate": 2.3464746939470288e-05,
+      "loss": 0.3767,
       "step": 135
     },
     {
-      "epoch": 0.22,
-      "learning_rate": 1.9660362743574163e-05,
-      "loss": 0.946,
+      "epoch": 1.42,
+      "learning_rate": 2.2934460154904436e-05,
+      "loss": 0.4248,
       "step": 140
     },
     {
-      "epoch": 0.23,
-      "learning_rate": 1.9626462761951583e-05,
-      "loss": 0.9368,
+      "epoch": 1.48,
+      "learning_rate": 2.2390028247360042e-05,
+      "loss": 0.3374,
       "step": 145
     },
     {
-      "epoch": 0.24,
-      "learning_rate": 1.959098255285636e-05,
-      "loss": 1.0338,
+      "epoch": 1.53,
+      "learning_rate": 2.183242180035951e-05,
+      "loss": 0.4582,
       "step": 150
     },
     {
-      "epoch": 0.25,
-      "learning_rate": 1.9553927940525557e-05,
-      "loss": 1.0359,
+      "epoch": 1.58,
+      "learning_rate": 2.1262634884275948e-05,
+      "loss": 0.4153,
       "step": 155
     },
     {
-      "epoch": 0.25,
-      "learning_rate": 1.9515305007641653e-05,
-      "loss": 1.0189,
+      "epoch": 1.63,
+      "learning_rate": 2.068168328415864e-05,
+      "loss": 0.409,
       "step": 160
     },
     {
-      "epoch": 0.26,
-      "learning_rate": 1.9475120094334046e-05,
-      "loss": 1.0388,
+      "epoch": 1.68,
+      "learning_rate": 2.0090602688846884e-05,
+      "loss": 0.4023,
       "step": 165
     },
     {
-      "epoch": 0.27,
-      "learning_rate": 1.9433379797138314e-05,
-      "loss": 1.0733,
+      "epoch": 1.73,
+      "learning_rate": 1.9490446844600375e-05,
+      "loss": 0.3426,
       "step": 170
     },
     {
-      "epoch": 0.28,
-      "learning_rate": 1.939009096791333e-05,
-      "loss": 0.9984,
+      "epoch": 1.78,
+      "learning_rate": 1.888228567653781e-05,
+      "loss": 0.4059,
       "step": 175
     },
     {
-      "epoch": 0.28,
-      "learning_rate": 1.9345260712716517e-05,
-      "loss": 1.0221,
+      "epoch": 1.83,
+      "learning_rate": 1.8267203381232774e-05,
+      "loss": 0.4449,
       "step": 180
     },
     {
-      "epoch": 0.29,
-      "learning_rate": 1.9298896390637363e-05,
-      "loss": 1.2205,
+      "epoch": 1.88,
+      "learning_rate": 1.764629649386713e-05,
+      "loss": 0.4362,
       "step": 185
     },
     {
-      "epoch": 0.3,
-      "learning_rate": 1.9251005612589382e-05,
-      "loss": 0.9938,
+      "epoch": 1.93,
+      "learning_rate": 1.7020671933387917e-05,
+      "loss": 0.4874,
       "step": 190
     },
     {
-      "epoch": 0.31,
-      "learning_rate": 1.9201596240060737e-05,
-      "loss": 1.0059,
+      "epoch": 1.98,
+      "learning_rate": 1.63914450291526e-05,
+      "loss": 0.3326,
       "step": 195
     },
     {
-      "epoch": 0.32,
-      "learning_rate": 1.9150676383823775e-05,
-      "loss": 1.0072,
+      "epoch": 2.04,
+      "learning_rate": 1.5633197410233404e-05,
+      "loss": 0.4035,
       "step": 200
     },
     {
-      "epoch": 0.32,
-      "learning_rate": 1.9098254402603573e-05,
-      "loss": 0.8886,
+      "epoch": 2.09,
+      "learning_rate": 1.5e-05,
+      "loss": 0.3291,
       "step": 205
     },
     {
-      "epoch": 0.33,
-      "learning_rate": 1.904433890170583e-05,
-      "loss": 1.1496,
+      "epoch": 2.14,
+      "learning_rate": 1.4366802589766598e-05,
+      "loss": 0.353,
       "step": 210
     },
     {
-      "epoch": 0.34,
-      "learning_rate": 1.898893873160428e-05,
-      "loss": 0.8883,
+      "epoch": 2.19,
+      "learning_rate": 1.373473400935433e-05,
+      "loss": 0.319,
       "step": 215
     },
     {
-      "epoch": 0.35,
-      "learning_rate": 1.893206298648781e-05,
-      "loss": 0.9564,
+      "epoch": 2.24,
+      "learning_rate": 1.3104921076168065e-05,
+      "loss": 0.341,
       "step": 220
     },
     {
-      "epoch": 0.36,
-      "learning_rate": 1.887372100276764e-05,
-      "loss": 0.9264,
+      "epoch": 2.29,
+      "learning_rate": 1.247848658636778e-05,
+      "loss": 0.3276,
       "step": 225
     },
     {
-      "epoch": 0.36,
-      "learning_rate": 1.8813922357544713e-05,
-      "loss": 1.1487,
+      "epoch": 2.34,
+      "learning_rate": 1.185654731320877e-05,
+      "loss": 0.3628,
       "step": 230
     },
     {
-      "epoch": 0.37,
-      "learning_rate": 1.875267686703754e-05,
-      "loss": 1.0543,
+      "epoch": 2.39,
+      "learning_rate": 1.124021201611919e-05,
+      "loss": 0.2727,
       "step": 235
     },
     {
-      "epoch": 0.38,
-      "learning_rate": 1.8689994584970843e-05,
-      "loss": 1.0098,
+      "epoch": 2.45,
+      "learning_rate": 1.0630579464064182e-05,
+      "loss": 0.3466,
       "step": 240
     },
     {
-      "epoch": 0.39,
-      "learning_rate": 1.8625885800925193e-05,
-      "loss": 0.915,
+      "epoch": 2.5,
+      "learning_rate": 1.0028736476720464e-05,
+      "loss": 0.3187,
       "step": 245
     },
     {
-      "epoch": 0.4,
-      "learning_rate": 1.8560361038647917e-05,
-      "loss": 1.0092,
+      "epoch": 2.55,
+      "learning_rate": 9.435755986953485e-06,
+      "loss": 0.3837,
       "step": 250
     },
     {
-      "epoch": 0.4,
-      "learning_rate": 1.8493431054325583e-05,
-      "loss": 1.0312,
+      "epoch": 2.6,
+      "learning_rate": 8.852695128051192e-06,
+      "loss": 0.2955,
       "step": 255
     },
     {
-      "epoch": 0.41,
-      "learning_rate": 1.8425106834818336e-05,
-      "loss": 1.0963,
+      "epoch": 2.65,
+      "learning_rate": 8.280593349124432e-06,
+      "loss": 0.3793,
       "step": 260
     },
     {
-      "epoch": 0.42,
-      "learning_rate": 1.8355399595856326e-05,
-      "loss": 0.8076,
+      "epoch": 2.7,
+      "learning_rate": 7.720470562033787e-06,
+      "loss": 0.3443,
       "step": 265
     },
     {
-      "epoch": 0.43,
-      "learning_rate": 1.8284320780198624e-05,
-      "loss": 0.8218,
+      "epoch": 2.75,
+      "learning_rate": 7.17332532314626e-06,
+      "loss": 0.2915,
       "step": 270
     },
     {
-      "epoch": 0.43,
-      "learning_rate": 1.8211882055754824e-05,
-      "loss": 0.9424,
+      "epoch": 2.8,
+      "learning_rate": 6.640133053163455e-06,
+      "loss": 0.3514,
       "step": 275
     },
     {
-      "epoch": 0.44,
-      "learning_rate": 1.8138095313669705e-05,
-      "loss": 1.0962,
+      "epoch": 2.85,
+      "learning_rate": 6.12184429819474e-06,
+      "loss": 0.3221,
       "step": 280
     },
     {
-      "epoch": 0.45,
-      "learning_rate": 1.8062972666371258e-05,
-      "loss": 1.0918,
+      "epoch": 2.9,
+      "learning_rate": 5.619383035175448e-06,
+      "loss": 0.2903,
       "step": 285
     },
     {
-      "epoch": 0.46,
-      "learning_rate": 1.798652644558236e-05,
-      "loss": 1.065,
+      "epoch": 2.95,
+      "learning_rate": 5.133645024651171e-06,
+      "loss": 0.3397,
       "step": 290
-    },
-    {
-      "epoch": 0.47,
-      "learning_rate": 1.790876920029647e-05,
-      "loss": 0.9168,
-      "step": 295
-    },
-    {
-      "epoch": 0.47,
-      "learning_rate": 1.7829713694717665e-05,
-      "loss": 1.0295,
-      "step": 300
-    },
-    {
-      "epoch": 0.48,
-      "learning_rate": 1.774937290616533e-05,
-      "loss": 0.9455,
-      "step": 305
-    },
-    {
-      "epoch": 0.49,
-      "learning_rate": 1.7667760022943864e-05,
-      "loss": 1.0272,
-      "step": 310
-    },
-    {
-      "epoch": 0.5,
-      "learning_rate": 1.7584888442177774e-05,
-      "loss": 1.127,
-      "step": 315
-    },
-    {
-      "epoch": 0.51,
-      "learning_rate": 1.7500771767612473e-05,
-      "loss": 1.0332,
-      "step": 320
-    },
-    {
-      "epoch": 0.51,
-      "learning_rate": 1.7415423807381162e-05,
-      "loss": 1.0603,
-      "step": 325
-    },
-    {
-      "epoch": 0.52,
-      "learning_rate": 1.7328858571738157e-05,
-      "loss": 1.1075,
-      "step": 330
-    },
-    {
-      "epoch": 0.53,
-      "learning_rate": 1.7241090270759055e-05,
-      "loss": 1.1279,
-      "step": 335
-    },
-    {
-      "epoch": 0.54,
-      "learning_rate": 1.715213331200807e-05,
-      "loss": 1.1037,
-      "step": 340
-    },
-    {
-      "epoch": 0.55,
-      "learning_rate": 1.7062002298172984e-05,
-      "loss": 0.9818,
-      "step": 345
-    },
-    {
-      "epoch": 0.55,
-      "learning_rate": 1.697071202466803e-05,
-      "loss": 0.9672,
-      "step": 350
-    },
-    {
-      "epoch": 0.56,
-      "learning_rate": 1.687827747720517e-05,
-      "loss": 0.8006,
-      "step": 355
-    },
-    {
-      "epoch": 0.57,
-      "learning_rate": 1.6784713829334124e-05,
-      "loss": 0.9823,
-      "step": 360
-    },
-    {
-      "epoch": 0.58,
-      "learning_rate": 1.6690036439951552e-05,
-      "loss": 1.0844,
-      "step": 365
-    },
-    {
-      "epoch": 0.59,
-      "learning_rate": 1.6594260850779837e-05,
-      "loss": 1.0858,
-      "step": 370
-    },
-    {
-      "epoch": 0.59,
-      "learning_rate": 1.6497402783815834e-05,
-      "loss": 0.9117,
-      "step": 375
-    },
-    {
-      "epoch": 0.6,
-      "learning_rate": 1.6399478138750015e-05,
-      "loss": 0.9313,
-      "step": 380
-    },
-    {
-      "epoch": 0.61,
-      "learning_rate": 1.63005029903565e-05,
-      "loss": 0.9776,
-      "step": 385
-    },
-    {
-      "epoch": 0.62,
-      "learning_rate": 1.620049358585427e-05,
-      "loss": 1.1445,
-      "step": 390
-    },
-    {
-      "epoch": 0.62,
-      "learning_rate": 1.609946634224015e-05,
-      "loss": 1.0567,
-      "step": 395
-    },
-    {
-      "epoch": 0.63,
-      "learning_rate": 1.5997437843593856e-05,
-      "loss": 0.9138,
-      "step": 400
-    },
-    {
-      "epoch": 0.64,
-      "learning_rate": 1.5894424838355654e-05,
-      "loss": 1.1003,
-      "step": 405
-    },
-    {
-      "epoch": 0.65,
-      "learning_rate": 1.5790444236577028e-05,
-      "loss": 1.0624,
-      "step": 410
-    },
-    {
-      "epoch": 0.66,
-      "learning_rate": 1.568551310714482e-05,
-      "loss": 0.8902,
-      "step": 415
-    },
-    {
-      "epoch": 0.66,
-      "learning_rate": 1.557964867497929e-05,
-      "loss": 1.0329,
-      "step": 420
-    },
-    {
-      "epoch": 0.67,
-      "learning_rate": 1.5472868318206566e-05,
-      "loss": 1.0023,
-      "step": 425
-    },
-    {
-      "epoch": 0.68,
-      "learning_rate": 1.5365189565305957e-05,
-      "loss": 0.9393,
-      "step": 430
-    },
-    {
-      "epoch": 0.69,
-      "learning_rate": 1.5256630092232567e-05,
-      "loss": 0.9747,
-      "step": 435
-    },
-    {
-      "epoch": 0.7,
-      "learning_rate": 1.5147207719515692e-05,
-      "loss": 1.0078,
-      "step": 440
-    },
-    {
-      "epoch": 0.7,
-      "learning_rate": 1.5036940409333533e-05,
-      "loss": 0.9576,
-      "step": 445
-    },
-    {
-      "epoch": 0.71,
-      "learning_rate": 1.4925846262564592e-05,
-      "loss": 0.992,
-      "step": 450
-    },
-    {
-      "epoch": 0.72,
-      "learning_rate": 1.4813943515816344e-05,
-      "loss": 1.0192,
-      "step": 455
-    },
-    {
-      "epoch": 0.73,
-      "learning_rate": 1.4701250538431617e-05,
-      "loss": 1.0154,
-      "step": 460
-    },
-    {
-      "epoch": 0.74,
-      "learning_rate": 1.4587785829473173e-05,
-      "loss": 1.1043,
-      "step": 465
-    },
-    {
-      "epoch": 0.74,
-      "learning_rate": 1.4473568014687018e-05,
-      "loss": 0.8351,
-      "step": 470
-    },
-    {
-      "epoch": 0.75,
-      "learning_rate": 1.4358615843444876e-05,
-      "loss": 1.0276,
-      "step": 475
-    },
-    {
-      "epoch": 0.76,
-      "learning_rate": 1.4242948185666419e-05,
-      "loss": 0.9423,
-      "step": 480
-    },
-    {
-      "epoch": 0.77,
-      "learning_rate": 1.4126584028721677e-05,
-      "loss": 0.9598,
-      "step": 485
-    },
-    {
-      "epoch": 0.77,
-      "learning_rate": 1.4009542474314173e-05,
-      "loss": 1.1755,
-      "step": 490
-    },
-    {
-      "epoch": 0.78,
-      "learning_rate": 1.3891842735345285e-05,
-      "loss": 1.1018,
-      "step": 495
-    },
-    {
-      "epoch": 0.79,
-      "learning_rate": 1.3773504132760379e-05,
-      "loss": 1.0445,
-      "step": 500
-    },
-    {
-      "epoch": 0.8,
-      "learning_rate": 1.3654546092377166e-05,
-      "loss": 0.9674,
-      "step": 505
-    },
-    {
-      "epoch": 0.81,
-      "learning_rate": 1.3534988141696891e-05,
-      "loss": 1.0473,
-      "step": 510
-    },
-    {
-      "epoch": 0.81,
-      "learning_rate": 1.3414849906698788e-05,
-      "loss": 0.9346,
-      "step": 515
-    },
-    {
-      "epoch": 0.82,
-      "learning_rate": 1.3294151108618379e-05,
-      "loss": 0.9643,
-      "step": 520
-    },
-    {
-      "epoch": 0.83,
-      "learning_rate": 1.3172911560710167e-05,
-      "loss": 1.113,
-      "step": 525
-    },
-    {
-      "epoch": 0.84,
-      "learning_rate": 1.3051151164995188e-05,
-      "loss": 1.0155,
-      "step": 530
-    },
-    {
-      "epoch": 0.85,
-      "learning_rate": 1.2928889908994003e-05,
-      "loss": 0.9675,
-      "step": 535
-    },
-    {
-      "epoch": 0.85,
-      "learning_rate": 1.280614786244566e-05,
-      "loss": 0.8424,
-      "step": 540
-    },
-    {
-      "epoch": 0.86,
-      "learning_rate": 1.2682945174013148e-05,
-      "loss": 1.1247,
-      "step": 545
-    },
-    {
-      "epoch": 0.87,
-      "learning_rate": 1.2559302067975914e-05,
-      "loss": 0.9641,
-      "step": 550
-    },
-    {
-      "epoch": 0.88,
-      "learning_rate": 1.243523884090995e-05,
-      "loss": 1.0329,
-      "step": 555
-    },
-    {
-      "epoch": 0.89,
-      "learning_rate": 1.2310775858356017e-05,
-      "loss": 1.0533,
-      "step": 560
-    },
-    {
-      "epoch": 0.89,
-      "learning_rate": 1.2185933551476545e-05,
-      "loss": 1.0216,
-      "step": 565
-    },
-    {
-      "epoch": 0.9,
-      "learning_rate": 1.2060732413701773e-05,
-      "loss": 1.0346,
-      "step": 570
-    },
-    {
-      "epoch": 0.91,
-      "learning_rate": 1.1935192997365666e-05,
-      "loss": 0.9749,
-      "step": 575
-    },
-    {
-      "epoch": 0.92,
-      "learning_rate": 1.1809335910332136e-05,
-      "loss": 1.0901,
-      "step": 580
-    },
-    {
-      "epoch": 0.93,
-      "learning_rate": 1.1683181812612186e-05,
-      "loss": 1.0043,
-      "step": 585
-    },
-    {
-      "epoch": 0.93,
-      "learning_rate": 1.1556751412972462e-05,
-      "loss": 0.9162,
-      "step": 590
-    },
-    {
-      "epoch": 0.94,
-      "learning_rate": 1.1430065465535827e-05,
-      "loss": 0.9228,
-      "step": 595
-    },
-    {
-      "epoch": 0.95,
-      "learning_rate": 1.1303144766374476e-05,
-      "loss": 0.9568,
-      "step": 600
-    },
-    {
-      "epoch": 0.96,
-      "learning_rate": 1.1176010150096158e-05,
-      "loss": 1.063,
-      "step": 605
-    },
-    {
-      "epoch": 0.96,
-      "learning_rate": 1.104868248642408e-05,
-      "loss": 0.9938,
-      "step": 610
-    },
-    {
-      "epoch": 0.97,
-      "learning_rate": 1.092118267677106e-05,
-      "loss": 1.1056,
-      "step": 615
-    },
-    {
-      "epoch": 0.98,
-      "learning_rate": 1.0793531650808469e-05,
-      "loss": 0.9269,
-      "step": 620
-    },
-    {
-      "epoch": 0.99,
-      "learning_rate": 1.0665750363030498e-05,
-      "loss": 1.0452,
-      "step": 625
-    },
-    {
-      "epoch": 1.0,
-      "learning_rate": 1.0537859789314424e-05,
-      "loss": 0.8855,
-      "step": 630
-    },
-    {
-      "epoch": 1.0,
-      "learning_rate": 1.0409880923477293e-05,
-      "loss": 1.0583,
-      "step": 635
-    },
-    {
-      "epoch": 1.01,
-      "learning_rate": 1.028183477382971e-05,
-      "loss": 0.8533,
-      "step": 640
-    },
-    {
-      "epoch": 1.02,
-      "learning_rate": 1.0153742359727226e-05,
-      "loss": 0.942,
-      "step": 645
-    },
-    {
-      "epoch": 1.03,
-      "learning_rate": 1.0025624708119901e-05,
-      "loss": 1.0044,
-      "step": 650
-    },
-    {
-      "epoch": 1.04,
-      "learning_rate": 9.897502850100648e-06,
-      "loss": 0.9652,
-      "step": 655
-    },
-    {
-      "epoch": 1.04,
-      "learning_rate": 9.76939781745289e-06,
-      "loss": 1.0202,
-      "step": 660
-    },
-    {
-      "epoch": 1.05,
-      "learning_rate": 9.641330639198083e-06,
-      "loss": 0.9401,
-      "step": 665
-    },
-    {
-      "epoch": 1.06,
-      "learning_rate": 9.513322338143714e-06,
-      "loss": 0.8343,
-      "step": 670
-    },
-    {
-      "epoch": 1.07,
-      "learning_rate": 9.385393927432307e-06,
-      "loss": 0.8904,
-      "step": 675
-    },
-    {
-      "epoch": 1.08,
-      "learning_rate": 9.257566407092032e-06,
-      "loss": 0.9143,
-      "step": 680
-    },
-    {
-      "epoch": 1.08,
-      "learning_rate": 9.129860760589441e-06,
-      "loss": 0.8408,
-      "step": 685
-    },
-    {
-      "epoch": 1.09,
-      "learning_rate": 9.002297951384945e-06,
-      "loss": 0.9426,
-      "step": 690
-    },
-    {
-      "epoch": 1.1,
-      "learning_rate": 8.874898919491564e-06,
-      "loss": 0.9476,
-      "step": 695
-    },
-    {
-      "epoch": 1.11,
-      "learning_rate": 8.74768457803754e-06,
-      "loss": 0.9992,
-      "step": 700
-    },
-    {
-      "epoch": 1.12,
-      "learning_rate": 8.62067580983333e-06,
-      "loss": 0.8491,
-      "step": 705
-    },
-    {
-      "epoch": 1.12,
-      "learning_rate": 8.493893463943617e-06,
-      "loss": 0.8829,
-      "step": 710
-    },
-    {
-      "epoch": 1.13,
-      "learning_rate": 8.367358352264834e-06,
-      "loss": 0.7933,
-      "step": 715
-    },
-    {
-      "epoch": 1.14,
-      "learning_rate": 8.241091246108796e-06,
-      "loss": 0.9976,
-      "step": 720
-    },
-    {
-      "epoch": 1.15,
-      "learning_rate": 8.115112872793006e-06,
-      "loss": 0.9947,
-      "step": 725
-    },
-    {
-      "epoch": 1.15,
-      "learning_rate": 7.989443912238151e-06,
-      "loss": 0.9171,
-      "step": 730
-    },
-    {
-      "epoch": 1.16,
-      "learning_rate": 7.864104993573422e-06,
-      "loss": 0.9348,
-      "step": 735
-    },
-    {
-      "epoch": 1.17,
-      "learning_rate": 7.73911669175013e-06,
-      "loss": 0.9357,
-      "step": 740
-    },
-    {
-      "epoch": 1.18,
-      "learning_rate": 7.614499524164251e-06,
-      "loss": 0.9273,
-      "step": 745
-    },
-    {
-      "epoch": 1.19,
-      "learning_rate": 7.490273947288389e-06,
-      "loss": 0.8991,
-      "step": 750
-    },
-    {
-      "epoch": 1.19,
-      "learning_rate": 7.366460353313762e-06,
-      "loss": 0.8373,
-      "step": 755
-    },
-    {
-      "epoch": 1.2,
-      "learning_rate": 7.2430790668027274e-06,
-      "loss": 0.8316,
-      "step": 760
-    },
-    {
-      "epoch": 1.21,
-      "learning_rate": 7.120150341352413e-06,
-      "loss": 0.7532,
-      "step": 765
-    },
-    {
-      "epoch": 1.22,
-      "learning_rate": 6.99769435627e-06,
-      "loss": 0.8851,
-      "step": 770
-    },
-    {
-      "epoch": 1.23,
-      "learning_rate": 6.875731213260193e-06,
-      "loss": 0.9678,
-      "step": 775
-    },
-    {
-      "epoch": 1.23,
-      "learning_rate": 6.754280933125441e-06,
-      "loss": 0.9968,
-      "step": 780
-    },
-    {
-      "epoch": 1.24,
-      "learning_rate": 6.633363452479431e-06,
-      "loss": 0.9126,
-      "step": 785
-    },
-    {
-      "epoch": 1.25,
-      "learning_rate": 6.512998620474396e-06,
-      "loss": 0.8765,
-      "step": 790
-    },
-    {
-      "epoch": 1.26,
-      "learning_rate": 6.393206195542791e-06,
-      "loss": 0.8761,
-      "step": 795
-    },
-    {
-      "epoch": 1.27,
-      "learning_rate": 6.27400584215386e-06,
-      "loss": 0.99,
-      "step": 800
-    },
-    {
-      "epoch": 1.27,
-      "learning_rate": 6.155417127585617e-06,
-      "loss": 0.7691,
-      "step": 805
-    },
-    {
-      "epoch": 1.28,
-      "learning_rate": 6.037459518712796e-06,
-      "loss": 0.7794,
-      "step": 810
-    },
-    {
-      "epoch": 1.29,
-      "learning_rate": 5.920152378811268e-06,
-      "loss": 0.9978,
-      "step": 815
-    },
-    {
-      "epoch": 1.3,
-      "learning_rate": 5.803514964379482e-06,
-      "loss": 0.8165,
-      "step": 820
-    },
-    {
-      "epoch": 1.31,
-      "learning_rate": 5.68756642197741e-06,
-      "loss": 0.8918,
-      "step": 825
-    },
-    {
-      "epoch": 1.31,
-      "learning_rate": 5.572325785083563e-06,
-      "loss": 0.9572,
-      "step": 830
-    },
-    {
-      "epoch": 1.32,
-      "learning_rate": 5.457811970970564e-06,
-      "loss": 0.9112,
-      "step": 835
-    },
-    {
-      "epoch": 1.33,
-      "learning_rate": 5.3440437775997636e-06,
-      "loss": 0.9659,
-      "step": 840
-    },
-    {
-      "epoch": 1.34,
-      "learning_rate": 5.231039880535511e-06,
-      "loss": 0.9516,
-      "step": 845
-    },
-    {
-      "epoch": 1.34,
-      "learning_rate": 5.118818829879442e-06,
-      "loss": 1.076,
-      "step": 850
-    },
-    {
-      "epoch": 1.35,
-      "learning_rate": 5.0073990472254075e-06,
-      "loss": 0.8136,
-      "step": 855
-    },
-    {
-      "epoch": 1.36,
-      "learning_rate": 4.8967988226354945e-06,
-      "loss": 0.9711,
-      "step": 860
-    },
-    {
-      "epoch": 1.37,
-      "learning_rate": 4.787036311637609e-06,
-      "loss": 0.8879,
-      "step": 865
-    },
-    {
-      "epoch": 1.38,
-      "learning_rate": 4.678129532245189e-06,
-      "loss": 1.0423,
-      "step": 870
-    },
-    {
-      "epoch": 1.38,
-      "learning_rate": 4.570096361999445e-06,
-      "loss": 0.892,
-      "step": 875
-    },
-    {
-      "epoch": 1.39,
-      "learning_rate": 4.462954535034692e-06,
-      "loss": 0.9188,
-      "step": 880
-    },
-    {
-      "epoch": 1.4,
-      "learning_rate": 4.356721639167202e-06,
-      "loss": 0.8706,
-      "step": 885
-    },
-    {
-      "epoch": 1.41,
-      "learning_rate": 4.251415113008096e-06,
-      "loss": 0.8926,
-      "step": 890
-    },
-    {
-      "epoch": 1.42,
-      "learning_rate": 4.147052243100706e-06,
-      "loss": 0.9126,
-      "step": 895
-    },
-    {
-      "epoch": 1.42,
-      "learning_rate": 4.043650161082913e-06,
-      "loss": 0.9201,
-      "step": 900
-    },
-    {
-      "epoch": 1.43,
-      "learning_rate": 3.941225840874925e-06,
-      "loss": 0.8758,
-      "step": 905
-    },
-    {
-      "epoch": 1.44,
-      "learning_rate": 3.839796095892905e-06,
-      "loss": 0.8818,
-      "step": 910
-    },
-    {
-      "epoch": 1.45,
-      "learning_rate": 3.7393775762889963e-06,
-      "loss": 0.9018,
-      "step": 915
-    },
-    {
-      "epoch": 1.46,
-      "learning_rate": 3.639986766218112e-06,
-      "loss": 0.8346,
-      "step": 920
-    },
-    {
-      "epoch": 1.46,
-      "learning_rate": 3.541639981131996e-06,
-      "loss": 0.9619,
-      "step": 925
-    },
-    {
-      "epoch": 1.47,
-      "learning_rate": 3.4443533651009474e-06,
-      "loss": 1.0202,
-      "step": 930
-    },
-    {
-      "epoch": 1.48,
-      "learning_rate": 3.348142888163726e-06,
-      "loss": 0.9186,
-      "step": 935
-    },
-    {
-      "epoch": 1.49,
-      "learning_rate": 3.2530243437059773e-06,
-      "loss": 0.9326,
-      "step": 940
-    },
-    {
-      "epoch": 1.49,
-      "learning_rate": 3.1590133458676787e-06,
-      "loss": 0.8347,
-      "step": 945
-    },
-    {
-      "epoch": 1.5,
-      "learning_rate": 3.066125326980027e-06,
-      "loss": 0.8914,
-      "step": 950
-    },
-    {
-      "epoch": 1.51,
-      "learning_rate": 2.9743755350321213e-06,
-      "loss": 0.8869,
-      "step": 955
-    },
-    {
-      "epoch": 1.52,
-      "learning_rate": 2.8837790311679625e-06,
-      "loss": 0.9833,
-      "step": 960
-    },
-    {
-      "epoch": 1.53,
-      "learning_rate": 2.7943506872140844e-06,
-      "loss": 0.8893,
-      "step": 965
-    },
-    {
-      "epoch": 1.53,
-      "learning_rate": 2.7061051832382836e-06,
-      "loss": 0.8616,
-      "step": 970
-    },
-    {
-      "epoch": 1.54,
-      "learning_rate": 2.6190570051398035e-06,
-      "loss": 0.9909,
-      "step": 975
-    },
-    {
-      "epoch": 1.55,
-      "learning_rate": 2.5332204422714368e-06,
-      "loss": 0.9515,
-      "step": 980
-    },
-    {
-      "epoch": 1.56,
-      "learning_rate": 2.4486095850938352e-06,
-      "loss": 1.0862,
-      "step": 985
-    },
-    {
-      "epoch": 1.57,
-      "learning_rate": 2.365238322862511e-06,
-      "loss": 0.9123,
-      "step": 990
-    },
-    {
-      "epoch": 1.57,
-      "learning_rate": 2.2831203413478555e-06,
-      "loss": 0.8965,
-      "step": 995
-    },
-    {
-      "epoch": 1.58,
-      "learning_rate": 2.202269120588546e-06,
-      "loss": 0.9261,
-      "step": 1000
-    },
-    {
-      "epoch": 1.59,
-      "learning_rate": 2.122697932678748e-06,
-      "loss": 0.9044,
-      "step": 1005
-    },
-    {
-      "epoch": 1.6,
-      "learning_rate": 2.0444198395894332e-06,
-      "loss": 0.923,
-      "step": 1010
-    },
-    {
-      "epoch": 1.61,
-      "learning_rate": 1.9674476910242055e-06,
-      "loss": 0.9581,
-      "step": 1015
-    },
-    {
-      "epoch": 1.61,
-      "learning_rate": 1.891794122309949e-06,
-      "loss": 1.0578,
-      "step": 1020
-    },
-    {
-      "epoch": 1.62,
-      "learning_rate": 1.8174715523227017e-06,
-      "loss": 0.886,
-      "step": 1025
-    },
-    {
-      "epoch": 1.63,
-      "learning_rate": 1.7444921814490256e-06,
-      "loss": 0.8365,
-      "step": 1030
-    },
-    {
-      "epoch": 1.64,
-      "learning_rate": 1.6728679895832622e-06,
-      "loss": 0.9913,
-      "step": 1035
-    },
-    {
-      "epoch": 1.65,
-      "learning_rate": 1.6026107341609842e-06,
-      "loss": 0.8921,
-      "step": 1040
-    },
-    {
-      "epoch": 1.65,
-      "learning_rate": 1.5337319482289503e-06,
-      "loss": 0.8323,
-      "step": 1045
-    },
-    {
-      "epoch": 1.66,
-      "learning_rate": 1.4662429385519084e-06,
-      "loss": 0.8869,
-      "step": 1050
-    },
-    {
-      "epoch": 1.67,
-      "learning_rate": 1.400154783756541e-06,
-      "loss": 0.9799,
-      "step": 1055
-    },
-    {
-      "epoch": 1.68,
-      "learning_rate": 1.3354783325128561e-06,
-      "loss": 0.9416,
-      "step": 1060
-    },
-    {
-      "epoch": 1.68,
-      "learning_rate": 1.2722242017533192e-06,
-      "loss": 0.9801,
-      "step": 1065
-    },
-    {
-      "epoch": 1.69,
-      "learning_rate": 1.2104027749300574e-06,
-      "loss": 0.9614,
-      "step": 1070
-    },
-    {
-      "epoch": 1.7,
-      "learning_rate": 1.150024200310348e-06,
-      "loss": 1.0792,
-      "step": 1075
-    },
-    {
-      "epoch": 1.71,
-      "learning_rate": 1.0910983893107419e-06,
-      "loss": 0.8622,
-      "step": 1080
-    },
-    {
-      "epoch": 1.72,
-      "learning_rate": 1.0336350148700668e-06,
-      "loss": 0.8627,
-      "step": 1085
-    },
-    {
-      "epoch": 1.72,
-      "learning_rate": 9.776435098615578e-07,
-      "loss": 0.857,
-      "step": 1090
-    },
-    {
-      "epoch": 1.73,
-      "learning_rate": 9.231330655444193e-07,
-      "loss": 1.0396,
-      "step": 1095
-    },
-    {
-      "epoch": 1.74,
-      "learning_rate": 8.701126300550322e-07,
-      "loss": 1.0174,
-      "step": 1100
-    },
-    {
-      "epoch": 1.75,
-      "learning_rate": 8.185909069380782e-07,
-      "loss": 1.031,
-      "step": 1105
-    },
-    {
-      "epoch": 1.76,
-      "learning_rate": 7.685763537178093e-07,
-      "loss": 0.9254,
-      "step": 1110
-    },
-    {
-      "epoch": 1.76,
-      "learning_rate": 7.200771805097206e-07,
-      "loss": 0.9294,
-      "step": 1115
-    },
-    {
-      "epoch": 1.77,
-      "learning_rate": 6.731013486728044e-07,
-      "loss": 0.8641,
-      "step": 1120
-    },
-    {
-      "epoch": 1.78,
-      "learning_rate": 6.276565695026671e-07,
-      "loss": 0.9669,
-      "step": 1125
-    },
-    {
-      "epoch": 1.79,
-      "learning_rate": 5.837503029656888e-07,
-      "loss": 0.8447,
-      "step": 1130
-    },
-    {
-      "epoch": 1.8,
-      "learning_rate": 5.413897564744253e-07,
-      "loss": 0.9566,
-      "step": 1135
-    },
-    {
-      "epoch": 1.8,
-      "learning_rate": 5.005818837044885e-07,
-      "loss": 0.9457,
-      "step": 1140
-    },
-    {
-      "epoch": 1.81,
-      "learning_rate": 4.613333834530631e-07,
-      "loss": 0.8523,
-      "step": 1145
-    },
-    {
-      "epoch": 1.82,
-      "learning_rate": 4.23650698539273e-07,
-      "loss": 0.9696,
-      "step": 1150
-    },
-    {
-      "epoch": 1.83,
-      "learning_rate": 3.8754001474655354e-07,
-      "loss": 0.7867,
-      "step": 1155
-    },
-    {
-      "epoch": 1.83,
-      "learning_rate": 3.530072598072454e-07,
-      "loss": 0.9534,
-      "step": 1160
-    },
-    {
-      "epoch": 1.84,
-      "learning_rate": 3.200581024295102e-07,
-      "loss": 0.9209,
-      "step": 1165
-    },
-    {
-      "epoch": 1.85,
-      "learning_rate": 2.886979513667998e-07,
-      "loss": 0.8257,
-      "step": 1170
-    },
-    {
-      "epoch": 1.86,
-      "learning_rate": 2.589319545299807e-07,
-      "loss": 0.9424,
-      "step": 1175
-    },
-    {
-      "epoch": 1.87,
-      "learning_rate": 2.3076499814227992e-07,
-      "loss": 0.873,
-      "step": 1180
-    },
-    {
-      "epoch": 1.87,
-      "learning_rate": 2.042017059371948e-07,
-      "loss": 0.843,
-      "step": 1185
-    },
-    {
-      "epoch": 1.88,
-      "learning_rate": 1.7924643839947632e-07,
-      "loss": 0.9136,
-      "step": 1190
-    },
-    {
-      "epoch": 1.89,
-      "learning_rate": 1.559032920493464e-07,
-      "loss": 1.0255,
-      "step": 1195
-    },
-    {
-      "epoch": 1.9,
-      "learning_rate": 1.3417609877002691e-07,
-      "loss": 0.8384,
-      "step": 1200
-    },
-    {
-      "epoch": 1.91,
-      "learning_rate": 1.1406842517872608e-07,
-      "loss": 0.7154,
-      "step": 1205
-    },
-    {
-      "epoch": 1.91,
-      "learning_rate": 9.558357204115464e-08,
-      "loss": 0.8805,
-      "step": 1210
-    },
-    {
-      "epoch": 1.92,
-      "learning_rate": 7.872457372969711e-08,
-      "loss": 0.7834,
-      "step": 1215
-    },
-    {
-      "epoch": 1.93,
-      "learning_rate": 6.34941977253023e-08,
-      "loss": 0.8752,
-      "step": 1220
-    },
-    {
-      "epoch": 1.94,
-      "learning_rate": 4.989494416318685e-08,
-      "loss": 0.9148,
-      "step": 1225
-    },
-    {
-      "epoch": 1.95,
-      "learning_rate": 3.7929045422432364e-08,
-      "loss": 0.8612,
-      "step": 1230
-    },
-    {
-      "epoch": 1.95,
-      "learning_rate": 2.7598465759526294e-08,
-      "loss": 0.9299,
-      "step": 1235
-    },
-    {
-      "epoch": 1.96,
-      "learning_rate": 1.8904900985918796e-08,
-      "loss": 0.9918,
-      "step": 1240
-    },
-    {
-      "epoch": 1.97,
-      "learning_rate": 1.184977818965205e-08,
-      "loss": 0.9008,
-      "step": 1245
-    },
-    {
-      "epoch": 1.98,
-      "learning_rate": 6.434255501095443e-09,
-      "loss": 0.9461,
-      "step": 1250
-    },
-    {
-      "epoch": 1.99,
-      "learning_rate": 2.659221902830966e-09,
-      "loss": 0.8181,
-      "step": 1255
-    },
-    {
-      "epoch": 1.99,
-      "learning_rate": 5.252970837255067e-10,
-      "loss": 0.8643,
-      "step": 1260
-    },
-    {
-      "epoch": 2.0,
-      "step": 1264,
-      "total_flos": 3.882961297263821e+17,
-      "train_loss": 0.9575079066466682,
-      "train_runtime": 77706.2136,
-      "train_samples_per_second": 0.26,
-      "train_steps_per_second": 0.016
     }
   ],
-  "max_steps": 1264,
-  "num_train_epochs": 2,
-  "total_flos": 3.882961297263821e+17,
+  "max_steps": 392,
+  "num_train_epochs": 4,
+  "total_flos": 1.821325738775675e+17,
   "trial_name": null,
   "trial_params": null
 }

training_args.bin

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:c2c8d84c6ea850e743bbd2858bd2ac8b6913bd2628de22bb48d811f906d778c2
+oid sha256:cb4937467d6816f4a87c65a795706d60d5c1945956041be1f6697c8ed7d29b1c
 size 4399

zero_to_fp32.py

@@ -0,0 +1,484 @@
+#!/usr/bin/env python
+
+# This script extracts fp32 consolidated weights from a zero 2 and 3 DeepSpeed checkpoints. It gets
+# copied into the top level checkpoint dir, so the user can easily do the conversion at any point in
+# the future. Once extracted, the weights don't require DeepSpeed and can be used in any
+# application.
+#
+# example: python zero_to_fp32.py . pytorch_model.bin
+
+import argparse
+import torch
+import glob
+import math
+import os
+import re
+from collections import OrderedDict
+
+# while this script doesn't use deepspeed to recover data, since the checkpoints are pickled with
+# DeepSpeed data structures it has to be available in the current python environment.
+import deepspeed
+from deepspeed.utils import logger
+from deepspeed.checkpoint.constants import (DS_VERSION,
+                                            OPTIMIZER_STATE_DICT,
+                                            PARAM_SHAPES,
+                                            SINGLE_PARTITION_OF_FP32_GROUPS,
+                                            FP32_FLAT_GROUPS,
+                                            ZERO_STAGE,
+                                            PARTITION_COUNT,
+                                            PARAM_SHAPES,
+                                            BUFFER_NAMES)
+
+debug = 0
+
+# load to cpu
+device = torch.device('cpu')
+
+
+def atoi(text):
+    return int(text) if text.isdigit() else text
+
+
+def natural_keys(text):
+    '''
+    alist.sort(key=natural_keys) sorts in human order
+    http://nedbatchelder.com/blog/200712/human_sorting.html
+    (See Toothy's implementation in the comments)
+    '''
+    return [atoi(c) for c in re.split(r'(\d+)', text)]
+
+
+def get_model_state_file(checkpoint_dir, zero_stage):
+    if not os.path.isdir(checkpoint_dir):
+        raise FileNotFoundError(f"Directory '{checkpoint_dir}' doesn't exist")
+
+    # there should be only one file
+    if zero_stage == 2:
+        file = os.path.join(checkpoint_dir, "mp_rank_00_model_states.pt")
+    elif zero_stage == 3:
+        file = os.path.join(checkpoint_dir, "zero_pp_rank_0_mp_rank_00_model_states.pt")
+
+    if not os.path.exists(file):
+        raise FileNotFoundError(f"can't find model states file at '{file}'")
+
+    return file
+
+
+def get_optim_files(checkpoint_dir):
+    # XXX: need to test that this simple glob rule works for multi-node setup too
+    optim_files = sorted(glob.glob(os.path.join(checkpoint_dir,
+                                                "*_optim_states.pt")),
+                         key=natural_keys)
+
+    if len(optim_files) == 0:
+        raise FileNotFoundError(
+            f"can't find '*_optim_states.pt' files in directory '{checkpoint_dir}'")
+
+    return optim_files
+
+
+def parse_model_state(file):
+    state_dict = torch.load(file, map_location=device)
+
+    if BUFFER_NAMES not in state_dict:
+        raise ValueError(f"{file} is not a model state checkpoint")
+    buffer_names = state_dict[BUFFER_NAMES]
+    if debug:
+        print("Found buffers:", buffer_names)
+
+    # recover just the buffers while restoring them to fp32 if they were saved in fp16
+    buffers = {
+        k: v.float()
+        for k,
+        v in state_dict["module"].items() if k in buffer_names
+    }
+    param_shapes = state_dict[PARAM_SHAPES]
+
+    ds_version = state_dict.get(DS_VERSION, None)
+
+    return buffers, param_shapes, ds_version
+
+
+def parse_optim_states(files, ds_checkpoint_dir):
+
+    total_files = len(files)
+    state_dicts = []
+    for f in files:
+        state_dicts.append(torch.load(f, map_location=device))
+
+    if not ZERO_STAGE in state_dicts[0][OPTIMIZER_STATE_DICT]:
+        raise ValueError(f"{files[0]} is not a zero checkpoint")
+    zero_stage = state_dicts[0][OPTIMIZER_STATE_DICT][ZERO_STAGE]
+    world_size = state_dicts[0][OPTIMIZER_STATE_DICT][PARTITION_COUNT]
+
+    # For ZeRO-2 each param group can have different partition_count as data parallelism for expert
+    # parameters can be different from data parallelism for non-expert parameters. So we can just
+    # use the max of the partition_count to get the dp world_size.
+
+    if type(world_size) is list:
+        world_size = max(world_size)
+
+    if world_size != total_files:
+        raise ValueError(
+            f"Expected {world_size} of '*_optim_states.pt' under '{ds_checkpoint_dir}' but found {total_files} files. "
+            "Possibly due to an overwrite of an old checkpoint, or a checkpoint didn't get saved by one or more processes."
+        )
+
+    # the groups are named differently in each stage
+    if zero_stage == 2:
+        fp32_groups_key = SINGLE_PARTITION_OF_FP32_GROUPS
+    elif zero_stage == 3:
+        fp32_groups_key = FP32_FLAT_GROUPS
+    else:
+        raise ValueError(f"unknown zero stage {zero_stage}")
+
+    if zero_stage == 2:
+        fp32_flat_groups = [
+            state_dicts[i][OPTIMIZER_STATE_DICT][fp32_groups_key]
+            for i in range(len(state_dicts))
+        ]
+    elif zero_stage == 3:
+        # if there is more than one param group, there will be multiple flattened tensors - one
+        # flattened tensor per group - for simplicity merge them into a single tensor
+        #
+        # XXX: could make the script more memory efficient for when there are multiple groups - it
+        # will require matching the sub-lists of param_shapes for each param group flattened tensor
+
+        fp32_flat_groups = [
+            torch.cat(state_dicts[i][OPTIMIZER_STATE_DICT][fp32_groups_key],
+                      0) for i in range(len(state_dicts))
+        ]
+
+    return zero_stage, world_size, fp32_flat_groups
+
+
+def _get_fp32_state_dict_from_zero_checkpoint(ds_checkpoint_dir):
+    """
+    Returns fp32 state_dict reconstructed from ds checkpoint
+
+    Args:
+        - ``ds_checkpoint_dir``: path to the deepspeed checkpoint folder (where the optimizer files are)
+
+    """
+    print(f"Processing zero checkpoint '{ds_checkpoint_dir}'")
+
+    optim_files = get_optim_files(ds_checkpoint_dir)
+    zero_stage, world_size, fp32_flat_groups = parse_optim_states(optim_files, ds_checkpoint_dir)
+    print(
+        f"Detected checkpoint of type zero stage {zero_stage}, world_size: {world_size}")
+
+    model_file = get_model_state_file(ds_checkpoint_dir, zero_stage)
+    buffers, param_shapes, ds_version = parse_model_state(model_file)
+    print(f'Parsing checkpoint created by deepspeed=={ds_version}')
+
+    if zero_stage == 2:
+        return _get_fp32_state_dict_from_zero2_checkpoint(world_size,
+                                                          param_shapes,
+                                                          fp32_flat_groups,
+                                                          buffers)
+    elif zero_stage == 3:
+        return _get_fp32_state_dict_from_zero3_checkpoint(world_size,
+                                                          param_shapes,
+                                                          fp32_flat_groups,
+                                                          buffers)
+
+
+def _get_fp32_state_dict_from_zero2_checkpoint(world_size,
+                                               param_shapes,
+                                               fp32_flat_groups,
+                                               buffers):
+
+    # Reconstruction protocol:
+    #
+    # XXX: document this
+
+    if debug:
+        for i in range(world_size):
+            for j in range(len(fp32_flat_groups[0])):
+                print(
+                    f"{FP32_FLAT_GROUPS}[{i}][{j}].shape={fp32_flat_groups[i][j].shape}")
+
+    # XXX: memory usage doubles here (zero2)
+    num_param_groups = len(fp32_flat_groups[0])
+    merged_single_partition_of_fp32_groups = []
+    for i in range(num_param_groups):
+        merged_partitions = [sd[i] for sd in fp32_flat_groups]
+        full_single_fp32_vector = torch.cat(merged_partitions, 0)
+        merged_single_partition_of_fp32_groups.append(full_single_fp32_vector)
+    avail_numel = sum([
+        full_single_fp32_vector.numel()
+        for full_single_fp32_vector in merged_single_partition_of_fp32_groups
+    ])
+
+    if debug:
+        wanted_params = sum([len(shapes) for shapes in param_shapes])
+        wanted_numel = sum(
+            [sum(shape.numel() for shape in shapes.values()) for shapes in param_shapes])
+        # not asserting if there is a mismatch due to possible padding
+        print(f"Have {avail_numel} numels to process.")
+        print(f"Need {wanted_numel} numels in {wanted_params} params.")
+
+    state_dict = OrderedDict()
+
+    # buffers
+    state_dict.update(buffers)
+    if debug:
+        print(f"added {len(buffers)} buffers")
+
+    # params
+    # XXX: for huge models that can't fit into the host's RAM we will have to recode this to support
+    # out-of-core computing solution
+    total_numel = 0
+    total_params = 0
+    for shapes, full_single_fp32_vector in zip(param_shapes, merged_single_partition_of_fp32_groups):
+        offset = 0
+        avail_numel = full_single_fp32_vector.numel()
+        for name, shape in shapes.items():
+
+            unpartitioned_numel = shape.numel()
+            total_numel += unpartitioned_numel
+            total_params += 1
+
+            if debug:
+                print(
+                    f"{name} full shape: {shape} unpartitioned numel {unpartitioned_numel} "
+                )
+            state_dict[name] = full_single_fp32_vector.narrow(
+                0,
+                offset,
+                unpartitioned_numel).view(shape)
+            offset += unpartitioned_numel
+
+        # Z2 started to align to 2*world_size to improve nccl performance. Therefore both offset and
+        # avail_numel can differ by anywhere between 0..2*world_size. Due to two unrelated complex
+        # paddings performed in the code it's almost impossible to predict the exact numbers w/o the
+        # live optimizer object, so we are checking that the numbers are within the right range
+        align_to = 2 * world_size
+
+        def zero2_align(x):
+            return align_to * math.ceil(x / align_to)
+
+        if debug:
+            print(f"original offset={offset}, avail_numel={avail_numel}")
+
+        offset = zero2_align(offset)
+        avail_numel = zero2_align(avail_numel)
+
+        if debug:
+            print(f"aligned  offset={offset}, avail_numel={avail_numel}")
+
+        # Sanity check
+        if offset != avail_numel:
+            raise ValueError(
+                f"consumed {offset} numels out of {avail_numel} - something is wrong")
+
+    print(
+        f"Reconstructed fp32 state dict with {total_params} params {total_numel} elements"
+    )
+
+    return state_dict
+
+
+def zero3_partitioned_param_info(unpartitioned_numel, world_size):
+    remainder = unpartitioned_numel % world_size
+    padding_numel = (world_size - remainder) if remainder else 0
+    partitioned_numel = math.ceil(unpartitioned_numel / world_size)
+    return partitioned_numel, padding_numel
+
+
+def _get_fp32_state_dict_from_zero3_checkpoint(world_size,
+                                               param_shapes,
+                                               fp32_flat_groups,
+                                               buffers):
+
+    # Reconstruction protocol: For zero3 we need to zip the partitions together at boundary of each
+    # param, re-consolidating each param, while dealing with padding if any
+
+    avail_numel = fp32_flat_groups[0].numel() * world_size
+    # merge list of dicts, preserving order
+    param_shapes = {k: v for d in param_shapes for k, v in d.items()}
+
+    if debug:
+        for i in range(world_size):
+            print(f"{FP32_FLAT_GROUPS}[{i}].shape={fp32_flat_groups[i].shape}")
+
+        wanted_params = len(param_shapes)
+        wanted_numel = sum(shape.numel() for shape in param_shapes.values())
+        # not asserting if there is a mismatch due to possible padding
+        print(f"Have {avail_numel} numels to process.")
+        print(f"Need {wanted_numel} numels in {wanted_params} params.")
+
+    state_dict = OrderedDict()
+
+    # buffers
+    state_dict.update(buffers)
+    if debug:
+        print(f"added {len(buffers)} buffers")
+
+    # params
+    # XXX: for huge models that can't fit into the host's RAM we will have to recode this to support
+    # out-of-core computing solution
+    offset = 0
+    total_numel = 0
+    total_params = 0
+    for name, shape in param_shapes.items():
+
+        unpartitioned_numel = shape.numel()
+        total_numel += unpartitioned_numel
+        total_params += 1
+
+        partitioned_numel, partitioned_padding_numel = zero3_partitioned_param_info(unpartitioned_numel, world_size)
+
+        if debug:
+            print(
+                f"{total_params} {name} full shape: {shape} partition0 numel={partitioned_numel} partitioned_padding_numel={partitioned_padding_numel}"
+            )
+
+        # XXX: memory usage doubles here
+        state_dict[name] = torch.cat(
+            tuple(fp32_flat_groups[i].narrow(0,
+                                             offset,
+                                             partitioned_numel)
+                  for i in range(world_size)),
+            0).narrow(0,
+                      0,
+                      unpartitioned_numel).view(shape)
+        offset += partitioned_numel
+
+    offset *= world_size
+
+    # Sanity check
+    if offset != avail_numel:
+        raise ValueError(
+            f"consumed {offset} numels out of {avail_numel} - something is wrong")
+
+    print(
+        f"Reconstructed fp32 state dict with {total_params} params {total_numel} elements"
+    )
+
+    return state_dict
+
+
+def get_fp32_state_dict_from_zero_checkpoint(checkpoint_dir, tag=None):
+    """
+    Convert ZeRO 2 or 3 checkpoint into a single fp32 consolidated state_dict that can be loaded with
+    ``load_state_dict()`` and used for training without DeepSpeed or shared with others, for example
+    via a model hub.
+
+    Args:
+        - ``checkpoint_dir``: path to the desired checkpoint folder
+        - ``tag``: checkpoint tag used as a unique identifier for checkpoint. If not provided will attempt to load tag in 'latest' file. e.g., ``global_step14``
+
+    Returns:
+        - pytorch ``state_dict``
+
+    Note: this approach may not work if your application doesn't have sufficient free CPU memory and
+    you may need to use the offline approach using the ``zero_to_fp32.py`` script that is saved with
+    the checkpoint.
+
+    A typical usage might be ::
+
+        from deepspeed.utils.zero_to_fp32 import get_fp32_state_dict_from_zero_checkpoint
+        # do the training and checkpoint saving
+        state_dict = get_fp32_state_dict_from_zero_checkpoint(checkpoint_dir) # already on cpu
+        model = model.cpu() # move to cpu
+        model.load_state_dict(state_dict)
+        # submit to model hub or save the model to share with others
+
+    In this example the ``model`` will no longer be usable in the deepspeed context of the same
+    application. i.e. you will need to re-initialize the deepspeed engine, since
+    ``model.load_state_dict(state_dict)`` will remove all the deepspeed magic from it.
+
+    If you want it all done for you, use ``load_state_dict_from_zero_checkpoint`` instead.
+
+    """
+    if tag is None:
+        latest_path = os.path.join(checkpoint_dir, 'latest')
+        if os.path.isfile(latest_path):
+            with open(latest_path, 'r') as fd:
+                tag = fd.read().strip()
+        else:
+            raise ValueError(f"Unable to find 'latest' file at {latest_path}")
+
+    ds_checkpoint_dir = os.path.join(checkpoint_dir, tag)
+
+    if not os.path.isdir(ds_checkpoint_dir):
+        raise FileNotFoundError(f"Directory '{ds_checkpoint_dir}' doesn't exist")
+
+    return _get_fp32_state_dict_from_zero_checkpoint(ds_checkpoint_dir)
+
+
+def convert_zero_checkpoint_to_fp32_state_dict(checkpoint_dir, output_file, tag=None):
+    """
+    Convert ZeRO 2 or 3 checkpoint into a single fp32 consolidated ``state_dict`` file that can be
+    loaded with ``torch.load(file)`` + ``load_state_dict()`` and used for training without DeepSpeed.
+
+    Args:
+        - ``checkpoint_dir``: path to the desired checkpoint folder. (one that contains the tag-folder, like ``global_step14``)
+        - ``output_file``: path to the pytorch fp32 state_dict output file (e.g. path/pytorch_model.bin)
+        - ``tag``: checkpoint tag used as a unique identifier for checkpoint. If not provided will attempt to load tag in the file named ``latest`` in the checkpoint folder, e.g., ``global_step14``
+    """
+
+    state_dict = get_fp32_state_dict_from_zero_checkpoint(checkpoint_dir, tag)
+    print(f"Saving fp32 state dict to {output_file}")
+    torch.save(state_dict, output_file)
+
+
+def load_state_dict_from_zero_checkpoint(model, checkpoint_dir, tag=None):
+    """
+    1. Put the provided model to cpu
+    2. Convert ZeRO 2 or 3 checkpoint into a single fp32 consolidated ``state_dict``
+    3. Load it into the provided model
+
+    Args:
+        - ``model``: the model object to update
+        - ``checkpoint_dir``: path to the desired checkpoint folder. (one that contains the tag-folder, like ``global_step14``)
+        - ``tag``: checkpoint tag used as a unique identifier for checkpoint. If not provided will attempt to load tag in the file named ``latest`` in the checkpoint folder, e.g., ``global_step14``
+
+    Returns:
+        - ``model`: modified model
+
+    Make sure you have plenty of CPU memory available before you call this function. If you don't
+    have enough use the ``zero_to_fp32.py`` utility to do the conversion. You will find it
+    conveniently placed for you in the checkpoint folder.
+
+    A typical usage might be ::
+
+        from deepspeed.utils.zero_to_fp32 import load_state_dict_from_zero_checkpoint
+        model = load_state_dict_from_zero_checkpoint(trainer.model, checkpoint_dir)
+        # submit to model hub or save the model to share with others
+
+    Note, that once this was run, the ``model`` will no longer be usable in the deepspeed context
+    of the same application. i.e. you will need to re-initialize the deepspeed engine, since
+    ``model.load_state_dict(state_dict)`` will remove all the deepspeed magic from it.
+
+    """
+    logger.info(f"Extracting fp32 weights")
+    state_dict = get_fp32_state_dict_from_zero_checkpoint(checkpoint_dir, tag)
+
+    logger.info(f"Overwriting model with fp32 weights")
+    model = model.cpu()
+    model.load_state_dict(state_dict, strict=False)
+
+    return model
+
+
+if __name__ == "__main__":
+
+    parser = argparse.ArgumentParser()
+    parser.add_argument(
+        "checkpoint_dir",
+        type=str,
+        help="path to the desired checkpoint folder, e.g., path/checkpoint-12")
+    parser.add_argument(
+        "output_file",
+        type=str,
+        help=
+        "path to the pytorch fp32 state_dict output file (e.g. path/checkpoint-12/pytorch_model.bin)"
+    )
+    parser.add_argument("-d", "--debug", action='store_true', help="enable debug")
+    args = parser.parse_args()
+
+    debug = args.debug
+
+    convert_zero_checkpoint_to_fp32_state_dict(args.checkpoint_dir, args.output_file)