Training in progress, step 100000, checkpoint

checkpoint-100000/config.json

@@ -1,32 +1,28 @@
 {
-  "activation_function": "gelu_new",
   "architectures": [
-    "GPT2LMHeadModel"
+    "RobertaForMaskedLM"
   ],
-  "attn_pdrop": 0.1,
+  "attention_probs_dropout_prob": 0.1,
   "bos_token_id": 0,
-  "embd_pdrop": 0.1,
+  "classifier_dropout": null,
   "eos_token_id": 1,
+  "hidden_act": "gelu",
+  "hidden_dropout_prob": 0.1,
+  "hidden_size": 256,
   "initializer_range": 0.02,
-  "layer_norm_epsilon": 1e-05,
-  "model_type": "gpt2",
-  "n_ctx": 512,
-  "n_embd": 256,
-  "n_head": 8,
-  "n_inner": 2048,
-  "n_layer": 8,
-  "n_positions": 512,
-  "reorder_and_upcast_attn": false,
-  "resid_pdrop": 0.1,
-  "scale_attn_by_inverse_layer_idx": false,
-  "scale_attn_weights": true,
-  "summary_activation": null,
-  "summary_first_dropout": 0.1,
-  "summary_proj_to_labels": true,
-  "summary_type": "cls_index",
-  "summary_use_proj": true,
+  "intermediate_size": 2048,
+  "layer_norm_eps": 1e-05,
+  "mask_token_id": 4,
+  "max_position_embeddings": 512,
+  "model_type": "roberta",
+  "num_attention_heads": 8,
+  "num_hidden_layers": 8,
+  "pad_token_id": 1,
+  "position_embedding_type": "absolute",
+  "tie_word_embeddings": false,
   "torch_dtype": "float32",
   "transformers_version": "4.45.2",
+  "type_vocab_size": 2,
   "use_cache": true,
   "vocab_size": 8192
 }

checkpoint-100000/model.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:c4055a337df2fd66617953610f35e72d1d1cbe9aae95395b2369475ef4520440
-size 51007160
+oid sha256:90b6015e1587475adb03a8f817875234ac28669dfcef50dbc1f43a4a3580e743
+size 59702184

checkpoint-100000/optimizer.pt

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:5be84cdd8a8b584f79c31d4bbe2760b9fd729eb1c7b67ce43edf5df135174973
-size 102078202
+oid sha256:6b06e9be14186869ec4131db44f91bf5f977ad5cccec0e475af695b8e64623fd
+size 119488058

checkpoint-100000/rng_state.pth

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:c0cc673951a7e857c303bc02e2b38fc179aac1c11a91d4fb17050ba5e3e6dc23
+oid sha256:fadd05ce0ffc734c64b1890c62a97168e0d033ad98b04ea76441b8cfbd7cfffd
 size 14308

checkpoint-100000/scheduler.pt

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:f69dc9ccdedaf3bfab458442f0bfb1377fc846c74b83d1cda9ba9247794ae3de
+oid sha256:cb98843300b5c35351a1d68de5818543c07387c84a09c3f6ded07dd086e6d493
 size 1000

checkpoint-100000/special_tokens_map.json

@@ -23,7 +23,13 @@
     "rstrip": false,
     "single_word": false
   },
-  "pad_token": "</s>",
+  "pad_token": {
+    "content": "<pad>",
+    "lstrip": false,
+    "normalized": false,
+    "rstrip": false,
+    "single_word": false
+  },
   "unk_token": {
     "content": "<unk>",
     "lstrip": false,

checkpoint-100000/tokenizer_config.json

@@ -57,7 +57,7 @@
   "eos_token": "</s>",
   "mask_token": "<mask>",
   "model_max_length": 128,
-  "pad_token": "</s>",
+  "pad_token": "<pad>",
   "tokenizer_class": "PreTrainedTokenizerFast",
   "unk_token": "<unk>"
 }

checkpoint-100000/trainer_state.json

@@ -1,7 +1,7 @@
 {
-  "best_metric": 3.9619035720825195,
-  "best_model_checkpoint": "/home/p318482/babyLM_controlled/models_trained_last/de_clm/de_childes_30/checkpoint-44000",
-  "epoch": 75.1190386781825,
+  "best_metric": 2.6056320667266846,
+  "best_model_checkpoint": "/home/p318482/babyLM_controlled/models_trained_last/de_mlm/de_childes_30/checkpoint-100000",
+  "epoch": 75.10326699211416,
   "eval_steps": 2000,
   "global_step": 100000,
   "is_hyper_param_search": false,
@@ -10,577 +10,577 @@
   "log_history": [
     {
       "epoch": 1.5020653398422832,
-      "eval_loss": 7.09099817276001,
-      "eval_runtime": 3.7115,
-      "eval_samples_per_second": 894.522,
-      "eval_steps_per_second": 56.042,
+      "eval_loss": 7.472109317779541,
+      "eval_runtime": 2.5707,
+      "eval_samples_per_second": 1291.498,
+      "eval_steps_per_second": 80.913,
       "step": 2000
     },
     {
       "epoch": 3.0041306796845664,
-      "grad_norm": 1.3897417783737183,
+      "grad_norm": 0.9529591798782349,
       "learning_rate": 1e-05,
-      "loss": 6.9957,
+      "loss": 7.3851,
       "step": 4000
     },
     {
       "epoch": 3.0041306796845664,
-      "eval_loss": 5.826653003692627,
-      "eval_runtime": 3.7561,
-      "eval_samples_per_second": 883.906,
-      "eval_steps_per_second": 55.377,
+      "eval_loss": 6.4175896644592285,
+      "eval_runtime": 2.5761,
+      "eval_samples_per_second": 1288.76,
+      "eval_steps_per_second": 80.742,
       "step": 4000
     },
     {
       "epoch": 4.50619601952685,
-      "eval_loss": 5.465098857879639,
-      "eval_runtime": 3.8583,
-      "eval_samples_per_second": 860.484,
-      "eval_steps_per_second": 53.91,
+      "eval_loss": 6.3026041984558105,
+      "eval_runtime": 2.5876,
+      "eval_samples_per_second": 1283.023,
+      "eval_steps_per_second": 80.382,
       "step": 6000
     },
     {
       "epoch": 6.008261359369133,
-      "grad_norm": 2.3630881309509277,
+      "grad_norm": 1.5569523572921753,
       "learning_rate": 2e-05,
-      "loss": 5.213,
+      "loss": 6.0706,
       "step": 8000
     },
     {
       "epoch": 6.008261359369133,
-      "eval_loss": 5.171680927276611,
-      "eval_runtime": 3.928,
-      "eval_samples_per_second": 845.218,
-      "eval_steps_per_second": 52.953,
+      "eval_loss": 6.200850486755371,
+      "eval_runtime": 2.6505,
+      "eval_samples_per_second": 1252.599,
+      "eval_steps_per_second": 78.476,
       "step": 8000
     },
     {
       "epoch": 7.510326699211416,
-      "eval_loss": 4.959029197692871,
-      "eval_runtime": 3.8954,
-      "eval_samples_per_second": 852.288,
-      "eval_steps_per_second": 53.396,
+      "eval_loss": 6.105844497680664,
+      "eval_runtime": 2.5888,
+      "eval_samples_per_second": 1282.425,
+      "eval_steps_per_second": 80.345,
       "step": 10000
     },
     {
       "epoch": 9.0123920390537,
-      "grad_norm": 2.578948736190796,
-      "learning_rate": 2.9995e-05,
-      "loss": 4.7279,
+      "grad_norm": 2.3049111366271973,
+      "learning_rate": 3e-05,
+      "loss": 5.8733,
       "step": 12000
     },
     {
       "epoch": 9.0123920390537,
-      "eval_loss": 4.791839599609375,
-      "eval_runtime": 3.7027,
-      "eval_samples_per_second": 896.638,
-      "eval_steps_per_second": 56.175,
+      "eval_loss": 6.0330352783203125,
+      "eval_runtime": 2.5874,
+      "eval_samples_per_second": 1283.132,
+      "eval_steps_per_second": 80.389,
       "step": 12000
     },
     {
       "epoch": 10.514457378895981,
-      "eval_loss": 4.660107612609863,
-      "eval_runtime": 3.7788,
-      "eval_samples_per_second": 878.594,
-      "eval_steps_per_second": 55.044,
+      "eval_loss": 5.968105792999268,
+      "eval_runtime": 2.6066,
+      "eval_samples_per_second": 1273.702,
+      "eval_steps_per_second": 79.798,
       "step": 14000
     },
     {
       "epoch": 12.016522718738265,
-      "grad_norm": 2.3461837768554688,
-      "learning_rate": 3.999e-05,
-      "loss": 4.4214,
+      "grad_norm": 2.316237688064575,
+      "learning_rate": 4e-05,
+      "loss": 5.723,
       "step": 16000
     },
     {
       "epoch": 12.016522718738265,
-      "eval_loss": 4.552372932434082,
-      "eval_runtime": 3.7928,
-      "eval_samples_per_second": 875.332,
-      "eval_steps_per_second": 54.84,
+      "eval_loss": 5.878941535949707,
+      "eval_runtime": 2.5723,
+      "eval_samples_per_second": 1290.696,
+      "eval_steps_per_second": 80.863,
       "step": 16000
     },
     {
       "epoch": 13.518588058580548,
-      "eval_loss": 4.458698272705078,
-      "eval_runtime": 3.8109,
-      "eval_samples_per_second": 871.192,
-      "eval_steps_per_second": 54.581,
+      "eval_loss": 5.8198137283325195,
+      "eval_runtime": 2.5819,
+      "eval_samples_per_second": 1285.857,
+      "eval_steps_per_second": 80.56,
       "step": 18000
     },
     {
       "epoch": 15.020653398422832,
-      "grad_norm": 2.1618399620056152,
-      "learning_rate": 4.9985e-05,
-      "loss": 4.195,
+      "grad_norm": 2.299781322479248,
+      "learning_rate": 5e-05,
+      "loss": 5.6127,
       "step": 20000
     },
     {
       "epoch": 15.020653398422832,
-      "eval_loss": 4.3683576583862305,
-      "eval_runtime": 3.8444,
-      "eval_samples_per_second": 863.584,
-      "eval_steps_per_second": 54.104,
+      "eval_loss": 5.802139759063721,
+      "eval_runtime": 2.6545,
+      "eval_samples_per_second": 1250.713,
+      "eval_steps_per_second": 78.358,
       "step": 20000
     },
     {
       "epoch": 16.522718738265116,
-      "eval_loss": 4.289999485015869,
-      "eval_runtime": 3.8372,
-      "eval_samples_per_second": 865.217,
-      "eval_steps_per_second": 54.206,
+      "eval_loss": 5.7662129402160645,
+      "eval_runtime": 2.5877,
+      "eval_samples_per_second": 1282.983,
+      "eval_steps_per_second": 80.38,
       "step": 22000
     },
     {
       "epoch": 18.0247840781074,
-      "grad_norm": 2.0583324432373047,
-      "learning_rate": 5.9980000000000005e-05,
-      "loss": 4.0109,
+      "grad_norm": 2.8674566745758057,
+      "learning_rate": 6e-05,
+      "loss": 5.5325,
       "step": 24000
     },
     {
       "epoch": 18.0247840781074,
-      "eval_loss": 4.227661609649658,
-      "eval_runtime": 3.7524,
-      "eval_samples_per_second": 884.777,
-      "eval_steps_per_second": 55.432,
+      "eval_loss": 5.731865406036377,
+      "eval_runtime": 2.5813,
+      "eval_samples_per_second": 1286.16,
+      "eval_steps_per_second": 80.579,
       "step": 24000
     },
     {
       "epoch": 19.52684941794968,
-      "eval_loss": 4.172571659088135,
-      "eval_runtime": 3.7915,
-      "eval_samples_per_second": 875.64,
-      "eval_steps_per_second": 54.859,
+      "eval_loss": 5.703192234039307,
+      "eval_runtime": 2.5846,
+      "eval_samples_per_second": 1284.544,
+      "eval_steps_per_second": 80.477,
       "step": 26000
     },
     {
       "epoch": 21.028914757791963,
-      "grad_norm": 2.0024306774139404,
-      "learning_rate": 6.997500000000001e-05,
-      "loss": 3.8596,
+      "grad_norm": 4.573985576629639,
+      "learning_rate": 7e-05,
+      "loss": 5.4632,
       "step": 28000
     },
     {
       "epoch": 21.028914757791963,
-      "eval_loss": 4.128843307495117,
-      "eval_runtime": 3.6349,
-      "eval_samples_per_second": 913.374,
-      "eval_steps_per_second": 57.223,
+      "eval_loss": 5.70460319519043,
+      "eval_runtime": 2.5824,
+      "eval_samples_per_second": 1285.631,
+      "eval_steps_per_second": 80.546,
       "step": 28000
     },
     {
       "epoch": 22.53098009763425,
-      "eval_loss": 4.089205265045166,
-      "eval_runtime": 3.7809,
-      "eval_samples_per_second": 878.103,
-      "eval_steps_per_second": 55.014,
+      "eval_loss": 5.570806503295898,
+      "eval_runtime": 2.5775,
+      "eval_samples_per_second": 1288.087,
+      "eval_steps_per_second": 80.699,
       "step": 30000
     },
     {
       "epoch": 24.03304543747653,
-      "grad_norm": 1.9181395769119263,
-      "learning_rate": 7.997e-05,
-      "loss": 3.7326,
+      "grad_norm": 5.407034397125244,
+      "learning_rate": 8e-05,
+      "loss": 5.2633,
       "step": 32000
     },
     {
       "epoch": 24.03304543747653,
-      "eval_loss": 4.058904647827148,
-      "eval_runtime": 3.7243,
-      "eval_samples_per_second": 891.451,
-      "eval_steps_per_second": 55.85,
+      "eval_loss": 4.99446439743042,
+      "eval_runtime": 2.5969,
+      "eval_samples_per_second": 1278.448,
+      "eval_steps_per_second": 80.096,
       "step": 32000
     },
     {
       "epoch": 25.535110777318813,
-      "eval_loss": 4.027942657470703,
-      "eval_runtime": 3.7785,
-      "eval_samples_per_second": 878.652,
-      "eval_steps_per_second": 55.048,
+      "eval_loss": 4.523237705230713,
+      "eval_runtime": 2.5888,
+      "eval_samples_per_second": 1282.462,
+      "eval_steps_per_second": 80.347,
       "step": 34000
     },
     {
       "epoch": 27.037176117161096,
-      "grad_norm": 1.8628945350646973,
-      "learning_rate": 8.9965e-05,
-      "loss": 3.6241,
+      "grad_norm": 5.287095069885254,
+      "learning_rate": 9e-05,
+      "loss": 4.3386,
       "step": 36000
     },
     {
       "epoch": 27.037176117161096,
-      "eval_loss": 4.002871513366699,
-      "eval_runtime": 3.8549,
-      "eval_samples_per_second": 861.239,
-      "eval_steps_per_second": 53.957,
+      "eval_loss": 4.140077590942383,
+      "eval_runtime": 2.5851,
+      "eval_samples_per_second": 1284.299,
+      "eval_steps_per_second": 80.462,
       "step": 36000
     },
     {
       "epoch": 28.539241457003378,
-      "eval_loss": 3.990269660949707,
-      "eval_runtime": 3.9421,
-      "eval_samples_per_second": 842.197,
-      "eval_steps_per_second": 52.764,
+      "eval_loss": 3.857847213745117,
+      "eval_runtime": 2.6038,
+      "eval_samples_per_second": 1275.044,
+      "eval_steps_per_second": 79.882,
       "step": 38000
     },
     {
       "epoch": 30.041306796845664,
-      "grad_norm": 1.9048078060150146,
-      "learning_rate": 9.996000000000001e-05,
-      "loss": 3.5294,
+      "grad_norm": 4.30466890335083,
+      "learning_rate": 0.0001,
+      "loss": 3.6823,
       "step": 40000
     },
     {
       "epoch": 30.041306796845664,
-      "eval_loss": 3.9731788635253906,
-      "eval_runtime": 3.8486,
-      "eval_samples_per_second": 862.647,
-      "eval_steps_per_second": 54.045,
+      "eval_loss": 3.6678993701934814,
+      "eval_runtime": 2.7074,
+      "eval_samples_per_second": 1226.275,
+      "eval_steps_per_second": 76.827,
       "step": 40000
     },
     {
       "epoch": 31.543372136687946,
-      "eval_loss": 3.969944715499878,
-      "eval_runtime": 3.8145,
-      "eval_samples_per_second": 870.363,
-      "eval_steps_per_second": 54.529,
+      "eval_loss": 3.5396695137023926,
+      "eval_runtime": 4.9347,
+      "eval_samples_per_second": 672.792,
+      "eval_steps_per_second": 42.151,
       "step": 42000
     },
     {
       "epoch": 33.04543747653023,
-      "grad_norm": 1.8309521675109863,
-      "learning_rate": 9.336333333333334e-05,
-      "loss": 3.4361,
+      "grad_norm": 4.037199974060059,
+      "learning_rate": 9.333333333333334e-05,
+      "loss": 3.3424,
       "step": 44000
     },
     {
       "epoch": 33.04543747653023,
-      "eval_loss": 3.9619035720825195,
-      "eval_runtime": 3.8269,
-      "eval_samples_per_second": 867.55,
-      "eval_steps_per_second": 54.353,
+      "eval_loss": 3.4012534618377686,
+      "eval_runtime": 4.8721,
+      "eval_samples_per_second": 681.437,
+      "eval_steps_per_second": 42.692,
       "step": 44000
     },
     {
       "epoch": 34.547502816372514,
-      "eval_loss": 3.9607603549957275,
-      "eval_runtime": 3.8521,
-      "eval_samples_per_second": 861.876,
-      "eval_steps_per_second": 53.997,
+      "eval_loss": 3.315351963043213,
+      "eval_runtime": 4.9228,
+      "eval_samples_per_second": 674.419,
+      "eval_steps_per_second": 42.253,
       "step": 46000
     },
     {
       "epoch": 36.0495681562148,
-      "grad_norm": 1.8812049627304077,
-      "learning_rate": 8.67e-05,
-      "loss": 3.3449,
+      "grad_norm": 5.304887771606445,
+      "learning_rate": 8.666666666666667e-05,
+      "loss": 3.1266,
       "step": 48000
     },
     {
       "epoch": 36.0495681562148,
-      "eval_loss": 3.9666340351104736,
-      "eval_runtime": 3.8283,
-      "eval_samples_per_second": 867.222,
-      "eval_steps_per_second": 54.332,
+      "eval_loss": 3.2563161849975586,
+      "eval_runtime": 4.958,
+      "eval_samples_per_second": 669.632,
+      "eval_steps_per_second": 41.953,
       "step": 48000
     },
     {
       "epoch": 37.55163349605708,
-      "eval_loss": 3.969193458557129,
-      "eval_runtime": 3.8919,
-      "eval_samples_per_second": 853.06,
-      "eval_steps_per_second": 53.445,
+      "eval_loss": 3.1665167808532715,
+      "eval_runtime": 4.695,
+      "eval_samples_per_second": 707.142,
+      "eval_steps_per_second": 44.303,
       "step": 50000
     },
     {
       "epoch": 39.05369883589936,
-      "grad_norm": 2.010619640350342,
-      "learning_rate": 8.003666666666667e-05,
-      "loss": 3.2664,
+      "grad_norm": 4.668087005615234,
+      "learning_rate": 8e-05,
+      "loss": 2.9693,
       "step": 52000
     },
     {
       "epoch": 39.05369883589936,
-      "eval_loss": 3.9745302200317383,
-      "eval_runtime": 3.8291,
-      "eval_samples_per_second": 867.037,
-      "eval_steps_per_second": 54.32,
+      "eval_loss": 3.1341300010681152,
+      "eval_runtime": 3.8879,
+      "eval_samples_per_second": 853.931,
+      "eval_steps_per_second": 53.499,
       "step": 52000
     },
     {
       "epoch": 40.55576417574164,
-      "eval_loss": 3.986243963241577,
-      "eval_runtime": 3.7036,
-      "eval_samples_per_second": 896.418,
-      "eval_steps_per_second": 56.161,
+      "eval_loss": 3.0564303398132324,
+      "eval_runtime": 4.2699,
+      "eval_samples_per_second": 777.528,
+      "eval_steps_per_second": 48.713,
       "step": 54000
     },
     {
       "epoch": 42.057829515583926,
-      "grad_norm": 2.0288503170013428,
-      "learning_rate": 7.337333333333334e-05,
-      "loss": 3.1963,
+      "grad_norm": 4.504222393035889,
+      "learning_rate": 7.333333333333333e-05,
+      "loss": 2.8544,
       "step": 56000
     },
     {
       "epoch": 42.057829515583926,
-      "eval_loss": 3.9971587657928467,
-      "eval_runtime": 3.7762,
-      "eval_samples_per_second": 879.202,
-      "eval_steps_per_second": 55.083,
+      "eval_loss": 3.0329349040985107,
+      "eval_runtime": 4.8343,
+      "eval_samples_per_second": 686.764,
+      "eval_steps_per_second": 43.026,
       "step": 56000
     },
     {
-      "epoch": 43.575666541494556,
-      "eval_loss": 4.007722854614258,
-      "eval_runtime": 3.9047,
-      "eval_samples_per_second": 850.256,
-      "eval_steps_per_second": 53.269,
+      "epoch": 43.55989485542621,
+      "eval_loss": 2.9552295207977295,
+      "eval_runtime": 4.8628,
+      "eval_samples_per_second": 682.729,
+      "eval_steps_per_second": 42.773,
       "step": 58000
     },
     {
-      "epoch": 45.07773188133684,
-      "grad_norm": 2.1440820693969727,
-      "learning_rate": 6.670666666666668e-05,
-      "loss": 3.1346,
+      "epoch": 45.0619601952685,
+      "grad_norm": 4.24030065536499,
+      "learning_rate": 6.666666666666667e-05,
+      "loss": 2.758,
       "step": 60000
     },
     {
-      "epoch": 45.07773188133684,
-      "eval_loss": 4.020359039306641,
-      "eval_runtime": 3.9205,
-      "eval_samples_per_second": 846.839,
-      "eval_steps_per_second": 53.055,
+      "epoch": 45.0619601952685,
+      "eval_loss": 2.9491591453552246,
+      "eval_runtime": 4.8121,
+      "eval_samples_per_second": 689.932,
+      "eval_steps_per_second": 43.225,
       "step": 60000
     },
     {
-      "epoch": 46.57979722117912,
-      "eval_loss": 4.0261549949646,
-      "eval_runtime": 2.3464,
-      "eval_samples_per_second": 1414.93,
-      "eval_steps_per_second": 88.646,
+      "epoch": 46.56402553511078,
+      "eval_loss": 2.8936784267425537,
+      "eval_runtime": 4.9154,
+      "eval_samples_per_second": 675.43,
+      "eval_steps_per_second": 42.316,
       "step": 62000
     },
     {
-      "epoch": 48.0818625610214,
-      "grad_norm": 2.2309792041778564,
-      "learning_rate": 6.004000000000001e-05,
-      "loss": 3.0792,
+      "epoch": 48.06609087495306,
+      "grad_norm": 4.06797456741333,
+      "learning_rate": 6e-05,
+      "loss": 2.684,
       "step": 64000
     },
     {
-      "epoch": 48.0818625610214,
-      "eval_loss": 4.040497303009033,
-      "eval_runtime": 2.2801,
-      "eval_samples_per_second": 1456.058,
-      "eval_steps_per_second": 91.223,
+      "epoch": 48.06609087495306,
+      "eval_loss": 2.866178512573242,
+      "eval_runtime": 4.8779,
+      "eval_samples_per_second": 680.615,
+      "eval_steps_per_second": 42.641,
       "step": 64000
     },
     {
-      "epoch": 49.583927900863685,
-      "eval_loss": 4.0522260665893555,
-      "eval_runtime": 2.3904,
-      "eval_samples_per_second": 1388.888,
-      "eval_steps_per_second": 87.015,
+      "epoch": 49.568156214795344,
+      "eval_loss": 2.8494906425476074,
+      "eval_runtime": 4.8964,
+      "eval_samples_per_second": 678.05,
+      "eval_steps_per_second": 42.48,
       "step": 66000
     },
     {
-      "epoch": 51.08599324070597,
-      "grad_norm": 2.314417839050293,
-      "learning_rate": 5.337666666666667e-05,
-      "loss": 3.0286,
+      "epoch": 51.07022155463763,
+      "grad_norm": 4.346592903137207,
+      "learning_rate": 5.333333333333333e-05,
+      "loss": 2.6223,
       "step": 68000
     },
     {
-      "epoch": 51.08599324070597,
-      "eval_loss": 4.06997537612915,
-      "eval_runtime": 3.8785,
-      "eval_samples_per_second": 855.993,
-      "eval_steps_per_second": 53.628,
+      "epoch": 51.07022155463763,
+      "eval_loss": 2.825268507003784,
+      "eval_runtime": 4.8304,
+      "eval_samples_per_second": 687.313,
+      "eval_steps_per_second": 43.061,
       "step": 68000
     },
     {
-      "epoch": 52.58805858054826,
-      "eval_loss": 4.0753583908081055,
-      "eval_runtime": 3.8883,
-      "eval_samples_per_second": 853.842,
-      "eval_steps_per_second": 53.494,
+      "epoch": 52.57228689447991,
+      "eval_loss": 2.78464937210083,
+      "eval_runtime": 4.904,
+      "eval_samples_per_second": 676.993,
+      "eval_steps_per_second": 42.414,
       "step": 70000
     },
     {
-      "epoch": 54.09012392039054,
-      "grad_norm": 2.489490509033203,
-      "learning_rate": 4.671333333333334e-05,
-      "loss": 2.9835,
+      "epoch": 54.07435223432219,
+      "grad_norm": 4.131275177001953,
+      "learning_rate": 4.666666666666667e-05,
+      "loss": 2.5704,
       "step": 72000
     },
     {
-      "epoch": 54.09012392039054,
-      "eval_loss": 4.092928886413574,
-      "eval_runtime": 3.9026,
-      "eval_samples_per_second": 850.718,
-      "eval_steps_per_second": 53.298,
+      "epoch": 54.07435223432219,
+      "eval_loss": 2.7910172939300537,
+      "eval_runtime": 4.8745,
+      "eval_samples_per_second": 681.097,
+      "eval_steps_per_second": 42.671,
       "step": 72000
     },
     {
-      "epoch": 55.59218926023282,
-      "eval_loss": 4.100366592407227,
-      "eval_runtime": 3.7565,
-      "eval_samples_per_second": 883.796,
-      "eval_steps_per_second": 55.37,
+      "epoch": 55.57641757416447,
+      "eval_loss": 2.750303030014038,
+      "eval_runtime": 2.5866,
+      "eval_samples_per_second": 1283.529,
+      "eval_steps_per_second": 80.414,
       "step": 74000
     },
     {
-      "epoch": 57.094254600075104,
-      "grad_norm": 2.537451982498169,
-      "learning_rate": 4.0050000000000004e-05,
-      "loss": 2.942,
+      "epoch": 57.078482914006756,
+      "grad_norm": 4.2763261795043945,
+      "learning_rate": 4e-05,
+      "loss": 2.5235,
       "step": 76000
     },
     {
-      "epoch": 57.094254600075104,
-      "eval_loss": 4.114384651184082,
-      "eval_runtime": 3.8675,
-      "eval_samples_per_second": 858.443,
-      "eval_steps_per_second": 53.782,
+      "epoch": 57.078482914006756,
+      "eval_loss": 2.7392032146453857,
+      "eval_runtime": 4.7672,
+      "eval_samples_per_second": 696.429,
+      "eval_steps_per_second": 43.632,
       "step": 76000
     },
     {
-      "epoch": 58.596319939917386,
-      "eval_loss": 4.126460075378418,
-      "eval_runtime": 3.8145,
-      "eval_samples_per_second": 870.359,
-      "eval_steps_per_second": 54.528,
+      "epoch": 58.580548253849045,
+      "eval_loss": 2.722296714782715,
+      "eval_runtime": 2.5739,
+      "eval_samples_per_second": 1289.861,
+      "eval_steps_per_second": 80.811,
       "step": 78000
     },
     {
-      "epoch": 60.09838527975967,
-      "grad_norm": 2.5971362590789795,
-      "learning_rate": 3.338666666666666e-05,
-      "loss": 2.9043,
+      "epoch": 60.08261359369133,
+      "grad_norm": 4.225383281707764,
+      "learning_rate": 3.3333333333333335e-05,
+      "loss": 2.4865,
       "step": 80000
     },
     {
-      "epoch": 60.09838527975967,
-      "eval_loss": 4.132453918457031,
-      "eval_runtime": 3.8363,
-      "eval_samples_per_second": 865.427,
-      "eval_steps_per_second": 54.22,
+      "epoch": 60.08261359369133,
+      "eval_loss": 2.714205503463745,
+      "eval_runtime": 2.5794,
+      "eval_samples_per_second": 1287.128,
+      "eval_steps_per_second": 80.639,
       "step": 80000
     },
     {
-      "epoch": 61.60045061960195,
-      "eval_loss": 4.143109321594238,
-      "eval_runtime": 3.9911,
-      "eval_samples_per_second": 831.849,
-      "eval_steps_per_second": 52.116,
+      "epoch": 61.58467893353361,
+      "eval_loss": 2.7068099975585938,
+      "eval_runtime": 2.5894,
+      "eval_samples_per_second": 1282.152,
+      "eval_steps_per_second": 80.328,
       "step": 82000
     },
     {
-      "epoch": 63.10251595944423,
-      "grad_norm": 2.654517650604248,
-      "learning_rate": 2.6723333333333332e-05,
-      "loss": 2.8693,
+      "epoch": 63.08674427337589,
+      "grad_norm": 4.279160976409912,
+      "learning_rate": 2.6666666666666667e-05,
+      "loss": 2.4549,
       "step": 84000
     },
     {
-      "epoch": 63.10251595944423,
-      "eval_loss": 4.152998447418213,
-      "eval_runtime": 3.976,
-      "eval_samples_per_second": 835.013,
-      "eval_steps_per_second": 52.314,
+      "epoch": 63.08674427337589,
+      "eval_loss": 2.7050325870513916,
+      "eval_runtime": 2.5878,
+      "eval_samples_per_second": 1282.952,
+      "eval_steps_per_second": 80.378,
       "step": 84000
     },
     {
-      "epoch": 64.60458129928652,
-      "eval_loss": 4.161225318908691,
-      "eval_runtime": 3.8263,
-      "eval_samples_per_second": 867.668,
-      "eval_steps_per_second": 54.36,
+      "epoch": 64.58880961321817,
+      "eval_loss": 2.674473524093628,
+      "eval_runtime": 2.5772,
+      "eval_samples_per_second": 1288.232,
+      "eval_steps_per_second": 80.709,
       "step": 86000
     },
     {
-      "epoch": 66.1066466391288,
-      "grad_norm": 2.6866908073425293,
-      "learning_rate": 2.006e-05,
-      "loss": 2.8393,
+      "epoch": 66.09087495306046,
+      "grad_norm": 5.214581489562988,
+      "learning_rate": 2e-05,
+      "loss": 2.427,
       "step": 88000
     },
     {
-      "epoch": 66.1066466391288,
-      "eval_loss": 4.171605110168457,
-      "eval_runtime": 2.3026,
-      "eval_samples_per_second": 1441.854,
-      "eval_steps_per_second": 90.333,
+      "epoch": 66.09087495306046,
+      "eval_loss": 2.6637983322143555,
+      "eval_runtime": 2.6044,
+      "eval_samples_per_second": 1274.757,
+      "eval_steps_per_second": 79.864,
       "step": 88000
     },
     {
-      "epoch": 67.60871197897109,
-      "eval_loss": 4.175863742828369,
-      "eval_runtime": 3.4485,
-      "eval_samples_per_second": 962.748,
-      "eval_steps_per_second": 60.317,
+      "epoch": 67.59294029290274,
+      "eval_loss": 2.659613847732544,
+      "eval_runtime": 4.1498,
+      "eval_samples_per_second": 800.041,
+      "eval_steps_per_second": 50.123,
       "step": 90000
     },
     {
-      "epoch": 69.11077731881336,
-      "grad_norm": 2.797072410583496,
-      "learning_rate": 1.3396666666666668e-05,
-      "loss": 2.8118,
+      "epoch": 69.09500563274503,
+      "grad_norm": 4.723524570465088,
+      "learning_rate": 1.3333333333333333e-05,
+      "loss": 2.4092,
       "step": 92000
     },
     {
-      "epoch": 69.11077731881336,
-      "eval_loss": 4.18150520324707,
-      "eval_runtime": 3.7163,
-      "eval_samples_per_second": 893.351,
-      "eval_steps_per_second": 55.969,
+      "epoch": 69.09500563274503,
+      "eval_loss": 2.647751808166504,
+      "eval_runtime": 4.8495,
+      "eval_samples_per_second": 684.606,
+      "eval_steps_per_second": 42.891,
       "step": 92000
     },
     {
-      "epoch": 70.61284265865565,
-      "eval_loss": 4.187265396118164,
-      "eval_runtime": 3.7036,
-      "eval_samples_per_second": 896.418,
-      "eval_steps_per_second": 56.161,
+      "epoch": 70.5970709725873,
+      "eval_loss": 2.653141736984253,
+      "eval_runtime": 4.9518,
+      "eval_samples_per_second": 670.462,
+      "eval_steps_per_second": 42.005,
       "step": 94000
     },
     {
-      "epoch": 72.11490799849794,
-      "grad_norm": 2.8451874256134033,
-      "learning_rate": 6.733333333333333e-06,
-      "loss": 2.7874,
+      "epoch": 72.0991363124296,
+      "grad_norm": 4.299484729766846,
+      "learning_rate": 6.666666666666667e-06,
+      "loss": 2.3873,
       "step": 96000
     },
     {
-      "epoch": 72.11490799849794,
-      "eval_loss": 4.191434860229492,
-      "eval_runtime": 3.7903,
-      "eval_samples_per_second": 875.931,
-      "eval_steps_per_second": 54.878,
+      "epoch": 72.0991363124296,
+      "eval_loss": 2.631507158279419,
+      "eval_runtime": 4.9961,
+      "eval_samples_per_second": 664.522,
+      "eval_steps_per_second": 41.633,
       "step": 96000
     },
     {
-      "epoch": 73.61697333834022,
-      "eval_loss": 4.193517684936523,
-      "eval_runtime": 3.7187,
-      "eval_samples_per_second": 892.782,
-      "eval_steps_per_second": 55.933,
+      "epoch": 73.60120165227187,
+      "eval_loss": 2.640511989593506,
+      "eval_runtime": 4.7868,
+      "eval_samples_per_second": 693.576,
+      "eval_steps_per_second": 43.453,
       "step": 98000
     },
     {
-      "epoch": 75.1190386781825,
-      "grad_norm": 2.7997725009918213,
-      "learning_rate": 7e-08,
-      "loss": 2.7676,
+      "epoch": 75.10326699211416,
+      "grad_norm": 4.979804992675781,
+      "learning_rate": 0.0,
+      "loss": 2.3734,
       "step": 100000
     },
     {
-      "epoch": 75.1190386781825,
-      "eval_loss": 4.194208145141602,
-      "eval_runtime": 4.0097,
-      "eval_samples_per_second": 827.988,
-      "eval_steps_per_second": 51.874,
+      "epoch": 75.10326699211416,
+      "eval_loss": 2.6056320667266846,
+      "eval_runtime": 4.2982,
+      "eval_samples_per_second": 772.421,
+      "eval_steps_per_second": 48.393,
       "step": 100000
     }
   ],
@@ -601,7 +601,7 @@
       "attributes": {}
     }
   },
-  "total_flos": 2.5851773206339584e+16,
+  "total_flos": 3.11878191808512e+16,
   "train_batch_size": 16,
   "trial_name": null,
   "trial_params": null

checkpoint-100000/training_args.bin

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:4e9ac166ba8168749d2e56e6bf1c4fed81d0a2370c265612a5c96db3fef84453
-size 5496
+oid sha256:85c0376b96bb4265cfb8870438af9d2475993e8b13e1915bca24764e62e88d20
+size 5368