Training in progress, step 44000, checkpoint

checkpoint-44000/config.json

@@ -1,28 +1,32 @@
 {
+  "activation_function": "gelu_new",
   "architectures": [
-    "RobertaForMaskedLM"
+    "GPT2LMHeadModel"
   ],
-  "attention_probs_dropout_prob": 0.1,
-  "bos_token_id": 3,
-  "classifier_dropout": null,
-  "eos_token_id": 4,
-  "hidden_act": "gelu",
-  "hidden_dropout_prob": 0.1,
-  "hidden_size": 256,
+  "attn_pdrop": 0.1,
+  "bos_token_id": 0,
+  "embd_pdrop": 0.1,
+  "eos_token_id": 1,
   "initializer_range": 0.02,
-  "intermediate_size": 2048,
-  "layer_norm_eps": 1e-05,
-  "mask_token_id": 6,
-  "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,
+  "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,
   "torch_dtype": "float32",
   "transformers_version": "4.45.2",
-  "type_vocab_size": 2,
   "use_cache": true,
   "vocab_size": 8192
 }

checkpoint-44000/generation_config.json

@@ -0,0 +1,6 @@
+{
+  "_from_model_config": true,
+  "bos_token_id": 0,
+  "eos_token_id": 1,
+  "transformers_version": "4.45.2"
+}

checkpoint-44000/model.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:13f35e77fea1ff52587a98f8ecbf082a87cdd953584774b1cd50031a8ba30b3f
-size 59702184
+oid sha256:94cc04fdb17016a865f7f7ad4ccf2cf208eb60bf2d25dd7e6628078c8a9e6833
+size 51007160

checkpoint-44000/optimizer.pt

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:de4199b62838aad3d6e124f7f76185420d22e14d4698862772d637c0ab1fa0e1
-size 119488058
+oid sha256:956b26c6e2d3cd3b0f357411c64adbdc163210ffcb87a0cf7e324af806235c5d
+size 102078202

checkpoint-44000/rng_state.pth

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:f8d9dac6f957322aa6bd1670471ef7a97ebbdd4a085f5a8538bf13f1a953902c
+oid sha256:65ca4a5a30b60c5e9306ecc289cd56375757093c6146738c194b1adce2633b98
 size 14308

checkpoint-44000/scheduler.pt

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:37eb4227469a4adaea5cd07e4be2f3f8beecf9254a51a6a3c106970fbc79dce5
+oid sha256:220eb48d1eab650aa4cfc612d14cb46a30cb04a376bf78cd0902a0f3d8987146
 size 1000

checkpoint-44000/special_tokens_map.json

@@ -1,6 +1,34 @@
 {
-  "bos_token": "[PAD]",
-  "eos_token": "[EOS]",
-  "mask_token": "[MASK]",
-  "pad_token": "[CLS]"
+  "additional_special_tokens": [
+    "<|endoftext|>"
+  ],
+  "bos_token": {
+    "content": "<s>",
+    "lstrip": false,
+    "normalized": false,
+    "rstrip": false,
+    "single_word": false
+  },
+  "eos_token": {
+    "content": "</s>",
+    "lstrip": false,
+    "normalized": false,
+    "rstrip": false,
+    "single_word": false
+  },
+  "mask_token": {
+    "content": "<mask>",
+    "lstrip": false,
+    "normalized": false,
+    "rstrip": false,
+    "single_word": false
+  },
+  "pad_token": "</s>",
+  "unk_token": {
+    "content": "<unk>",
+    "lstrip": false,
+    "normalized": false,
+    "rstrip": false,
+    "single_word": false
+  }
 }

checkpoint-44000/tokenizer_config.json

@@ -1,7 +1,7 @@
 {
   "added_tokens_decoder": {
     "0": {
-      "content": "[UNK]",
+      "content": "<s>",
       "lstrip": false,
       "normalized": false,
       "rstrip": false,
@@ -9,7 +9,7 @@
       "special": true
     },
     "1": {
-      "content": "[CLS]",
+      "content": "</s>",
       "lstrip": false,
       "normalized": false,
       "rstrip": false,
@@ -17,7 +17,7 @@
       "special": true
     },
     "2": {
-      "content": "[SEP]",
+      "content": "<pad>",
       "lstrip": false,
       "normalized": false,
       "rstrip": false,
@@ -25,7 +25,7 @@
       "special": true
     },
     "3": {
-      "content": "[PAD]",
+      "content": "<unk>",
       "lstrip": false,
       "normalized": false,
       "rstrip": false,
@@ -33,7 +33,7 @@
       "special": true
     },
     "4": {
-      "content": "[EOS]",
+      "content": "<mask>",
       "lstrip": false,
       "normalized": false,
       "rstrip": false,
@@ -41,15 +41,7 @@
       "special": true
     },
     "5": {
-      "content": "[BOS]",
-      "lstrip": false,
-      "normalized": false,
-      "rstrip": false,
-      "single_word": false,
-      "special": true
-    },
-    "6": {
-      "content": "[MASK]",
+      "content": "<|endoftext|>",
       "lstrip": false,
       "normalized": false,
       "rstrip": false,
@@ -57,11 +49,15 @@
       "special": true
     }
   },
-  "bos_token": "[PAD]",
+  "additional_special_tokens": [
+    "<|endoftext|>"
+  ],
+  "bos_token": "<s>",
   "clean_up_tokenization_spaces": false,
-  "eos_token": "[EOS]",
-  "mask_token": "[MASK]",
+  "eos_token": "</s>",
+  "mask_token": "<mask>",
   "model_max_length": 128,
-  "pad_token": "[CLS]",
-  "tokenizer_class": "PreTrainedTokenizerFast"
+  "pad_token": "</s>",
+  "tokenizer_class": "PreTrainedTokenizerFast",
+  "unk_token": "<unk>"
 }

checkpoint-44000/trainer_state.json

@@ -1,7 +1,7 @@
 {
-  "best_metric": 2.9188082218170166,
-  "best_model_checkpoint": "/home/p318482/babyLM_controlled/models_trained_recent/mlm_unmasking/de_mlm_big/childes_30/checkpoint-44000",
-  "epoch": 36.303630363036305,
+  "best_metric": 3.9619035720825195,
+  "best_model_checkpoint": "/home/p318482/babyLM_controlled/models_trained_last/de_clm/de_childes_30/checkpoint-44000",
+  "epoch": 33.04543747653023,
   "eval_steps": 2000,
   "global_step": 44000,
   "is_hyper_param_search": false,
@@ -9,274 +9,265 @@
   "is_world_process_zero": true,
   "log_history": [
     {
-      "epoch": 1.6501650165016502,
-      "eval_loss": 5.888791561126709,
-      "eval_runtime": 1.8805,
-      "eval_samples_per_second": 1301.26,
-      "eval_steps_per_second": 81.362,
+      "epoch": 1.5020653398422832,
+      "eval_loss": 7.09099817276001,
+      "eval_runtime": 3.7115,
+      "eval_samples_per_second": 894.522,
+      "eval_steps_per_second": 56.042,
       "step": 2000
     },
     {
-      "epoch": 3.3003300330033003,
-      "grad_norm": 1.0182065963745117,
-      "learning_rate": 4e-05,
-      "loss": 6.5372,
+      "epoch": 3.0041306796845664,
+      "grad_norm": 1.3897417783737183,
+      "learning_rate": 1e-05,
+      "loss": 6.9957,
       "step": 4000
     },
     {
-      "epoch": 3.3003300330033003,
-      "eval_loss": 5.845461368560791,
-      "eval_runtime": 1.8476,
-      "eval_samples_per_second": 1324.395,
-      "eval_steps_per_second": 82.808,
+      "epoch": 3.0041306796845664,
+      "eval_loss": 5.826653003692627,
+      "eval_runtime": 3.7561,
+      "eval_samples_per_second": 883.906,
+      "eval_steps_per_second": 55.377,
       "step": 4000
     },
     {
-      "epoch": 4.9504950495049505,
-      "eval_loss": 5.773000717163086,
-      "eval_runtime": 1.8524,
-      "eval_samples_per_second": 1320.962,
-      "eval_steps_per_second": 82.594,
+      "epoch": 4.50619601952685,
+      "eval_loss": 5.465098857879639,
+      "eval_runtime": 3.8583,
+      "eval_samples_per_second": 860.484,
+      "eval_steps_per_second": 53.91,
       "step": 6000
     },
     {
-      "epoch": 6.600660066006601,
-      "grad_norm": 1.9582704305648804,
-      "learning_rate": 8e-05,
-      "loss": 5.7792,
+      "epoch": 6.008261359369133,
+      "grad_norm": 2.3630881309509277,
+      "learning_rate": 2e-05,
+      "loss": 5.213,
       "step": 8000
     },
     {
-      "epoch": 6.600660066006601,
-      "eval_loss": 5.6879730224609375,
-      "eval_runtime": 1.8512,
-      "eval_samples_per_second": 1321.837,
-      "eval_steps_per_second": 82.649,
+      "epoch": 6.008261359369133,
+      "eval_loss": 5.171680927276611,
+      "eval_runtime": 3.928,
+      "eval_samples_per_second": 845.218,
+      "eval_steps_per_second": 52.953,
       "step": 8000
     },
     {
-      "epoch": 8.250825082508252,
-      "eval_loss": 5.123713970184326,
-      "eval_runtime": 1.8558,
-      "eval_samples_per_second": 1318.591,
-      "eval_steps_per_second": 82.446,
+      "epoch": 7.510326699211416,
+      "eval_loss": 4.959029197692871,
+      "eval_runtime": 3.8954,
+      "eval_samples_per_second": 852.288,
+      "eval_steps_per_second": 53.396,
       "step": 10000
     },
     {
-      "epoch": 9.900990099009901,
-      "grad_norm": 2.0609753131866455,
-      "learning_rate": 0.00012,
-      "loss": 5.147,
+      "epoch": 9.0123920390537,
+      "grad_norm": 2.578948736190796,
+      "learning_rate": 2.9995e-05,
+      "loss": 4.7279,
       "step": 12000
     },
     {
-      "epoch": 9.900990099009901,
-      "eval_loss": 4.784373760223389,
-      "eval_runtime": 1.8697,
-      "eval_samples_per_second": 1308.749,
-      "eval_steps_per_second": 81.83,
+      "epoch": 9.0123920390537,
+      "eval_loss": 4.791839599609375,
+      "eval_runtime": 3.7027,
+      "eval_samples_per_second": 896.638,
+      "eval_steps_per_second": 56.175,
       "step": 12000
     },
     {
-      "epoch": 11.551155115511552,
-      "eval_loss": 4.513790607452393,
-      "eval_runtime": 1.8742,
-      "eval_samples_per_second": 1305.614,
-      "eval_steps_per_second": 81.634,
+      "epoch": 10.514457378895981,
+      "eval_loss": 4.660107612609863,
+      "eval_runtime": 3.7788,
+      "eval_samples_per_second": 878.594,
+      "eval_steps_per_second": 55.044,
       "step": 14000
     },
     {
-      "epoch": 13.201320132013201,
-      "grad_norm": 2.4001076221466064,
-      "learning_rate": 0.00016,
-      "loss": 4.4762,
+      "epoch": 12.016522718738265,
+      "grad_norm": 2.3461837768554688,
+      "learning_rate": 3.999e-05,
+      "loss": 4.4214,
       "step": 16000
     },
     {
-      "epoch": 13.201320132013201,
-      "eval_loss": 4.007374286651611,
-      "eval_runtime": 1.8539,
-      "eval_samples_per_second": 1319.908,
-      "eval_steps_per_second": 82.528,
+      "epoch": 12.016522718738265,
+      "eval_loss": 4.552372932434082,
+      "eval_runtime": 3.7928,
+      "eval_samples_per_second": 875.332,
+      "eval_steps_per_second": 54.84,
       "step": 16000
     },
     {
-      "epoch": 14.851485148514852,
-      "eval_loss": 3.8194680213928223,
-      "eval_runtime": 1.8778,
-      "eval_samples_per_second": 1303.112,
-      "eval_steps_per_second": 81.478,
+      "epoch": 13.518588058580548,
+      "eval_loss": 4.458698272705078,
+      "eval_runtime": 3.8109,
+      "eval_samples_per_second": 871.192,
+      "eval_steps_per_second": 54.581,
       "step": 18000
     },
     {
-      "epoch": 16.501650165016503,
-      "grad_norm": 2.3759870529174805,
-      "learning_rate": 0.0002,
-      "loss": 3.8668,
+      "epoch": 15.020653398422832,
+      "grad_norm": 2.1618399620056152,
+      "learning_rate": 4.9985e-05,
+      "loss": 4.195,
       "step": 20000
     },
     {
-      "epoch": 16.501650165016503,
-      "eval_loss": 3.6909382343292236,
-      "eval_runtime": 1.8525,
-      "eval_samples_per_second": 1320.929,
-      "eval_steps_per_second": 82.592,
+      "epoch": 15.020653398422832,
+      "eval_loss": 4.3683576583862305,
+      "eval_runtime": 3.8444,
+      "eval_samples_per_second": 863.584,
+      "eval_steps_per_second": 54.104,
       "step": 20000
     },
     {
-      "epoch": 18.151815181518153,
-      "eval_loss": 3.5956225395202637,
-      "eval_runtime": 1.8688,
-      "eval_samples_per_second": 1309.397,
-      "eval_steps_per_second": 81.871,
+      "epoch": 16.522718738265116,
+      "eval_loss": 4.289999485015869,
+      "eval_runtime": 3.8372,
+      "eval_samples_per_second": 865.217,
+      "eval_steps_per_second": 54.206,
       "step": 22000
     },
     {
-      "epoch": 19.801980198019802,
-      "grad_norm": 2.246126174926758,
-      "learning_rate": 0.00024,
-      "loss": 3.6038,
+      "epoch": 18.0247840781074,
+      "grad_norm": 2.0583324432373047,
+      "learning_rate": 5.9980000000000005e-05,
+      "loss": 4.0109,
       "step": 24000
     },
     {
-      "epoch": 19.801980198019802,
-      "eval_loss": 3.4480886459350586,
-      "eval_runtime": 1.8595,
-      "eval_samples_per_second": 1315.949,
-      "eval_steps_per_second": 82.28,
+      "epoch": 18.0247840781074,
+      "eval_loss": 4.227661609649658,
+      "eval_runtime": 3.7524,
+      "eval_samples_per_second": 884.777,
+      "eval_steps_per_second": 55.432,
       "step": 24000
     },
     {
-      "epoch": 21.45214521452145,
-      "eval_loss": 3.37424898147583,
-      "eval_runtime": 1.8589,
-      "eval_samples_per_second": 1316.347,
-      "eval_steps_per_second": 82.305,
+      "epoch": 19.52684941794968,
+      "eval_loss": 4.172571659088135,
+      "eval_runtime": 3.7915,
+      "eval_samples_per_second": 875.64,
+      "eval_steps_per_second": 54.859,
       "step": 26000
     },
     {
-      "epoch": 23.102310231023104,
-      "grad_norm": 2.1454977989196777,
-      "learning_rate": 0.00028000000000000003,
-      "loss": 3.3867,
+      "epoch": 21.028914757791963,
+      "grad_norm": 2.0024306774139404,
+      "learning_rate": 6.997500000000001e-05,
+      "loss": 3.8596,
       "step": 28000
     },
     {
-      "epoch": 23.102310231023104,
-      "eval_loss": 3.263887882232666,
-      "eval_runtime": 1.8666,
-      "eval_samples_per_second": 1310.945,
-      "eval_steps_per_second": 81.968,
+      "epoch": 21.028914757791963,
+      "eval_loss": 4.128843307495117,
+      "eval_runtime": 3.6349,
+      "eval_samples_per_second": 913.374,
+      "eval_steps_per_second": 57.223,
       "step": 28000
     },
     {
-      "epoch": 24.752475247524753,
-      "eval_loss": 3.1954398155212402,
-      "eval_runtime": 1.8583,
-      "eval_samples_per_second": 1316.823,
-      "eval_steps_per_second": 82.335,
+      "epoch": 22.53098009763425,
+      "eval_loss": 4.089205265045166,
+      "eval_runtime": 3.7809,
+      "eval_samples_per_second": 878.103,
+      "eval_steps_per_second": 55.014,
       "step": 30000
     },
     {
-      "epoch": 26.402640264026402,
-      "grad_norm": 2.3603739738464355,
-      "learning_rate": 0.00032,
-      "loss": 3.2158,
+      "epoch": 24.03304543747653,
+      "grad_norm": 1.9181395769119263,
+      "learning_rate": 7.997e-05,
+      "loss": 3.7326,
       "step": 32000
     },
     {
-      "epoch": 26.402640264026402,
-      "eval_loss": 3.1180179119110107,
-      "eval_runtime": 1.8529,
-      "eval_samples_per_second": 1320.612,
-      "eval_steps_per_second": 82.572,
+      "epoch": 24.03304543747653,
+      "eval_loss": 4.058904647827148,
+      "eval_runtime": 3.7243,
+      "eval_samples_per_second": 891.451,
+      "eval_steps_per_second": 55.85,
       "step": 32000
     },
     {
-      "epoch": 28.05280528052805,
-      "eval_loss": 3.090503454208374,
-      "eval_runtime": 1.8647,
-      "eval_samples_per_second": 1312.256,
-      "eval_steps_per_second": 82.05,
+      "epoch": 25.535110777318813,
+      "eval_loss": 4.027942657470703,
+      "eval_runtime": 3.7785,
+      "eval_samples_per_second": 878.652,
+      "eval_steps_per_second": 55.048,
       "step": 34000
     },
     {
-      "epoch": 29.702970297029704,
-      "grad_norm": 2.3500428199768066,
-      "learning_rate": 0.00035999999999999997,
-      "loss": 3.0917,
+      "epoch": 27.037176117161096,
+      "grad_norm": 1.8628945350646973,
+      "learning_rate": 8.9965e-05,
+      "loss": 3.6241,
       "step": 36000
     },
     {
-      "epoch": 29.702970297029704,
-      "eval_loss": 3.0517561435699463,
-      "eval_runtime": 1.8516,
-      "eval_samples_per_second": 1321.58,
-      "eval_steps_per_second": 82.633,
+      "epoch": 27.037176117161096,
+      "eval_loss": 4.002871513366699,
+      "eval_runtime": 3.8549,
+      "eval_samples_per_second": 861.239,
+      "eval_steps_per_second": 53.957,
       "step": 36000
     },
     {
-      "epoch": 31.353135313531354,
-      "eval_loss": 3.031769275665283,
-      "eval_runtime": 1.855,
-      "eval_samples_per_second": 1319.169,
-      "eval_steps_per_second": 82.482,
+      "epoch": 28.539241457003378,
+      "eval_loss": 3.990269660949707,
+      "eval_runtime": 3.9421,
+      "eval_samples_per_second": 842.197,
+      "eval_steps_per_second": 52.764,
       "step": 38000
     },
     {
-      "epoch": 33.00330033003301,
-      "grad_norm": 1.963541030883789,
-      "learning_rate": 0.0004,
-      "loss": 3.0074,
+      "epoch": 30.041306796845664,
+      "grad_norm": 1.9048078060150146,
+      "learning_rate": 9.996000000000001e-05,
+      "loss": 3.5294,
       "step": 40000
     },
     {
-      "epoch": 33.00330033003301,
-      "eval_loss": 2.988098382949829,
-      "eval_runtime": 1.8581,
-      "eval_samples_per_second": 1316.953,
-      "eval_steps_per_second": 82.343,
+      "epoch": 30.041306796845664,
+      "eval_loss": 3.9731788635253906,
+      "eval_runtime": 3.8486,
+      "eval_samples_per_second": 862.647,
+      "eval_steps_per_second": 54.045,
       "step": 40000
     },
     {
-      "epoch": 34.65346534653465,
-      "eval_loss": 2.9400546550750732,
-      "eval_runtime": 1.8573,
-      "eval_samples_per_second": 1317.525,
-      "eval_steps_per_second": 82.379,
+      "epoch": 31.543372136687946,
+      "eval_loss": 3.969944715499878,
+      "eval_runtime": 3.8145,
+      "eval_samples_per_second": 870.363,
+      "eval_steps_per_second": 54.529,
       "step": 42000
     },
     {
-      "epoch": 36.303630363036305,
-      "grad_norm": 1.974672555923462,
-      "learning_rate": 0.00044,
-      "loss": 2.9357,
+      "epoch": 33.04543747653023,
+      "grad_norm": 1.8309521675109863,
+      "learning_rate": 9.336333333333334e-05,
+      "loss": 3.4361,
       "step": 44000
     },
     {
-      "epoch": 36.303630363036305,
-      "eval_loss": 2.9188082218170166,
-      "eval_runtime": 1.877,
-      "eval_samples_per_second": 1303.662,
-      "eval_steps_per_second": 81.512,
+      "epoch": 33.04543747653023,
+      "eval_loss": 3.9619035720825195,
+      "eval_runtime": 3.8269,
+      "eval_samples_per_second": 867.55,
+      "eval_steps_per_second": 54.353,
       "step": 44000
     }
   ],
   "logging_steps": 4000,
-  "max_steps": 400000,
+  "max_steps": 100000,
   "num_input_tokens_seen": 0,
-  "num_train_epochs": 331,
+  "num_train_epochs": 76,
   "save_steps": 4000,
   "stateful_callbacks": {
-    "EarlyStoppingCallback": {
-      "args": {
-        "early_stopping_patience": 3,
-        "early_stopping_threshold": 0.001
-      },
-      "attributes": {
-        "early_stopping_patience_counter": 0
-      }
-    },
     "TrainerControl": {
       "args": {
         "should_epoch_stop": false,
@@ -288,7 +279,7 @@
       "attributes": {}
     }
   },
-  "total_flos": 1.371994998571008e+16,
+  "total_flos": 1.13747817056256e+16,
   "train_batch_size": 16,
   "trial_name": null,
   "trial_params": null

checkpoint-44000/training_args.bin

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:82b00b33ff5a34cb27834747d142985f370938cb613e5baed1026ce1070e9a42
+oid sha256:2a77d647bb2e817d836b30fa0c2702de6364acb70c98f3c308390594352b8543
 size 5368