Training in progress, step 44000, checkpoint

checkpoint-44000/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-44000/model.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:a4954eb5e958316f0a26322389d76b01d49a20f02080202418ce763477464fc1
-size 51007160
+oid sha256:bda27e98a63c647d8be540fcab44fa7b6ea512951369166207a31b25074c3204
+size 59702184

checkpoint-44000/optimizer.pt

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:dc36bd5c94e7764132921eefbc5c1d40ae03c81cf1e66007ec1f62883dc944a5
-size 102078202
+oid sha256:56ebbde5be847c3f5f41a27f08c86fb341800d10b67b8627183cc2b8a929c7b4
+size 119488058

checkpoint-44000/rng_state.pth

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:d7888343c0cb40489e247b37a65446402d8768ecda76516f9a461543c999ed29
+oid sha256:39dfee9bc71be2c3b4a47af39909ea9deaa5da4640ea0bd6b5457aaa7d71adb9
 size 14244

checkpoint-44000/scheduler.pt

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:220eb48d1eab650aa4cfc612d14cb46a30cb04a376bf78cd0902a0f3d8987146
+oid sha256:31646587383179df2de5e87c2e8bc3e2a76c4621dc77ed7445d443349476151f
 size 1000

checkpoint-44000/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-44000/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-44000/trainer_state.json

@@ -1,7 +1,7 @@
 {
-  "best_metric": 3.9599227905273438,
-  "best_model_checkpoint": "/home/p318482/babyLM_controlled/models_trained_last/de_clm/de_childes_42/checkpoint-44000",
-  "epoch": 33.04543747653023,
+  "best_metric": 3.3299686908721924,
+  "best_model_checkpoint": "/home/p318482/babyLM_controlled/models_trained_last/de_mlm/de_childes_42/checkpoint-44000",
+  "epoch": 33.055576417574166,
   "eval_steps": 2000,
   "global_step": 44000,
   "is_hyper_param_search": false,
@@ -10,255 +10,255 @@
   "log_history": [
     {
       "epoch": 1.5020653398422832,
-      "eval_loss": 7.03363037109375,
-      "eval_runtime": 3.6768,
-      "eval_samples_per_second": 902.97,
-      "eval_steps_per_second": 56.572,
+      "eval_loss": 7.505205154418945,
+      "eval_runtime": 4.9012,
+      "eval_samples_per_second": 677.383,
+      "eval_steps_per_second": 42.438,
       "step": 2000
     },
     {
       "epoch": 3.0041306796845664,
-      "grad_norm": 1.6605541706085205,
+      "grad_norm": 1.3569484949111938,
       "learning_rate": 1e-05,
-      "loss": 6.9417,
+      "loss": 7.4063,
       "step": 4000
     },
     {
       "epoch": 3.0041306796845664,
-      "eval_loss": 5.811168670654297,
-      "eval_runtime": 3.7246,
-      "eval_samples_per_second": 891.378,
-      "eval_steps_per_second": 55.845,
+      "eval_loss": 6.433775901794434,
+      "eval_runtime": 4.7916,
+      "eval_samples_per_second": 692.873,
+      "eval_steps_per_second": 43.409,
       "step": 4000
     },
     {
       "epoch": 4.50619601952685,
-      "eval_loss": 5.458603382110596,
-      "eval_runtime": 3.8082,
-      "eval_samples_per_second": 871.804,
-      "eval_steps_per_second": 54.619,
+      "eval_loss": 6.305135250091553,
+      "eval_runtime": 4.8722,
+      "eval_samples_per_second": 681.421,
+      "eval_steps_per_second": 42.691,
       "step": 6000
     },
     {
       "epoch": 6.008261359369133,
-      "grad_norm": 2.27949595451355,
+      "grad_norm": 1.548327088356018,
       "learning_rate": 2e-05,
-      "loss": 5.2077,
+      "loss": 6.0672,
       "step": 8000
     },
     {
       "epoch": 6.008261359369133,
-      "eval_loss": 5.171741008758545,
-      "eval_runtime": 3.8003,
-      "eval_samples_per_second": 873.606,
-      "eval_steps_per_second": 54.732,
+      "eval_loss": 6.1662516593933105,
+      "eval_runtime": 4.8572,
+      "eval_samples_per_second": 683.527,
+      "eval_steps_per_second": 42.823,
       "step": 8000
     },
     {
       "epoch": 7.510326699211416,
-      "eval_loss": 4.9582695960998535,
-      "eval_runtime": 3.7933,
-      "eval_samples_per_second": 875.227,
-      "eval_steps_per_second": 54.834,
+      "eval_loss": 6.095215797424316,
+      "eval_runtime": 4.7932,
+      "eval_samples_per_second": 692.655,
+      "eval_steps_per_second": 43.395,
       "step": 10000
     },
     {
       "epoch": 9.0123920390537,
-      "grad_norm": 2.439126491546631,
-      "learning_rate": 2.9995e-05,
-      "loss": 4.7315,
+      "grad_norm": 1.639375925064087,
+      "learning_rate": 3e-05,
+      "loss": 5.8691,
       "step": 12000
     },
     {
       "epoch": 9.0123920390537,
-      "eval_loss": 4.800222873687744,
-      "eval_runtime": 3.705,
-      "eval_samples_per_second": 896.095,
-      "eval_steps_per_second": 56.141,
+      "eval_loss": 6.0121965408325195,
+      "eval_runtime": 4.8217,
+      "eval_samples_per_second": 688.56,
+      "eval_steps_per_second": 43.139,
       "step": 12000
     },
     {
       "epoch": 10.514457378895981,
-      "eval_loss": 4.670550346374512,
-      "eval_runtime": 3.7395,
-      "eval_samples_per_second": 887.823,
-      "eval_steps_per_second": 55.623,
+      "eval_loss": 5.951181888580322,
+      "eval_runtime": 4.7668,
+      "eval_samples_per_second": 696.478,
+      "eval_steps_per_second": 43.635,
       "step": 14000
     },
     {
       "epoch": 12.016522718738265,
-      "grad_norm": 2.3431427478790283,
-      "learning_rate": 3.999e-05,
-      "loss": 4.4253,
+      "grad_norm": 1.6758147478103638,
+      "learning_rate": 4e-05,
+      "loss": 5.7209,
       "step": 16000
     },
     {
       "epoch": 12.016522718738265,
-      "eval_loss": 4.555232524871826,
-      "eval_runtime": 3.7089,
-      "eval_samples_per_second": 895.14,
-      "eval_steps_per_second": 56.081,
+      "eval_loss": 5.869147300720215,
+      "eval_runtime": 4.6953,
+      "eval_samples_per_second": 707.092,
+      "eval_steps_per_second": 44.3,
       "step": 16000
     },
     {
       "epoch": 13.518588058580548,
-      "eval_loss": 4.4522013664245605,
-      "eval_runtime": 3.4306,
-      "eval_samples_per_second": 967.764,
-      "eval_steps_per_second": 60.631,
+      "eval_loss": 5.852941513061523,
+      "eval_runtime": 4.8028,
+      "eval_samples_per_second": 691.267,
+      "eval_steps_per_second": 43.308,
       "step": 18000
     },
     {
       "epoch": 15.020653398422832,
-      "grad_norm": 2.3183419704437256,
-      "learning_rate": 4.9985e-05,
-      "loss": 4.1962,
+      "grad_norm": 2.3887548446655273,
+      "learning_rate": 5e-05,
+      "loss": 5.6105,
       "step": 20000
     },
     {
       "epoch": 15.020653398422832,
-      "eval_loss": 4.373892784118652,
-      "eval_runtime": 3.1907,
-      "eval_samples_per_second": 1040.508,
-      "eval_steps_per_second": 65.188,
+      "eval_loss": 5.796474933624268,
+      "eval_runtime": 4.7859,
+      "eval_samples_per_second": 693.704,
+      "eval_steps_per_second": 43.461,
       "step": 20000
     },
     {
       "epoch": 16.522718738265116,
-      "eval_loss": 4.292479515075684,
-      "eval_runtime": 3.6404,
-      "eval_samples_per_second": 911.984,
-      "eval_steps_per_second": 57.136,
+      "eval_loss": 5.740424156188965,
+      "eval_runtime": 4.6987,
+      "eval_samples_per_second": 706.584,
+      "eval_steps_per_second": 44.268,
       "step": 22000
     },
     {
       "epoch": 18.0247840781074,
-      "grad_norm": 2.1154417991638184,
-      "learning_rate": 5.9980000000000005e-05,
-      "loss": 4.0105,
+      "grad_norm": 4.52785587310791,
+      "learning_rate": 6e-05,
+      "loss": 5.5302,
       "step": 24000
     },
     {
       "epoch": 18.0247840781074,
-      "eval_loss": 4.227701663970947,
-      "eval_runtime": 3.6867,
-      "eval_samples_per_second": 900.545,
-      "eval_steps_per_second": 56.42,
+      "eval_loss": 5.742419719696045,
+      "eval_runtime": 3.8111,
+      "eval_samples_per_second": 871.137,
+      "eval_steps_per_second": 54.577,
       "step": 24000
     },
     {
       "epoch": 19.52684941794968,
-      "eval_loss": 4.170625686645508,
-      "eval_runtime": 3.5669,
-      "eval_samples_per_second": 930.781,
-      "eval_steps_per_second": 58.314,
+      "eval_loss": 5.7255754470825195,
+      "eval_runtime": 2.6268,
+      "eval_samples_per_second": 1263.889,
+      "eval_steps_per_second": 79.183,
       "step": 26000
     },
     {
       "epoch": 21.028914757791963,
-      "grad_norm": 2.051388740539551,
-      "learning_rate": 6.997500000000001e-05,
-      "loss": 3.857,
+      "grad_norm": 4.69941520690918,
+      "learning_rate": 7e-05,
+      "loss": 5.4587,
       "step": 28000
     },
     {
       "epoch": 21.028914757791963,
-      "eval_loss": 4.124534606933594,
-      "eval_runtime": 3.7721,
-      "eval_samples_per_second": 880.155,
-      "eval_steps_per_second": 55.142,
+      "eval_loss": 5.6831231117248535,
+      "eval_runtime": 2.5816,
+      "eval_samples_per_second": 1286.034,
+      "eval_steps_per_second": 80.571,
       "step": 28000
     },
     {
       "epoch": 22.53098009763425,
-      "eval_loss": 4.084589004516602,
-      "eval_runtime": 3.7783,
-      "eval_samples_per_second": 878.7,
-      "eval_steps_per_second": 55.051,
+      "eval_loss": 5.396602153778076,
+      "eval_runtime": 2.6907,
+      "eval_samples_per_second": 1233.859,
+      "eval_steps_per_second": 77.302,
       "step": 30000
     },
     {
       "epoch": 24.03304543747653,
-      "grad_norm": 1.966176986694336,
-      "learning_rate": 7.997e-05,
-      "loss": 3.7298,
+      "grad_norm": 5.43873929977417,
+      "learning_rate": 8e-05,
+      "loss": 5.0899,
       "step": 32000
     },
     {
       "epoch": 24.03304543747653,
-      "eval_loss": 4.058874130249023,
-      "eval_runtime": 3.6613,
-      "eval_samples_per_second": 906.785,
-      "eval_steps_per_second": 56.811,
+      "eval_loss": 4.704195022583008,
+      "eval_runtime": 2.6893,
+      "eval_samples_per_second": 1234.521,
+      "eval_steps_per_second": 77.343,
       "step": 32000
     },
     {
       "epoch": 25.535110777318813,
-      "eval_loss": 4.026635646820068,
-      "eval_runtime": 3.722,
-      "eval_samples_per_second": 891.982,
-      "eval_steps_per_second": 55.883,
+      "eval_loss": 4.2316813468933105,
+      "eval_runtime": 4.2325,
+      "eval_samples_per_second": 784.414,
+      "eval_steps_per_second": 49.144,
       "step": 34000
     },
     {
       "epoch": 27.037176117161096,
-      "grad_norm": 1.8346107006072998,
-      "learning_rate": 8.9965e-05,
-      "loss": 3.6218,
+      "grad_norm": 4.492648124694824,
+      "learning_rate": 9e-05,
+      "loss": 4.0988,
       "step": 36000
     },
     {
       "epoch": 27.037176117161096,
-      "eval_loss": 4.0028157234191895,
-      "eval_runtime": 3.7484,
-      "eval_samples_per_second": 885.713,
-      "eval_steps_per_second": 55.49,
+      "eval_loss": 3.90928316116333,
+      "eval_runtime": 4.8525,
+      "eval_samples_per_second": 684.177,
+      "eval_steps_per_second": 42.864,
       "step": 36000
     },
     {
-      "epoch": 28.539241457003378,
-      "eval_loss": 3.986341953277588,
-      "eval_runtime": 3.7709,
-      "eval_samples_per_second": 880.423,
-      "eval_steps_per_second": 55.159,
+      "epoch": 28.549380398047315,
+      "eval_loss": 3.749643564224243,
+      "eval_runtime": 4.8877,
+      "eval_samples_per_second": 679.256,
+      "eval_steps_per_second": 42.556,
       "step": 38000
     },
     {
-      "epoch": 30.041306796845664,
-      "grad_norm": 1.8153942823410034,
-      "learning_rate": 9.996000000000001e-05,
-      "loss": 3.5278,
+      "epoch": 30.051445737889598,
+      "grad_norm": 4.555140018463135,
+      "learning_rate": 0.0001,
+      "loss": 3.555,
       "step": 40000
     },
     {
-      "epoch": 30.041306796845664,
-      "eval_loss": 3.9731833934783936,
-      "eval_runtime": 3.8054,
-      "eval_samples_per_second": 872.434,
-      "eval_steps_per_second": 54.659,
+      "epoch": 30.051445737889598,
+      "eval_loss": 3.5960817337036133,
+      "eval_runtime": 4.777,
+      "eval_samples_per_second": 695.0,
+      "eval_steps_per_second": 43.542,
       "step": 40000
     },
     {
-      "epoch": 31.543372136687946,
-      "eval_loss": 3.971970796585083,
-      "eval_runtime": 3.6892,
-      "eval_samples_per_second": 899.912,
-      "eval_steps_per_second": 56.38,
+      "epoch": 31.55351107773188,
+      "eval_loss": 3.454169988632202,
+      "eval_runtime": 4.8006,
+      "eval_samples_per_second": 691.578,
+      "eval_steps_per_second": 43.328,
       "step": 42000
     },
     {
-      "epoch": 33.04543747653023,
-      "grad_norm": 1.8285443782806396,
-      "learning_rate": 9.336333333333334e-05,
-      "loss": 3.4351,
+      "epoch": 33.055576417574166,
+      "grad_norm": 4.173329830169678,
+      "learning_rate": 9.333333333333334e-05,
+      "loss": 3.2522,
       "step": 44000
     },
     {
-      "epoch": 33.04543747653023,
-      "eval_loss": 3.9599227905273438,
-      "eval_runtime": 3.7507,
-      "eval_samples_per_second": 885.169,
-      "eval_steps_per_second": 55.456,
+      "epoch": 33.055576417574166,
+      "eval_loss": 3.3299686908721924,
+      "eval_runtime": 4.544,
+      "eval_samples_per_second": 730.638,
+      "eval_steps_per_second": 45.775,
       "step": 44000
     }
   ],
@@ -279,7 +279,7 @@
       "attributes": {}
     }
   },
-  "total_flos": 1.13747817056256e+16,
+  "total_flos": 1.3722640439574528e+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 sha256:77f63c790664816f6ed91c4986d1e19906c6b86d33dddcf3c8935838a9bb688c
-size 5368
+oid sha256:977cda690c0a8825ee003e5fad5dfe7020c0f0650747f283a88b45076025d0cf
+size 5432