Training in progress, step 24000, checkpoint

checkpoint-24000/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-24000/model.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:7886392bdaf0ae08aa8ba82a65b8cf271e802301fa43806f10b69d2474b86d46
-size 51007160
+oid sha256:29d08208155aa9556b668dbc389aa5bb1e9e3f9d8e576fa8ec451fe5eda1dc3d
+size 59702184

checkpoint-24000/optimizer.pt

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:cbf5f075b4b0f93d2901f43bed5e31dea110d33758d43212894e817ea2fc8f34
-size 102078202
+oid sha256:f6a91b4002386c0c758b78a772290af3ab890d4f254daabdf1f6ae541c995212
+size 119488058

checkpoint-24000/rng_state.pth

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:0db9531a89f0d4f1bc6c808dd8ce8380e4f07eb65221e2dfd968b54540cbecc0
+oid sha256:33583abbe5c0aa36f645a9169439ff3ddf99e181f54bc304e31b8cb6611113be
 size 14244

checkpoint-24000/scheduler.pt

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:618e55c101bcc3e08431d378f638b238dffcf3e1a0e4ffd18eb2959c5e5d635d
+oid sha256:dc5386d7baeb0b79834cc1e4db992c0d7571cbac82976ce0ffe3a288952a79dd
 size 1000

checkpoint-24000/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-24000/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-24000/trainer_state.json

@@ -1,6 +1,6 @@
 {
-  "best_metric": 4.227701663970947,
-  "best_model_checkpoint": "/home/p318482/babyLM_controlled/models_trained_last/de_clm/de_childes_42/checkpoint-24000",
+  "best_metric": 5.742419719696045,
+  "best_model_checkpoint": "/home/p318482/babyLM_controlled/models_trained_last/de_mlm/de_childes_42/checkpoint-24000",
   "epoch": 18.0247840781074,
   "eval_steps": 2000,
   "global_step": 24000,
@@ -10,140 +10,140 @@
   "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
     }
   ],
@@ -164,7 +164,7 @@
       "attributes": {}
     }
   },
-  "total_flos": 6204426155335680.0,
+  "total_flos": 7485076603404288.0,
   "train_batch_size": 16,
   "trial_name": null,
   "trial_params": null

checkpoint-24000/training_args.bin

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:77f63c790664816f6ed91c4986d1e19906c6b86d33dddcf3c8935838a9bb688c
+oid sha256:05c16929805669913e701682558f0da2712e7b14e77c8dab493f55b49b82b680
 size 5368