Training in progress, step 4000, checkpoint

checkpoint-4000/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-4000/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-4000/model.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:23af1571ed15cba6ba24cbfbb2f6bd9f6651465bbf2b6851d73ddaee4d85962c
-size 59702184
+oid sha256:2801ae148db1f89a38af1d83271bf1f654930608ffbdc05c9e66f5d71a959394
+size 51007160

checkpoint-4000/optimizer.pt

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:9d56e4f335898915f22ec83603158db11174cffbc6ac281312c40225423e5fca
-size 119488058
+oid sha256:8b876312d057bcb9f013a41112a442bcccd0a75c2fc2ec6355efb1188d0d5f7b
+size 102078202

checkpoint-4000/rng_state.pth

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:3ca7b686a4cc62b4c3685e8789313658b08950174ca88afc5c651137237eebda
+oid sha256:7b7eff52f03357b9512488137ad64d9a0c3fdb4c2ff008186fc619978748328b
 size 14308

checkpoint-4000/scheduler.pt

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:0ef2be0e7d64cfb402210e997232b889afcd2b3deb5853020826de505e8ab58e
+oid sha256:52ebf870af61d11049051e91e6bc1c00c9fd777dcbd430d66598fe46f09c050b
 size 1000

checkpoint-4000/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-4000/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-4000/trainer_state.json

@@ -1,7 +1,7 @@
 {
-  "best_metric": 5.845461368560791,
-  "best_model_checkpoint": "/home/p318482/babyLM_controlled/models_trained_recent/mlm_unmasking/de_mlm_big/childes_30/checkpoint-4000",
-  "epoch": 3.3003300330033003,
+  "best_metric": 5.826653003692627,
+  "best_model_checkpoint": "/home/p318482/babyLM_controlled/models_trained_last/de_clm/de_childes_30/checkpoint-4000",
+  "epoch": 3.0041306796845664,
   "eval_steps": 2000,
   "global_step": 4000,
   "is_hyper_param_search": false,
@@ -9,44 +9,35 @@
   "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
     }
   ],
   "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,
@@ -58,7 +49,7 @@
       "attributes": {}
     }
   },
-  "total_flos": 1247308058787840.0,
+  "total_flos": 1034071110057984.0,
   "train_batch_size": 16,
   "trial_name": null,
   "trial_params": null

checkpoint-4000/training_args.bin

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