Training in progress, step 4000, checkpoint

checkpoint-4000/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-4000/model.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:915b9094516fe4724e1d656a5c6be21e75899ed1cd7fbf5c73d828ea81c79a9f
-size 51007160
+oid sha256:a4322e58e8c8c11f969f243d0c88b4ca73ebbe8f708a4c75189703c2ba49634c
+size 59702184

checkpoint-4000/optimizer.pt

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:d28d0f378242e29505b7b470d88e095b398800b10749d58c834f4eb96bfe8e2e
-size 102078202
+oid sha256:34f98683e50186919282ff3e1d6aefbdca086bc4d1e0bd8cc86352271c57dbb5
+size 119488058

checkpoint-4000/rng_state.pth

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:a5ab8a8e5484d5e164e1cd9e250999235ce4f2af93325f5f6bf6714b180ae6bf
+oid sha256:c454acd8bdb2c83e244050feac5ad7bbd9e1d724cb18112ec574c4af31302df8
 size 14244

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

@@ -1,6 +1,6 @@
 {
-  "best_metric": 5.811168670654297,
-  "best_model_checkpoint": "/home/p318482/babyLM_controlled/models_trained_last/de_clm/de_childes_42/checkpoint-4000",
+  "best_metric": 6.433775901794434,
+  "best_model_checkpoint": "/home/p318482/babyLM_controlled/models_trained_last/de_mlm/de_childes_42/checkpoint-4000",
   "epoch": 3.0041306796845664,
   "eval_steps": 2000,
   "global_step": 4000,
@@ -10,25 +10,25 @@
   "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
     }
   ],
@@ -49,7 +49,7 @@
       "attributes": {}
     }
   },
-  "total_flos": 1034071110057984.0,
+  "total_flos": 1247512767234048.0,
   "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 sha256:77f63c790664816f6ed91c4986d1e19906c6b86d33dddcf3c8935838a9bb688c
+oid sha256:05c16929805669913e701682558f0da2712e7b14e77c8dab493f55b49b82b680
 size 5368