Training in progress, step 20000

hydra_config_1736994567.7338758.yaml

@@ -0,0 +1,45 @@
+experiment:
+  seed: 42
+  name: gpt2_600k-1800000-05
+  group: childes-size-english
+  dry_run: false
+  offline_run: false
+  evaluate_segmentation: true
+  evaluate_babyslm: true
+  blimp_tasks: null
+  resume_checkpoint_path: null
+  resume_run_id: null
+dataset:
+  name: phonemetransformers/CHILDES
+  subconfig: English
+  text_column: phonemized_utterance
+  is_phonemes: true
+  max_age: 120
+  remove_child_utterances: true
+tokenizer:
+  name: phonemetransformers/CHILDES-English-phoneme-tokenizer
+data_preprocessing:
+  max_input_length: 128
+  join_utts: static
+  remove_word_boundaries: true
+  subsample: 1800000
+  subsample_type: tokens
+model:
+  name: gpt2_lm
+  model_kwargs:
+    n_layer: 3
+    n_head: 4
+    n_embd: 128
+    n_positions: 256
+    n_inner: 512
+    resid_pdrop: 0.5
+    embd_pdrop: 0.5
+    attn_pdrop: 0.5
+trainer:
+  batch_size: 32
+  lr: 0.001
+  num_warmup_steps: 60000
+  max_training_steps: 200000
+  logging_steps: 2000
+  save_steps: 20000
+  eval_steps: 20000

model.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:113199b76b1bb44635f1b72d11306f9860c732259136d442bc49defd29216603
+oid sha256:856712af683125c18e590cd1524cbeec07c61d39d88def90eb9aef639b5c5da7
 size 2539288

training_args.bin

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:25e37002c4f6a6781d1013281479faace7ed07d33e8cba5110290704a2a8219f
-size 5368
+oid sha256:602d8e5e97b607fe733fd1514ffd785fde603b0a2cebe60509a470ad5152744e
+size 5432