Training in progress, step 20000

config.json

@@ -13,7 +13,7 @@
   "n_embd": 128,
   "n_head": 4,
   "n_inner": 512,
-  "n_layer": 2,
+  "n_layer": 3,
   "n_positions": 256,
   "reorder_and_upcast_attn": false,
   "resid_pdrop": 0.1,

hydra_config_1736991437.8517768.yaml

@@ -0,0 +1,45 @@
+experiment:
+  seed: 42
+  name: gpt2_600k-full-01
+  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: null
+  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: 0.1
+    embd_pdrop: 0.1
+    attn_pdrop: 0.1
+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 sha256:5c0822fb8c072fe683336997e0faa542afb47b034be1124d29f6040048aff8a1
-size 1745032
+oid sha256:b2685dbdce2dcec9f43ee5c00cfd6b5f2a6b154c1b161b0d03fdb5c2f75af795
+size 2539288

training_args.bin

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:d7095aa795c045f1f628bade73a30e10368976550dbb343a2b69d1fa74f4a2d1
-size 5432
+oid sha256:8bf1d0ec1e2a96d0c2cbc50709e295183517fe8466e0ff5b95a9335e38c55841
+size 5368