Training in progress, step 20000

hydra_config_1736986263.118243.yaml

@@ -0,0 +1,45 @@
+experiment:
+  seed: 42
+  name: gpt2_400k-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: 2
+    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:b00fff19b2f2c0d4663f1902e90a8acbc54c4c8dc009d148f2aabf68c36f091c
+oid sha256:0a0cca6ba3dd6e9660dfd197305de6d7a4da211627ba771b1568b864e5679fa2
 size 1745032

training_args.bin

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:c8e141a17943c61188ad5d9d2e45cb3b519a1a3e8513de85ae2672237eb1dc82
-size 5368
+oid sha256:e39e5f79376736909aff47ed0035fa12fc36e51fb24724077ea552997f4276cb
+size 5432