Training in progress, step 50000

blimp_results.json

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:b7ab61a9f5abc94daaace067debc70a12e7ab5b883f04303561040d26ae2b618
-size 80217640
+oid sha256:760662f5d2930739acf8f1de7e408cc4274b742cf2719b24a8d08d8bf65f7d25
+size 80215269

config.json

@@ -13,7 +13,7 @@
   "n_embd": 512,
   "n_head": 8,
   "n_inner": 2048,
-  "n_layer": 6,
+  "n_layer": 8,
   "n_positions": 256,
   "reorder_and_upcast_attn": false,
   "resid_pdrop": 0.1,

hydra_config_1724868027.8639908.yaml

@@ -0,0 +1,41 @@
+experiment:
+  seed: 42
+  name: gpt2_25M-bpe-text
+  group: babylm-small
+  dry_run: false
+  offline_run: false
+  evaluate_segmentation: false
+  evaluate_babyslm: true
+  blimp_tasks: blimp_filtered,blimp_supplement
+  resume_checkpoint_path: null
+  resume_run_id: null
+dataset:
+  name: transformersegmentation/BabyLM-phonemized
+  subconfig: strict_small
+  text_column: text
+  is_phonemes: false
+  max_age: null
+tokenizer:
+  name: transformersegmentation/BabyLM-BPE-ortho-tokenizer
+data_preprocessing:
+  max_input_length: 128
+  join_utts: static
+  remove_word_boundaries: false
+  subsample: null
+  subsample_type: examples
+model:
+  name: gpt2_lm
+  model_kwargs:
+    n_layer: 8
+    n_head: 8
+    n_embd: 512
+    n_positions: 256
+    n_inner: 2048
+trainer:
+  batch_size: 32
+  lr: 0.001
+  num_warmup_steps: 90000
+  max_training_steps: 400000
+  logging_steps: 4000
+  save_steps: 50000
+  eval_steps: 50000

hydra_config_1724870633.4389598.yaml

@@ -0,0 +1,41 @@
+experiment:
+  seed: 42
+  name: gpt2_25M-bpe-text
+  group: babylm-small
+  dry_run: false
+  offline_run: false
+  evaluate_segmentation: false
+  evaluate_babyslm: true
+  blimp_tasks: blimp_filtered,blimp_supplement
+  resume_checkpoint_path: null
+  resume_run_id: null
+dataset:
+  name: transformersegmentation/BabyLM-phonemized
+  subconfig: strict_small
+  text_column: text
+  is_phonemes: false
+  max_age: null
+tokenizer:
+  name: transformersegmentation/BabyLM-BPE-ortho-tokenizer
+data_preprocessing:
+  max_input_length: 128
+  join_utts: static
+  remove_word_boundaries: false
+  subsample: null
+  subsample_type: examples
+model:
+  name: gpt2_lm
+  model_kwargs:
+    n_layer: 8
+    n_head: 8
+    n_embd: 512
+    n_positions: 256
+    n_inner: 2048
+trainer:
+  batch_size: 32
+  lr: 0.001
+  num_warmup_steps: 90000
+  max_training_steps: 400000
+  logging_steps: 4000
+  save_steps: 50000
+  eval_steps: 50000

model.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:37fc397e0772db251f34de842915546f79cfa9e25a4d75313f8bcc56a1f7564c
-size 108961160
+oid sha256:155163e8e243b53ebd3030460d0369058098d912a8654de8f8f613097d5bd283
+size 134182632

training_args.bin

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:975f4c8fc39f5136ba79137eff2767d6095d8cadb172037c87f30b58b803cd77
+oid sha256:4bdb9bb7a47c8d79dae8b6e300ed84f1e99318dafb96dc9dbd2c7137c0067d4e
 size 5368