Training in progress, step 50000

blimp_results.json

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:10604f23a65adc5593915011d1243b48d61f219befad2e7f10a73d6feda78ff4
-size 80226394
+oid sha256:e3c3e5231a5833ea81c3c627178853a917de1900b8646da982207cc3d00067e3
+size 80216800

config.json

@@ -3,20 +3,20 @@
   "architectures": [
     "GPT2LMHeadModel"
   ],
-  "attn_pdrop": 0.3,
+  "attn_pdrop": 0.1,
   "bos_token_id": 0,
-  "embd_pdrop": 0.3,
+  "embd_pdrop": 0.1,
   "eos_token_id": 0,
   "initializer_range": 0.02,
   "layer_norm_epsilon": 1e-05,
   "model_type": "gpt2",
-  "n_embd": 768,
-  "n_head": 12,
-  "n_inner": 3072,
-  "n_layer": 12,
+  "n_embd": 512,
+  "n_head": 8,
+  "n_inner": 2048,
+  "n_layer": 6,
   "n_positions": 256,
   "reorder_and_upcast_attn": false,
-  "resid_pdrop": 0.3,
+  "resid_pdrop": 0.1,
   "scale_attn_by_inverse_layer_idx": false,
   "scale_attn_weights": true,
   "summary_activation": null,

hydra_config_1724933085.2729318.yaml

@@ -0,0 +1,41 @@
+experiment:
+  seed: 42
+  name: gpt2_19M-bpe-text-dyn
+  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: dynamic
+  remove_word_boundaries: false
+  subsample: null
+  subsample_type: examples
+model:
+  name: gpt2_lm
+  model_kwargs:
+    n_layer: 6
+    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 sha256:1ec58f542f7ff628ac674a695b02be0f0c4a42633bd5239f2ef4802047a0bcd5
-size 390177408
+oid sha256:8a1856b83481be0f7a52a705416d3ffb7c8d242e334e4ea6fdab8e61552617d9
+size 108961160

training_args.bin

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:b6bf434d08adffb63320a056f43c3d312c362758a72db108fcc5c86bb963bff1
+oid sha256:5dc80136f7f23f477ebdd661d182228c4de21c404425ed1d103d090bb1541237
 size 5368