Add easy question scaling law dataset from ExpEmergence paper (U-shaped and Inverted-U Scaling behind Emergent Abilities of LLMs)

README.md

@@ -54,6 +54,27 @@ dataset_info:
     num_examples: 24
   download_size: 15342
   dataset_size: 9126
+- config_name: easy_question_scaling_law
+  features:
+  - name: group
+    dtype: string
+  - name: log_flops
+    dtype: float64
+  - name: flops
+    dtype: float64
+  - name: brier_score
+    dtype: float64
+  - name: model_name
+    dtype: string
+  splits:
+  - name: train
+    num_bytes: 27304
+    num_examples: 389
+  - name: test
+    num_bytes: 8530
+    num_examples: 127
+  download_size: 13707
+  dataset_size: 35834
 - config_name: lr_bsz_scaling_law
   features:
   - name: group
@@ -189,6 +210,12 @@ configs:
     path: domain_mixture_scaling_law/train-*
   - split: test
     path: domain_mixture_scaling_law/test-*
+- config_name: easy_question_scaling_law
+  data_files:
+  - split: train
+    path: easy_question_scaling_law/train-*
+  - split: test
+    path: easy_question_scaling_law/test-*
 - config_name: lr_bsz_scaling_law
   data_files:
   - split: train

easy_question_scaling_law/test-00000-of-00001.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:946e8f735c0963cf86a56a83c20058a500a73365891ae24c9cb33192f8bcef81
+size 5287

easy_question_scaling_law/train-00000-of-00001.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:40b81eb072d48abd9f47acd8200851978d37cbcedf5e2c5bb8b0b4f16ba17c07
+size 8420