luo352 commited on
Commit
f740fa5
·
verified ·
1 Parent(s): f1d05bc

Document 8k PG-OPD scaling data

Browse files
Files changed (1) hide show
  1. README.md +2 -0
README.md CHANGED
@@ -30,6 +30,8 @@ not interchangeable.
30
 
31
  `Qwen35-4B/pg_opd.parquet` contains 2,394 rows:
32
 
 
 
33
  - 1,197 verified evidence-privileged rows;
34
  - 1,197 native fallback rows;
35
  - evidence was selected from the 4B Stage-A filtering run.
 
30
 
31
  `Qwen35-4B/pg_opd.parquet` contains 2,394 rows:
32
 
33
+ `Qwen35-4B/pg_opd_8k_forward_kl.parquet` contains 8,000 rows for the forward-KL data-scaling ablation (4,000 verified privilege rows and 4,000 native-control rows). This is an auxiliary scaling artifact; the final reported recipe uses JSD (beta=0.5).
34
+
35
  - 1,197 verified evidence-privileged rows;
36
  - 1,197 native fallback rows;
37
  - evidence was selected from the 4B Stage-A filtering run.