Datasets:

HarshTri007
/

FlakeForge_Blog

Dataset card Files Files and versions

HarshTri007 commited on 15 days ago

Commit

78d7d90

·

verified ·

1 Parent(s): 35e7c3c

Update BLOG.md

Files changed (1) hide show

BLOG.md +3 -3

BLOG.md CHANGED Viewed

@@ -343,11 +343,11 @@ The logs below show what one rollout “looks like” at the environment level:
 - **`[ENV] REWARD total=... breakdown={...}`**: the scalar learning signal plus per-term contributions (format/compile/stability/entropy/anti-hack/oracle/etc).
 - **`[ENV] PREFLIGHT ... should_train: False ... reason: ...`**: the gate that skips cases that are stable, deterministic bugs, or infra-broken (so GRPO budget goes to true flaky-like episodes).
-![Training log excerpt 1](images/IMG1.jpeg)
-![Training log excerpt 2](images/IMG2.jpeg)
-![Training log excerpt 2](images/IMG3.jpeg)
 ---

 - **`[ENV] REWARD total=... breakdown={...}`**: the scalar learning signal plus per-term contributions (format/compile/stability/entropy/anti-hack/oracle/etc).
 - **`[ENV] PREFLIGHT ... should_train: False ... reason: ...`**: the gate that skips cases that are stable, deterministic bugs, or infra-broken (so GRPO budget goes to true flaky-like episodes).
+![Training log excerpt 1](https://drive.google.com/file/d/1I5_ft33gA41sgkt82HCxJEHyUzA2CfDD/view?usp=sharing)
+![Training log excerpt 2](https://drive.google.com/file/d/1hf0ePnWugxn-SO4RVB084ZBdrwZZbbm6/view?usp=sharing)
+![Training log excerpt 2](https://drive.google.com/file/d/1r3SLUC7rRKqLdo4WnAUob7l9Vm84TOiq/view?usp=sharing)
 ---