p4_probe2_soak_runner: the wrapper as submitted for probe chain v8 / soak / rehearsal v4
Browse files
kernels/p4_probe2_soak_runner.py
ADDED
|
@@ -0,0 +1,18 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
# Kaggle kernel body for P4 probe 2, the soak (2x T4, ~1.2 h). Same pinned body as the smoke, one
|
| 2 |
+
# environment difference: MODE=soak, which runs 180 steps at the real step shape without gradient
|
| 3 |
+
# checkpointing, through three push/verify/prune cycles, a forced cold resume and the horizon's
|
| 4 |
+
# validation pass -- and gates on the memory of the leg that never evaluates. That gate is D-018.
|
| 5 |
+
import hashlib, os, runpy, subprocess, sys
|
| 6 |
+
|
| 7 |
+
os.chdir("/kaggle/working")
|
| 8 |
+
sys.path.insert(0, "/kaggle/working")
|
| 9 |
+
REV = "add04ae7a24bd9925115773a93e870fc1dd01f59"
|
| 10 |
+
SHA = "f55bd29c94deff0a9309e41be2818c49db6720bf871cd60cd43de654810bfaf2"
|
| 11 |
+
MODE = "soak"
|
| 12 |
+
url = f"https://huggingface.co/Cion-lab/ounce100m-code/resolve/{REV}/kernels/p4_stop_probe.py"
|
| 13 |
+
assert subprocess.run(["curl", "-sfL", url, "-o", "p4_stop_probe.py"]).returncode == 0, "fetch probe"
|
| 14 |
+
got = hashlib.sha256(open("p4_stop_probe.py", "rb").read()).hexdigest()
|
| 15 |
+
assert got == SHA, ("PROBE SHA MISMATCH", got[:16], SHA[:16])
|
| 16 |
+
os.environ["P4_PROBE_MODE"] = MODE
|
| 17 |
+
print("PROBE_BOOTSTRAP_OK", REV[:12], got[:12], "mode", MODE, flush=True)
|
| 18 |
+
runpy.run_path("p4_stop_probe.py", run_name="__main__")
|