diff options
Diffstat (limited to 'ep_run/ep_c_check.py')
| -rw-r--r-- | ep_run/ep_c_check.py | 22 |
1 files changed, 22 insertions, 0 deletions
diff --git a/ep_run/ep_c_check.py b/ep_run/ep_c_check.py new file mode 100644 index 0000000..90892d9 --- /dev/null +++ b/ep_run/ep_c_check.py @@ -0,0 +1,22 @@ +import time, os, re, subprocess +os.chdir("/home/yurenh2/ept/ep_run"); LOG="runs/ep_resreg_c.log"; TARGET=2.30 +def alive(): return subprocess.run(["pgrep","-f","ckpt runs/ep_resreg_c.pt"],capture_output=True).returncode==0 +def latest(): + try: ls=[l for l in open(LOG) if l.startswith("step")] + except Exception: return None + if not ls: return None + l=ls[-1] + ms=re.search(r"step\s+(\d+)",l); mv=re.search(r"val CE ([\d.eE+-]+)",l); mb=re.search(r"best ([\d.eE+-]+)",l) + mr=re.search(r"res=([\d.eE+-]+)",l); mi=re.search(r"([\d.]+) it/s",l) + if not (ms and mv and mb and mr): return None + return (int(ms.group(1)),float(mv.group(1)),float(mb.group(1)),float(mr.group(1)), mi.group(1) if mi else "?") +fired=None; t0=time.time() +while fired is None and time.time()-t0<12*3600: + time.sleep(90) + d=latest() + if not alive(): fired=f"ep_resreg_c EXITED; last {d}"; break + if not d: continue + step,val,best,res,its=d + if val>12 or res>0.3: fired=f"ep_resreg_c BLEW @step{step} val{val:.2f} res{res:.1e} best{best:.4f} ({its} it/s)"; break + if best<=TARGET: fired=f"ep_resreg_c reached {TARGET}: step{step} best{best:.4f} val{val:.4f} res{res:.1e} -> {its} it/s (vs eager 0.07) = compile 安全+加速确认"; break +print(f"=== EP_RESREG_C -> {TARGET} (+it/s) ==="); print(fired or "12h timeout"); print("last:",latest()) |
