AlexWortega commited on
Commit
fd625f9
·
verified ·
1 Parent(s): 63e4224

fix: yaml 1.1 parses '2250.0e12' as string -> float() cast in MFU calc (crashed first log step)

Browse files
Files changed (1) hide show
  1. tinyvla_b200/scripts/train_fast.py +3 -2
tinyvla_b200/scripts/train_fast.py CHANGED
@@ -319,8 +319,9 @@ def main():
319
  torch.cuda.synchronize()
320
  el = time.time() - t0
321
  sps = seen * world / el # глобально по всем рангам
322
- gflops = cfg.get("gflops_per_sample", 360.0)
323
- mfu = sps * gflops * 1e9 / (cfg.get("peak_flops", 989e12)) * 100
 
324
  eta = (steps - step) * eff_batch / sps / 3600
325
  log(f"step {step}/{steps} loss {info['loss']:.4f} | {sps:.0f} samples/s "
326
  f"| {step/el:.2f} it/s | MFU~{mfu:.0f}% | eta {eta:.2f} h "
 
319
  torch.cuda.synchronize()
320
  el = time.time() - t0
321
  sps = seen * world / el # глобально по всем рангам
322
+ # float(): yaml 1.1 парсит "2250.0e12" без знака экспоненты как строку
323
+ gflops = float(cfg.get("gflops_per_sample", 360.0))
324
+ mfu = sps * gflops * 1e9 / float(cfg.get("peak_flops", 989e12)) * 100
325
  eta = (steps - step) * eff_batch / sps / 3600
326
  log(f"step {step}/{steps} loss {info['loss']:.4f} | {sps:.0f} samples/s "
327
  f"| {step/el:.2f} it/s | MFU~{mfu:.0f}% | eta {eta:.2f} h "