eac123 commited on
Commit
b70fd3c
·
verified ·
1 Parent(s): da72c90

stage4: aligned at 100k steps

Browse files
Files changed (1) hide show
  1. run_all.py +15 -2
run_all.py CHANGED
@@ -132,6 +132,18 @@ def stage3(args):
132
  run_eval(args, f"out/{name}.pt")
133
 
134
 
 
 
 
 
 
 
 
 
 
 
 
 
135
  def stage_eval(args):
136
  if not glob.glob("out/*.pt"):
137
  # recovery mode (fresh container): pull every checkpoint from the Hub,
@@ -153,13 +165,14 @@ def stage_eval(args):
153
  def main():
154
  p = argparse.ArgumentParser()
155
  p.add_argument("stage", choices=["lrsweep", "stage1", "stage2", "stage3",
156
- "eval"])
157
  p.add_argument("--hub-repo", default=None)
158
  a = p.parse_args()
159
  if a.hub_repo:
160
  os.environ.setdefault("TRACKIO_HUB_REPO", a.hub_repo)
161
  {"lrsweep": stage_lrsweep, "stage1": stage1, "stage2": stage2,
162
- "stage3": stage3, "eval": stage_eval}[a.stage](a)
 
163
 
164
 
165
  if __name__ == "__main__":
 
132
  run_eval(args, f"out/{name}.pt")
133
 
134
 
135
+ def stage4(args):
136
+ """Aligned objective at 4x the step budget (1 seed): distinguishes
137
+ 'harder optimization, needs more steps' from 'fundamentally harder'
138
+ after the 25k-step aligned runs plateaued near 0.10 id_acc."""
139
+ lr = chosen_lr(args.hub_repo)
140
+ name = "rec1aligned-long-seed0"
141
+ if not os.path.exists(f"out/{name}.pt"):
142
+ run(train_cmd({"arch": "rec", "depth": 1, "iters": "aligned",
143
+ "seed": 0}, name, 4 * STEPS, lr, args.hub_repo))
144
+ run_eval(args, f"out/{name}.pt")
145
+
146
+
147
  def stage_eval(args):
148
  if not glob.glob("out/*.pt"):
149
  # recovery mode (fresh container): pull every checkpoint from the Hub,
 
165
  def main():
166
  p = argparse.ArgumentParser()
167
  p.add_argument("stage", choices=["lrsweep", "stage1", "stage2", "stage3",
168
+ "stage4", "eval"])
169
  p.add_argument("--hub-repo", default=None)
170
  a = p.parse_args()
171
  if a.hub_repo:
172
  os.environ.setdefault("TRACKIO_HUB_REPO", a.hub_repo)
173
  {"lrsweep": stage_lrsweep, "stage1": stage1, "stage2": stage2,
174
+ "stage3": stage3, "stage4": stage4,
175
+ "eval": stage_eval}[a.stage](a)
176
 
177
 
178
  if __name__ == "__main__":