gitadmin/foxhunt
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

fix(diag): replace hardcoded replay_len=0 with actual GPU PER capacity

Browse Source 
The old host replay.len() was replaced with 0usize during T1 cleanup.
GPU PER is working fine — just not reporting its length in the diag.

Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
This commit is contained in:
jgrusewski
2026-05-26 00:58:48 +02:00
parent 33e14155fe
commit 1ddf74fb51
1 changed files with 3 additions and 3 deletions
Download Patch File Download Diff File Expand all files Collapse all files

6
crates/ml-alpha/examples/alpha_rl_train.rs
View File

@@ -885,7 +885,7 @@ fn main() -> Result<()> {
"stale": isv[RL_LR_V_STEPS_SINCE_BEST_INDEX],
"warmup": isv[RL_LR_V_WARMUP_COUNTER_INDEX] },
},
"replay_len": 0usize,
"replay_len": trainer.gpu_replay.capacity.min(step + 1),
// Post-scale, POST-clamp reward stats — what V regression
// and Q distributional projection actually saw this step.
// The `scaled_pre_clamp_max` field (below) is the SAME
@@ -1152,7 +1152,7 @@ fn main() -> Result<()> {
isv[RL_PPO_CLIP_INDEX],
isv[RL_PER_ALPHA_INDEX],
isv[RL_REWARD_SCALE_INDEX],
0usize,
trainer.gpu_replay.capacity.min(step + 1),
done_count,
reward_sum,
sps,
@@ -1298,7 +1298,7 @@ fn main() -> Result<()> {
summary.final_l_aux = s.l_aux;
summary.final_l_total = s.l_total;
}
summary.final_replay_len = 0usize;
summary.final_replay_len = trainer.gpu_replay.capacity.min(cli.n_steps);
summary.completed_clean = summary.n_steps_completed == summary.n_steps_planned;
write_summary(&cli.out, &summary)?;