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

chore: remove unused NaN diagnostic accessor methods

Browse Source 
Removes cql_grad_scratch_ptr, iqn_trunk_m_ptr, save_h_s1_ptr,
stream_cu — added during NaN investigation, no longer called.

Retains run_nan_tracing (lightweight GPU-side checks on first 20 steps)
and all isfinite guards as production safety nets.

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
This commit is contained in:
jgrusewski
2026-04-05 13:40:12 +02:00
parent 7b6e676952
commit c9191e7c58
1 changed files with 0 additions and 20 deletions
Download Patch File Download Diff File Expand all files Collapse all files

20
crates/ml/src/cuda_pipeline/gpu_dqn_trainer.rs
View File

@@ -1026,26 +1026,6 @@ impl GpuDqnTrainer {
self.params_buf.raw_ptr()
}
/// Raw pointer to CQL gradient scratch buffer (for NaN diagnostics).
pub fn cql_grad_scratch_ptr(&self) -> u64 {
self.cql_grad_scratch.raw_ptr()
}
/// Raw pointer to IQN trunk gradient scratch (for NaN diagnostics).
pub fn iqn_trunk_m_ptr(&self) -> u64 {
self.ptrs.iqn_trunk_m
}
/// Raw pointer to saved h_s1 activations (for NaN diagnostics).
pub fn save_h_s1_ptr(&self) -> u64 {
self.save_h_s1.raw_ptr()
}
/// Raw CUDA stream handle (for synchronous diagnostics).
pub fn stream_cu(&self) -> cudarc::driver::sys::CUstream {
self.stream.cu_stream()
}
/// Total number of trainable parameters.
pub fn total_params(&self) -> usize {
self.total_params