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

docs(dqn): T6 polish — fix 4 stale Pearl 2 prose references in gpu_dqn_trainer.rs

Browse Source 
Four comment/docstring sites still described the pre-SP7 Pearl 2 as
writing "20 floats / 5 outputs / c51/cql/ens" after the T6 kernel-
signature shrink. Updated to reflect the current 6-arg kernel writing
8 floats (IQN + flatness only) into scratch[115..119, 127..131).

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
This commit is contained in:
jgrusewski
2026-05-03 01:50:58 +02:00
parent 805a7429f4
commit d704ffd1bc
2 changed files with 12 additions and 11 deletions
Download Patch File Download Diff File Expand all files Collapse all files

19
crates/ml/src/cuda_pipeline/gpu_dqn_trainer.rs
View File

@@ -304,17 +304,16 @@ static SP5_PEARL_1_ATOM_CUBIN: &[u8] =
static SP5_PEARL_3_SIGMA_CUBIN: &[u8] =
include_bytes!(concat!(env!("OUT_DIR"), "/pearl_3_sigma_kernel.cubin"));
/// SP5 Task A3 (2026-05-01): Pearl 2 per-branch loss budget cubin.
/// SP5 Task A3 + SP7 (2026-05-03): Pearl 2 IQN budget + flatness producer cubin.
/// Single-block 4-thread kernel. Reads Q_VAR_PER_BRANCH[b] (Pearl 1's
/// q_branch_stats output, ISV[222..226)) + NOISY_SIGMA[b] (Pearl 3 output,
/// ISV[210..214)). Computes per-branch flatness = clamp(var_q / (σ²+EPS_DIV),
/// 0, 1) and derives (budget_iqn, budget_c51, budget_cql=0, budget_ens,
/// flatness) per branch. Writes 20 floats to
/// producer_step_scratch_buf[111..131). Chained apply_pearls_ad_kernel
/// (20 single-slot launches) smooths via Pearls A+D into ISV[190..210).
/// 0, 1) and derives (budget_iqn, flatness) per branch. Writes 8 floats to
/// producer_step_scratch_buf[115..119, 127..131). Chained apply_pearls_ad_kernel
/// (8 single-slot launches) smooths via Pearls A+D into ISV[194..198, 206..210).
/// SP7: C51/CQL/ENS budget ownership transferred to loss_balance_controller_kernel.
/// Must run AFTER launch_sp5_pearl_1_atom (Q_VAR) AND
/// launch_sp5_pearl_3_sigma (NOISY_SIGMA).
/// Layer A only — loss-budget consumer migration deferred to Layer B.
static SP5_PEARL_2_BUDGET_CUBIN: &[u8] =
include_bytes!(concat!(env!("OUT_DIR"), "/pearl_2_budget_kernel.cubin"));
@@ -1069,7 +1068,7 @@ pub const SP5_WIENER_TOTAL_FLOATS: usize =
/// 4 floats for pearl_3_sigma output (SCRATCH_PEARL_3_SIGMA=103, sigma × 4 branches)
/// 4 floats for pearl_3_sf output (SCRATCH_PEARL_3_SF=107, sigma_fraction × 4 branches)
/// Task A3 adds:
/// 20 floats for pearl_2_budget output (5 outputs × 4 branches at [111..131))
/// 8 floats for pearl_2_budget output (2 outputs × 4 branches; SP7 retired C51/CQL/ENS slots)
/// [111..115) reserved-for-future (T6 retired SCRATCH_PEARL_2_C51)
/// budget_iqn[4] → SCRATCH_PEARL_2_IQN=115
/// [119..127) reserved-for-future (T6 retired SCRATCH_PEARL_2_CQL/_ENS)
@@ -14331,9 +14330,9 @@ impl GpuDqnTrainer {
.map_err(|e| MLError::ModelError(format!("pearl_3_sigma_update load: {e}")))?
};
// SP5 Task A3 (2026-05-01): load pearl_2_budget loss budget controller kernel.
// SP5 Task A3 + SP7 (2026-05-03): load pearl_2_budget IQN budget + flatness kernel.
// Single-block 4-thread kernel; reads Q_VAR_PER_BRANCH (ISV[222..226)) +
// NOISY_SIGMA (ISV[210..214)) from ISV; writes 20 floats to scratch[111..131).
// NOISY_SIGMA (ISV[210..214)) from ISV; writes 8 floats to scratch[115..119, 127..131).
// Must be invoked AFTER pearl_1_atom (Q_VAR) AND pearl_3_sigma (NOISY_SIGMA).
let pearl_2_budget_kernel = {
let module = stream.context().load_cubin(SP5_PEARL_2_BUDGET_CUBIN.to_vec())
@@ -16082,7 +16081,7 @@ impl GpuDqnTrainer {
// [87..103) pearl_1_atom_update (Pearl 1 / A1)
// [103..107) pearl_3_sigma σ (Pearl 3 / A2)
// [107..111) pearl_3_sigma SF (Pearl 3 / A2)
// [111..131) pearl_2_budget (Pearl 2 / A3 — c51/iqn/cql/ens/flatness)
// [111..131) pearl_2_budget (Pearl 2 / A3 — iqn/flatness; SP7 retired c51/cql/ens)
// [131..147) pearl_4 cosine + l2 (auxiliary grad_cosine_sim, A4)
// [147..171) pearl_4 β1/β2/ε (Pearl 4 / A4)
// [171..199) pearl_5 skew/ex_kurt + IQN τ schedule (A5)

4
docs/dqn-wire-up-audit.md
View File

@@ -3910,6 +3910,8 @@ extended. No producer kernel yet — that arrives in the next commit.
apply_pearls smoothing loop shrinks from 5 slot-blocks to 2 (IQN +
flatness). SCRATCH_PEARL_2_C51/_CQL/_ENS deleted as orphans per
feedback_wire_everything_up. CQL/C51/ENS budget ownership now lives
entirely in the SP7 controller.
entirely in the SP7 controller. T6 polish: 4 stale prose references
(cubin static docstring, scratch layout header, constructor load
comment, scratch map) updated to reflect 8-float / 2-output contract.
- T7: launch site + sentinel-aware bootstrap with bootstrap constants matching the kernel's cold-start basis (defined in T7) + stale doc.
- T9–T10: smoke + 50-epoch verification.