Strata

0 Fork 0

mirror of https://github.com/Niko1221/Strata.git synced 2026-11-07 06:15:58 -10:00

Files

T

bitaccesscomau 386f9959f1 Exact two-pass QSA prompt selection (opt-in STRATA_SELECT_TWO_PASS=1)

Prompt-path block scores in single-pass TF32 with a per-block error bound,
3xTF32 rescoring only near each query's top-k cut-off, so qsa_block_topk
returns the same cells in the same order, ties included. Off unless the
variable is set; says once where it cannot run (HIP, Metal, SYCL, layer-major
path). Adds qsa_two_pass_test, qsa_two_pass_status_test and the GPU parity
test qsa_two_pass_parity.

Ported from PR 4119 (bitaccesscomau); the log line of the per-prompt counters
is shortened without dropping a field.

l620 (sm_120, IQ3_S), greedy ids identical off vs on in every pair:
prompt 128K +1.65% (n=12), 256K +4.10% (n=8), 32K -0.14% (noise).

(cherry picked from commit bbf7c55f1351d913233250fe7869642f7c2b98a1)

2026-11-06 15:47:43 +01:00

artifact

Gyro-L: read the pair-trellis types TQP5/TQP6/TQP7 (ggml 147-149) on the CPU, CUDA and the MMQ prompt path (#3806 )

2026-10-27 01:57:57 +01:00

core

Run manifest (lite): STRATA_MANIFEST=<file> writes the selected fast paths, fallback reasons, tensor formats, engine choice and budgets as JSON (P0.1)

2026-11-06 00:21:53 +01:00

dsv41

DeepSeek PCIe lane and whole-token judge from Project Maya (#312 Peasant Smith, #329 mw00), staged merge v1.1.17 to v1.1.19

2026-10-31 01:10:17 +01:00

hip_compat

RDNA4 routed-expert, draft, shared-expert and head kernels (PR 3259 and PR 3264, author bsorensen110)