Niko1221 6fb110e557 Opt-in research switches: STRATA_PAR_COUNTERS, STRATA_PIPED_CONCURRENT_MIN, multi-prompt layer-major read (Prefill::run_multi, BCO co-reads, server STRATA_COALESCE_PREFILL), STRATA_CRASH_BT

2026-10-30 11:08:53 +01:00

..

adapt_decay.hpp

engine: --adapt-decay outside 0..1 falls back to 0.7 with a line on stderr (#1284)

2026-10-26 18:35:58 +01:00

batch_groups.hpp

batch groups: two NVIDIA cards pipeline by default; the serial group is opt-in (STRATA_BATCH_GROUPS_NV_SERIAL=1, STRATA_DEFAULTS=026)

2026-10-27 04:17:17 +01:00

batch_read.hpp

batch-groups: decode and give way between the pieces of a prompt read (#2050, opt-in STRATA_PIPED_READ_PIECE), on top of #2326's concurrent read

2026-10-14 12:26:10 +02:00

conv_cache.hpp

Conversation cache: skip a restore only when a card would have to create its copy stream and has no room for it (#2409)

2026-10-27 02:02:46 +01:00

draft_kv_plan.hpp

split: batched draft-layer prompt K/V on the last stage (STRATA_SPLIT_MTP_BATCH=1, opt-in)

2026-10-09 03:06:46 +02:00

gpu_report.hpp

Later-card VRAM: audit fixes for failed allocations, the reserve arithmetic as a tested helper, a per-card out-of-memory report (#3638 #3702 #3797)

2026-10-27 02:02:46 +01:00

logits_selection.hpp

Strata: Qwen3.8-Flash-Next on a 12-24 GB NVIDIA GPU + 64 GB RAM, one-click install

2026-09-24 18:42:20 +02:00

message_boundary.hpp

Reuse long chat history when the last user message is edited

2026-10-05 19:57:35 +02:00

par_counters.hpp

Opt-in research switches: STRATA_PAR_COUNTERS, STRATA_PIPED_CONCURRENT_MIN, multi-prompt layer-major read (Prefill::run_multi, BCO co-reads, server STRATA_COALESCE_PREFILL), STRATA_CRASH_BT

2026-10-30 11:08:53 +01:00

pipe_probe.hpp

Verify and batch windows of up to 16 rows: HIP builds default to 16, narrow builds unchanged

2026-10-29 15:06:02 +01:00

prefill_plan.hpp

Prefill planner: request-sized chunks. A prompt that fits today's chunk keeps one chunk; a longer one reads in equal chunks of at most the plan's size (no stub tail). Buffers stay sized for today's chunk.

2026-10-18 09:20:24 +02:00

split_ring.hpp

split: decide STRATA_SPLIT_RING before sizing the lend regions; log borrowed vs RAM-kept slots

2026-10-09 03:06:43 +02:00

split_rules.hpp

layer split auto: a two-GPU rig whose first pick has a tiny stage tries each card's share of the room before the three-layer rule (#2644)

2026-10-15 16:31:45 +02:00

stage_reserve.hpp

Later-card VRAM: audit fixes for failed allocations, the reserve arithmetic as a tested helper, a per-card out-of-memory report (#3638 #3702 #3797)

2026-10-27 02:02:46 +01:00