Repository navigation

[perf][rocm]: accelerate strict VIME rollout with direct paged CK and graph-safe execution - #390

Merged

inaniloquentee merged 17 commits into

testfrom

perf/rocm-vime-rollout-overheads

Sep 8, 2026

Commits on Sep 6, 2026
- perf(rocm): reduce strict paged decode overheads
  Flink-ddd
  committed
  c9922db
- perf(rocm): write deterministic GEMM roots directly
  Flink-ddd
  committed
  b8e1054
- perf(rocm): reuse rotary tables across q and k
  Flink-ddd
  committed
  1136880
- perf(rocm): write paged attention groups into caller output
  Flink-ddd
  committed
  47e1fd1
- perf(rocm): cache prepared LM-head weights
  Flink-ddd
  committed
  af866d2
- perf(rocm): bypass RoPE autograd in rollout
  Flink-ddd
  committed
  336a2d7
- perf(rocm): cache rotary inverse frequencies
  Flink-ddd
  committed
  c5b122e
- perf(rocm): reuse paged bounds validation across layers
  Flink-ddd
  committed
  9df220f
- perf(rocm): skip singleton rollout LSE copies
  Flink-ddd
  committed
  b9ed179
- perf(rocm): gather paged KV directly in head-major order
  Flink-ddd
  committed
  5c1e4a8
- bench(rocm): model the vllm pair-axis KV cache
  Flink-ddd
  committed
  47bc20a
- perf(rocm): cache immutable attention provenance inputs
  Flink-ddd
  committed
  c70efd7
Commits on Sep 7, 2026
- perf(rocm): batch strict decode KV groups
  Flink-ddd
  committed
  53b691e
- perf(rocm): collapse fresh paged prefill rows
  Flink-ddd
  committed
  75cdabe
- perf-rocm-optimize-strict-VIME-rollout-scheduling
  inaniloquentee
  committed
  e92bf56
- fix-rocm-adapter-identifiers
  inaniloquentee
  committed
  a2ef32c
Commits on Sep 8, 2026
- fix(rocm): make piecewise HIP graph replay exact
  inaniloquentee
  committed
  fed8362

Navigation Menu

Repository navigation

[perf][rocm]: accelerate strict VIME rollout with direct paged CK and graph-safe execution - #390

[perf][rocm]: accelerate strict VIME rollout with direct paged CK and graph-safe execution#390
inaniloquentee merged 17 commits into
testfrom
perf/rocm-vime-rollout-overheads

Commits on Sep 6, 2026

perf(rocm): reduce strict paged decode overheads

perf(rocm): write deterministic GEMM roots directly

perf(rocm): reuse rotary tables across q and k

perf(rocm): write paged attention groups into caller output

perf(rocm): cache prepared LM-head weights

perf(rocm): bypass RoPE autograd in rollout

perf(rocm): cache rotary inverse frequencies

perf(rocm): reuse paged bounds validation across layers

perf(rocm): skip singleton rollout LSE copies

perf(rocm): gather paged KV directly in head-major order

bench(rocm): model the vllm pair-axis KV cache

perf(rocm): cache immutable attention provenance inputs

Commits on Sep 7, 2026

perf(rocm): batch strict decode KV groups

perf(rocm): collapse fresh paged prefill rows

perf-rocm-optimize-strict-VIME-rollout-scheduling

fix-rocm-adapter-identifiers

Commits on Sep 8, 2026

fix(rocm): make piecewise HIP graph replay exact