Repository navigation

ggml-hrx: update upstream and enable Flint through 32K context - #158

Draft

AaronStGeorge wants to merge 23 commits into

hrx-graph-develop-v2from

users/astgeorg/flint-hrx-support

Draft

ggml-hrx: update upstream and enable Flint through 32K context#158
AaronStGeorge wants to merge 23 commits into
hrx-graph-develop-v2from
users/astgeorg/flint-hrx-support

Commits on Aug 17, 2026
- ggml-hrx: add AMD ROCm HRX native ggml backend
  stellaraccident
  committed
  33a1f2b
Commits on Oct 7, 2026
- ggml-hrx: squash development through 8108322b7
  
  committed
  53d443c
Commits on Oct 8, 2026
- ggml-hrx: adopt canonical Loom divisibility predicates (#139)
  AaronStGeorge
  authored
  d9e2dc4
Commits on Oct 12, 2026
- hrx: accept empty MoE expert views when matching graphs (#140)
  AaronStGeorge
  authored
  d1b0eb2
Commits on Oct 13, 2026
- ggml-hrx: prove packet bounds before view linearization (#141)
  AaronStGeorge
  authored
  775aa77
Commits on Oct 14, 2026
- ggml-hrx: size fused router launch for target expert tiles (#146)
  AaronStGeorge
  authored
  1bdb972
- ggml-hrx: adopt exclusively owned Loom link-index builders (#147)
  AaronStGeorge
  authored
  dbe1d6e
Commits on Oct 15, 2026
- ggml-hrx: reuse block metadata in dense and paired tiled matmul (#148)
  AaronStGeorge
  authored
  43b0ec2
Commits on Oct 21, 2026
- ggml-hrx: reuse IQ3_S metadata and Q5 loads (#150)
  AaronStGeorge
  authored
  e086347
- ggml-hrx: unify MoE partition descriptors and cover all experts (#153)
  AaronStGeorge
  authored
  7fd4e51
Commits on Oct 23, 2026
- hrx: publish decode attention output before Q8 packing (#154)
  AaronStGeorge
  authored
  1ec5a7a
Commits on Oct 26, 2026
- ggml-hrx: fix suite dispatch expectations and concat bounds (#156)
  AaronStGeorge
  authored
  dace247
Commits on Oct 27, 2026
Commits on Oct 28, 2026