Repository navigation

Update the llama.cpp to utilize OV frontend op translations - #342

Open

preetha-intel wants to merge 28 commits into

ravi9:dev_backend_openvinofrom

preetha-intel:ov-gguf-frontend

Open

Update the llama.cpp to utilize OV frontend op translations#342
preetha-intel wants to merge 28 commits into
ravi9:dev_backend_openvinofrom
preetha-intel:ov-gguf-frontend

Commits on Oct 4, 2026
- ggml-openvino: fix SWA mask detection
  osabnis
  committed
  15dff06
- ggml-openvino: fix SWA KV cache write indices
  osabnis
  committed
  3f023f5
- ggml-openvino: classify SWA layers only when the mask shows the window
  osabnis
  committed
  706a4d9
Commits on Oct 5, 2026
- ggml-openvino: classify SWA layers from the first graph
  osabnis
  committed
  e98f645
Commits on Oct 6, 2026
- ggml-openvino: skip unselected graph branches and support DUP
  ravi9
  committed
  8c174e4
- ggml-openvino: make inp_scale_rows token dim dynamic
  ravi9
  committed
  130d8bd
- ggml-openvino: skip GPU MUL_MAT op tests with unbound Q4_1/Q4_K weights
  ravi9
  committed
  0087a1b
- ggml-openvino: create FILL in the output type
  ravi9
  committed
  672083f
- ggml-openvino: reject CONCAT with a quantized type
  ravi9
  committed
  5f2e77e
- ggml-openvino: handle the single recurrent state gather of build_rs
  ravi9
  committed
  627e811
- openvino: align eltwise operand ranks to work around a GPU-plugin defect
  
  cavusmustafa
  authored and
  ravi9
  committed
  5802ab8
- openvino: match the MoE fusion on the rank-3 stateful graph
  
  cavusmustafa
  authored and
  ravi9
  committed
  409ba45
- ggml-openvino: do not unsqueeze an RMS norm output in AlignEltwiseOperandRanks
  ravi9
  committed
  39dacd4
- docs : update OpenVINO validated models
  ravi9
  committed
  a4493f9
- ggml-openvino: derive attention sizes from the KQ mask shape
  osabnis
  committed
  84232f4
- ggml-openvino: only reuse Split outputs for exact chunk-aligned views
  osabnis
  committed
  0aa29c7
- ggml-openvino: mark SWA fallback as a model-level heuristic
  osabnis
  committed
  e127d6b
Commits on Oct 7, 2026
- Merge pull request #336 from osabnis/ov-gemma4-swa-fixes
  cavusmustafa
  authored
  7906a48
Commits on Oct 9, 2026
- ggml-openvino: initialize ModelExtraInputInfo shape fields
  osabnis
  committed
  e706353
- ggml-openvino: remove GGML_OPENVINO_REQUANT_EMBD
  osabnis
  committed
  dfb6ad9
- Merge pull request #339 from osabnis/ov-dynamic-decode
  cavusmustafa
  authored
  ab759cf
Commits on Oct 12, 2026
- ggml-openvino: keep manual GQA attention 4-D
  osabnis
  committed
  2e4dbf0
- Merge pull request #343 from osabnis/ov-gqa-attn-4d
  cavusmustafa
  authored
  f308135
- ggml-openvino: use a 4-D stateless KV cache on GPU
  osabnis
  committed
  dd04231
- Merge pull request #344 from osabnis/ov-kv-cache-4d
  cavusmustafa
  authored
  8619618
Commits on Oct 13, 2026
- Remove legacy OpenVINO bridge code
  preetha-intel
  committed
  c541da7
Commits on Oct 15, 2026
- ggml-openvino: add eltwise rank alignment and update decoder/utils
  preetha-intel
  committed
  8faf3c5
Commits on Oct 23, 2026
- ggml-openvino: apply scale before MatMul on GPU/NPU to keep f16 in range
  preetha-intel
  committed
  6abf2ae

Commits

Commits on Oct 4, 2026

Commits on Oct 5, 2026

Commits on Oct 6, 2026

Commits on Oct 7, 2026

Commits on Oct 9, 2026

Commits on Oct 12, 2026

Commits on Oct 13, 2026

Commits on Oct 15, 2026

Commits on Oct 23, 2026