llama.cpp
OpenVINO: Qwen3.5, memory optimization, and test-recurrent-state-rollback
#26952
Merged
Merged
OpenVINO: Qwen3.5, memory optimization, and test-recurrent-state-rollback #26952
ggerganov
merged
59 commits into
ggml-org:master
from
ravi9:pr-26-08-11
OpenVINO backend: 1) enable gpt-oss moe on OV bk; 2) enable mxfp4 sup…
65db5901
OpenVINO backend: disable TOPK_MOE op test
44fcf817
OpenVINO Backend: Add op FILL support
c867f51b
OpenVINO backend: enable set rows with multi dims
53ce035f
fix the name missmatch in setrow + view
c1d775f8
OpenVINO backend: enable op GGML_UNARY_OP_SIGMOID
186db4fa
OpenVINO Backend: enable SQR & SQRT
68da958c
OpenVINO backend: 1) ensure unique node names for OpenVINO; 2) add or…
7631aaff
OpenVINO backend: enable fallback for openVINO to CPU backend
0529e9ee
OpenVINO backend: fix accurace issue in gemma3n arch test
bd532e42
fix mpt failed case
026f1d07
OpenVINO backend: clean nodeinfo
df1fbfd5
OpenVINO Backend: enable zero-size copy for view
b2015595
add concat ssm_conv in compute_dynamic_dim
116c0488
OpenVINO backend: disable EXP with FP32, which failed in op test. Roo…
98c9e937
OpenVINO backend: fix CPY op test failed issue
2736fa5f
OpenVINO backend: fix GATED_DELTA_NET op test failed issue
63fdf3d7
handle in-place op, handle qwen35 dynamic clearing of cache in cgraph
f223472c
handle qwen35 dynamic clearing of cache correctly
98cbce09
Enable qwen35 dense multi seq
af70b49e
Fix qwen35 9b gqa
26f6a73a
Fix after rebase
1be15a90
Disable SOLVE_TRI
3ab8fce5
openvino: fix NEOX RoPE accuracy on GPU stateful (mixed-rank Multiply)
1192ab5c
OpenVINO backend: 1) remove the unique name in llama.cpp; 2) add new …
0ab29cb4
OpenVINO Backenb: remove changes in llama.cpp
f2088c52
Doc change (use x64 Native Tools Command Prompt for VS)
430d1dbd
Cleaner sentence
0f0047e6
OpenVINO Backend: cache key upgrade includes all src name
7871c831
OpenVINO Backend: enable llama arch test on ci
7da52bc2
OpenVINO Backend: move parameter node creating from decoder into tran…
a5bd6541
OpenVINO Backend: create extra input ov node move from decoder to tra…
3d999cb8
fix for op regression due to is_model_splitted
9dd17d0c
openvino: fix CPY writeback for recurrent state rollback
e2e52aca
qwen35 moe
e2b116eb
optimize MoE expert aggregation with ReduceSum
0303d278
Skip GET_ROWS inaccurate test
36feef47
openvino: fallback dynamic MUL_MAT_ID shapes
2d6c8bf8
OpenVINO Backend: fix error in arch test model mpt
38856999
fix error caused by cpy in arch test model kimi-linear
518e5078
OpenVINO Backend: fix error in arch test model minimax-m3
3485ef3c
openvino: fix GPU mul_mat_id op tests
4a0e802a
ggml-openvino: add GGML_OPENVINO_RELEASE_WEIGHTS to reclaim host weig…
524f869d
ggml-openvino: stream weight requantization to cut the compile-time R…
ef9f854e
ggml-openvino: avoid redundant token_embd requantization at compile
868b0223
ggml-openvino: gate compile-memory optimizations behind GGML_OPENVINO…
63e06ce1
ggml-openvino: add frontend model cache (GGML_OPENVINO_MODEL_CACHE_DIR)
747b4e9a
ggml-openvino: harden frontend model cache correctness
4be0fd7d
ggml-openvino: add memory optimization umbrella switch
db15f2e5
ggml-openvino: rename compiled model cache env
a0108b07
docs: document OpenVINO memory optimization env vars
30905fee
ggml-openvino: fix Qwen3VL crash and deepstack correctness bug
12f3c783
ggml-openvino: fix IMROPE inp_pos padding for NPU static shapes
6491854e
disable test-llama-archs tests.
289ddb3d
openvino: gate fallback with env var
c66a9c98
Revert changes in test-llama-archs
00457cef
Apply editor config
2e0c1be7
reject CPY with quantized destination as unsupported
ae02f5ef
Merge pull request #283 from ravi9/ravi9-patch-1
37b164ff
Reviewers
No reviews
Assignees
No one assigned
Labels
documentation
ggml
OpenVINO
Milestone
No milestone
Login to write a write a comment.
Login via GitHub