Skip to content

ggml-zdnn: fix fattn test failure - #31583

Merged
taronaeo merged 1 commit into
ggml-org:masterfrom
taronaeo:fix/zdnn-fattn-kv-rows
Nov 3, 2026
Merged

taronaeo merged 1 commit into
ggml-org:masterfrom
taronaeo:fix/zdnn-fattn-kv-rows

Conversation

@taronaeo

@taronaeo taronaeo commented Nov 3, 2026

Copy link
Copy Markdown
Member

Overview

This PR fixes a test failure with Flash Attention on the IBM zDNN backend.

Additional information

Failure:

FLASH_ATTN_EXT(hsk=128,hsv=128,nh=1,nr23=[8,1],kv=4096,nb=4,mask=1,sinks=0,max_bias=8.000000,logit_softcap=0.000000,prec=f32,type_K=f16,type_V=f16,permute=[0,1,2,3],kv_view=1,v_is_view_of_k=0,n_kv_max=512): not supported [zDNN] 
WARN: aiu_ops_func_specific() (aiu_ops.c:194): ZDNN_ELEMENT_RANGE_VIOLATION: Range violation on tensor data
/home/linux1/zdnn-float-inputs/master/ggml/src/ggml-zdnn/fattn.cpp:169: GGML_ASSERT(status == ZDNN_OK) failed
/home/linux1/zdnn-float-inputs/master/build/bin/libggml-base.so.0(+0x152fe) [0x3ffb6b152fe]
/home/linux1/zdnn-float-inputs/master/build/bin/libggml-base.so.0(+0x157ae) [0x3ffb6b157ae]
/home/linux1/zdnn-float-inputs/master/build/bin/libggml-base.so.0(ggml_abort+0x124) [0x3ffb6b15964]
/home/linux1/zdnn-float-inputs/master/build/bin/libggml-zdnn.so.0(_Z24ggml_zdnn_flash_attn_extP25ggml_backend_zdnn_contextP11ggml_tensor+0x2cda) [0x3ffb608d78a]
/home/linux1/zdnn-float-inputs/master/build/bin/libggml-zdnn.so.0(+0xef58) [0x3ffb608ef58]
/home/linux1/zdnn-float-inputs/master/build/bin/libggml-base.so.0(ggml_backend_graph_compute+0x16) [0x3ffb6b329c6]
/home/linux1/zdnn-float-inputs/master/build/bin/libggml-base.so.0(ggml_backend_compare_graph_backend+0x1e8) [0x3ffb6b39a58]
/home/linux1/zdnn-float-inputs/master/build/bin/test-backend-ops(+0x13193c) [0x2aa2923193c]
/home/linux1/zdnn-float-inputs/master/build/bin/test-backend-ops(+0x8a328) [0x2aa2918a328]
/home/linux1/zdnn-float-inputs/master/build/bin/test-backend-ops(main+0x350e) [0x2aa2912535e]
/usr/lib/s390x-linux-gnu/libc.so.6(+0x2c1c0) [0x3ffb642c1c0]
/usr/lib/s390x-linux-gnu/libc.so.6(__libc_start_main+0xac) [0x3ffb642c31c]
/home/linux1/zdnn-float-inputs/master/build/bin/test-backend-ops(+0x25e00) [0x2aa29125e00]

Requirements

Signed-off-by: Aaron Teo <aaron.teo1@ibm.com>
@taronaeo
taronaeo requested a review from a team as a code owner November 3, 2026 08:40
@github-actions github-actions Bot added ggml changes relating to the ggml tensor library for machine learning IBM zDNN issues specific to IBM zDNN Accelerator labels Nov 3, 2026

@AlekseiNikiforovIBM AlekseiNikiforovIBM left a comment

Copy link
Copy Markdown
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

LGTM

@taronaeo
taronaeo merged commit a464960 into ggml-org:master Nov 3, 2026
25 checks passed
edwardyoon pushed a commit to edwardyoon/focus-llama that referenced this pull request Nov 6, 2026
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

ggml changes relating to the ggml tensor library for machine learning IBM zDNN issues specific to IBM zDNN Accelerator

Projects

None yet

Development

Successfully merging this pull request may close these issues.

3 participants