Skip to content

Fix accuracy issue and enable Arctic and Grok for arch tests #183

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
zhaixuejun1993 merged 3 commits into from
May 25, 2026
Merged

Fix accuracy issue and enable Arctic and Grok for arch tests #183

Changes from all commits
Commits
Show all changes
3 commits
Select commit Hold shift + click to select a range
25cd873
Fix gpt-oss accuracy issue
yangwang201911 May 22, 2026
5dd95ea
OpenVINO backend: enable arctic for arch test
zhaixuejun1993 May 24, 2026
6665562
OpenVINO backend: enable grok for arch test
zhaixuejun1993 May 25, 2026
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
56 changes: 50 additions & 6 deletions ggml/src/ggml-openvino/ggml-openvino.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -871,6 +871,20 @@ static bool is_op_unsupported_case(const ggml_tensor * op) {
// ERR = 0.000000197 > 0.000000100 GET_ROWS(type=q5_K,n=256,m=5,r=4,be1=1,be2=1,v=0)
return true;
}

// Keep the MoE routing weights gather on CPU for GPU runs. Splitting
// only at the later SUM/CLAMP/DIV nodes still leaves this routing path
// numerically unstable for arctic-style MoE graphs.
if (strncmp(op->name, "ffn_moe_weights", sizeof("ffn_moe_weights") - 1) == 0) {
return true;
}
break;
}
case GGML_OP_RESHAPE: {
if (strncmp(op->name, "ffn_moe_weights", sizeof("ffn_moe_weights") - 1) == 0 ||
strncmp(op->name, "ffn_norm_exps", sizeof("ffn_norm_exps") - 1) == 0) {
return true;
}
break;
}
case GGML_OP_ADD:
Expand Down Expand Up @@ -908,8 +922,7 @@ static bool is_op_unsupported_case(const ggml_tensor * op) {

// qwen3next MoE weight normalization is numerically sensitive on the GPU
// path. Keep the normalization divide on CPU to match the reference.
if (ggml_openvino_get_device_name() == "GPU" &&
strncmp(op->name, "ffn_moe_weights_norm", sizeof("ffn_moe_weights_norm") - 1) == 0) {
if (strncmp(op->name, "ffn_moe_weights_norm", sizeof("ffn_moe_weights_norm") - 1) == 0) {
return true;
}
break;
Expand All @@ -919,11 +932,23 @@ static bool is_op_unsupported_case(const ggml_tensor * op) {
// GGML_LOG_WARN("OpenVINO backend does not support SOFT_MAX with sinks\n");
return true;
}

if (strncmp(op->name, "ffn_moe_probs", sizeof("ffn_moe_probs") - 1) == 0) {
return true;
}

// GPU execution of the MoE routing weights softmax is numerically unstable
// when fused with the surrounding GET_ROWS/reshape path. Keep this softmax
// on CPU so the scheduler splits at the same boundary that restores parity.
if (op->src[0] != nullptr && op->src[0]->op == GGML_OP_RESHAPE &&
op->src[0]->src[0] != nullptr &&
strncmp(op->src[0]->src[0]->name, "ffn_moe_weights", sizeof("ffn_moe_weights") - 1) == 0) {
return true;
}
break;
}
case GGML_OP_SUM_ROWS: {
if (ggml_openvino_get_device_name() == "GPU" &&
strncmp(op->name, "ffn_moe_weights_sum", sizeof("ffn_moe_weights_sum") - 1) == 0) {
if (strncmp(op->name, "ffn_moe_weights_sum", sizeof("ffn_moe_weights_sum") - 1) == 0) {
return true;
}

Expand All @@ -934,13 +959,16 @@ static bool is_op_unsupported_case(const ggml_tensor * op) {
break;
}
case GGML_OP_CLAMP: {
if (ggml_openvino_get_device_name() == "GPU" &&
strncmp(op->name, "ffn_moe_weights_sum_clamped", sizeof("ffn_moe_weights_sum_clamped") - 1) == 0) {
if (strncmp(op->name, "ffn_moe_weights_sum_clamped", sizeof("ffn_moe_weights_sum_clamped") - 1) == 0) {
return true;
}
break;
}
case GGML_OP_FLASH_ATTN_EXT: {
// qwen3next currently shows large accuracy drift in OpenVINO flash attention.
// Keep FLASH_ATTN_EXT on CPU until parity is restored.
// return true;

if (op->src[4] != nullptr) {
// GGML_LOG_WARN("OpenVINO backend does not support FLASH_ATTN_EXT with sinks\n");
return true;
Expand Down Expand Up @@ -987,6 +1015,12 @@ static bool is_op_unsupported_case(const ggml_tensor * op) {
break;
}
case GGML_OP_MUL_MAT: {
if (ggml_openvino_get_device_name() == "GPU" && op->src[1]->op == GGML_OP_SOFT_MAX &&
op->src[0]->op == GGML_OP_CONT && op->src[0]->src[0] != nullptr &&
op->src[0]->src[0]->op == GGML_OP_TRANSPOSE && op->src[0]->src[0]->src[0] != nullptr &&
op->src[0]->src[0]->src[0]->op == GGML_OP_PERMUTE) {
return true;
}
if (op->src[0]->type == GGML_TYPE_F16 && op->src[1]->type == GGML_TYPE_F16) {
// Has accuracy issue, try enabling this and see `test-backend-ops -o "MUL_MAT"`
// GGML_LOG_WARN("OpenVINO backend does not support MUL_MAT with two F16 tensors\n");
Expand All @@ -1006,6 +1040,11 @@ static bool is_op_unsupported_case(const ggml_tensor * op) {
break;
}
case GGML_OP_MUL_MAT_ID: {
if (strncmp(op->name, "ffn_moe_gate_up", sizeof("ffn_moe_gate_up") - 1) == 0 ||
strncmp(op->name, "ffn_moe_down", sizeof("ffn_moe_down") - 1) == 0) {
return true;
}

if (mul_mat_id_requires_large_tmp(op)) {
return true;
}
Expand Down Expand Up @@ -1078,6 +1117,11 @@ static bool is_op_unsupported_case(const ggml_tensor * op) {
}
break;
}
case GGML_OP_SSM_CONV: {
// qwen3next is numerically unstable with OpenVINO SSM_CONV.
// Keep this op on CPU until the OpenVINO implementation is fixed.
return true;
}
default:
break;
}
Expand Down
Loading