Skip to content

ggml : remove GGML_KQ_MASK_PAD constant #17910

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
ggerganov merged 2 commits into from
Dec 10, 2025
Merged

ggml : remove GGML_KQ_MASK_PAD constant #17910

Show file tree
Hide file tree
Changes from 1 commit
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
2b17c65
ggml : remove GGML_KQ_MASK_PAD constant
ggerganov Dec 10, 2025
0bfa070
cont : remove comment
ggerganov Dec 10, 2025
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
2 changes: 0 additions & 2 deletions ggml/include/ggml.h
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -2305,8 +2305,6 @@ extern "C" {
float stop,
float step);

#define GGML_KQ_MASK_PAD 1

// q: [n_embd_k, n_batch, n_head, ne3 ]
// k: [n_embd_k, n_kv, n_head_kv, ne3 ]
// v: [n_embd_v, n_kv, n_head_kv, ne3 ] !! not transposed !!

@jeffbolznv jeffbolznv Dec 10, 2025

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.

There's a use of GGML_KQ_MASK_PAD in the comment on the next line.

Expand Down
2 changes: 0 additions & 2 deletions ggml/src/ggml.c
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -5260,8 +5260,6 @@ struct ggml_tensor * ggml_flash_attn_ext(

if (mask) {
GGML_ASSERT(ggml_is_contiguous(mask));
GGML_ASSERT(mask->ne[1] >= GGML_PAD(q->ne[1], GGML_KQ_MASK_PAD) &&
"the Flash-Attention kernel requires the mask to be padded to GGML_KQ_MASK_PAD and at least n_queries big");
//GGML_ASSERT(ggml_can_repeat_rows(mask, qk));

GGML_ASSERT(q->ne[2] % mask->ne[2] == 0);
Expand Down
8 changes: 0 additions & 8 deletions src/llama-context.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -93,14 +93,6 @@ llama_context::llama_context(
// with causal attention, the batch size is limited by the context size
cparams.n_batch = cparams.causal_attn ? std::min(cparams.n_ctx, params.n_batch) : params.n_batch;

// the batch has to be at least GGML_KQ_MASK_PAD because we will be padding the KQ_mask
// this is required by GPU kernels in order to avoid out-of-bounds accesses (e.g. ggml_flash_attn_ext)
// ref: https://github.com/ggerganov/llama.cpp/pull/5021
// TODO: this padding is not needed for the cache-less context so we should probably move it to llama_memory
if (cparams.n_batch < GGML_KQ_MASK_PAD) {
LLAMA_LOG_WARN("%s: n_batch is less than GGML_KQ_MASK_PAD - increasing to %d\n", __func__, GGML_KQ_MASK_PAD);
cparams.n_batch = GGML_KQ_MASK_PAD;
}
cparams.n_ubatch = std::min(cparams.n_batch, params.n_ubatch == 0 ? params.n_batch : params.n_ubatch);

cparams.op_offload = params.op_offload;
Expand Down
20 changes: 10 additions & 10 deletions src/llama-graph.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -385,7 +385,7 @@ bool llm_graph_input_attn_kv::can_reuse(const llm_graph_params & params) {
//res &= self_v_idxs->ne[0] == params.ubatch.n_tokens; // TODO: need to move this to the unified cache and check there

res &= self_kq_mask->ne[0] == mctx->get_n_kv();
res &= self_kq_mask->ne[1] == GGML_PAD(params.ubatch.n_tokens, GGML_KQ_MASK_PAD);
res &= self_kq_mask->ne[1] == params.ubatch.n_tokens;

return res;
}
Expand Down Expand Up @@ -416,10 +416,10 @@ bool llm_graph_input_attn_kv_iswa::can_reuse(const llm_graph_params & params) {
//res &= self_v_idxs_swa->ne[0] == params.ubatch.n_tokens; // TODO: need to move this to the unified cache and check there

res &= self_kq_mask->ne[0] == mctx->get_base()->get_n_kv();
res &= self_kq_mask->ne[1] == GGML_PAD(params.ubatch.n_tokens, GGML_KQ_MASK_PAD);
res &= self_kq_mask->ne[1] == params.ubatch.n_tokens;

res &= self_kq_mask_swa->ne[0] == mctx->get_swa()->get_n_kv();
res &= self_kq_mask_swa->ne[1] == GGML_PAD(params.ubatch.n_tokens, GGML_KQ_MASK_PAD);
res &= self_kq_mask_swa->ne[1] == params.ubatch.n_tokens;

return res;
}
Expand Down Expand Up @@ -452,7 +452,7 @@ void llm_graph_input_attn_cross::set_input(const llama_ubatch * ubatch) {
}
}

for (int i = n_tokens; i < GGML_PAD(n_tokens, GGML_KQ_MASK_PAD); ++i) {
for (int i = n_tokens; i < n_tokens; ++i) {

@gabe-l-hart gabe-l-hart Jan 12, 2026

Copy link
Copy Markdown
Collaborator

Choose a reason for hiding this comment

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

@ggerganov Totally random, but while looking at some of the ModernBert stuff, I (and @hansolosan) came across this line which now seems to do nothing with the removal of the padding. Should this whole set of nested loops be removed?

@gabe-l-hart gabe-l-hart Jan 12, 2026

Copy link
Copy Markdown
Collaborator

Choose a reason for hiding this comment

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

cc @ryan-mangeno in case this has anything to do with your ongoing Modern Bert divergence issues

@gabe-l-hart gabe-l-hart Jan 12, 2026

Copy link
Copy Markdown
Collaborator

Choose a reason for hiding this comment

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

I don't think it related to Modern Bert after all. From what I can tell, this (as well as build_inp_pos_bucket_enc) are only used in t5-dec.cpp and t5-enc.cpp respectively. Both of these input builders have similar properties (outer h loops that only loop once for example).

@gabe-l-hart gabe-l-hart Jan 12, 2026

Copy link
Copy Markdown
Collaborator

Choose a reason for hiding this comment

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

I don't think this is actually causing any problems, but I made a PR to clean it up for future readers: #18795

@ryan-mangeno ryan-mangeno Jan 12, 2026

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.

I don't think it related to Modern Bert after all. From what I can tell, this (as well as build_inp_pos_bucket_enc) are only used in t5-dec.cpp and t5-enc.cpp respectively. Both of these input builders have similar properties (outer h loops that only loop once for example).

alright, thanks for letting me know though @gabe-l-hart ! I just got back to school but I have been doing quite a bit of debugging between transformers and llama.cpp in terms of finding where they diverge, if its useful ... I have been inspecting the pooling process much closer and using the llama-eval-callback tool, but there is no pooling flag to be passed, would it make sense to add support for this / expose it. In my case I can hardcode it in llama-embedding.cpp and view the tensors made during build_pooling.

for (int j = 0; j < n_enc; ++j) {
data[h*(n_enc*n_tokens) + i*n_enc + j] = -INFINITY;
}
Expand Down Expand Up @@ -1470,13 +1470,13 @@ llm_graph_input_attn_no_cache * llm_graph_context::build_attn_inp_no_cache() con
auto inp = std::make_unique<llm_graph_input_attn_no_cache>(hparams, cparams);

// note: there is no KV cache, so the number of KV values is equal to the number of tokens in the batch
inp->self_kq_mask = ggml_new_tensor_4d(ctx0, GGML_TYPE_F32, n_tokens, GGML_PAD(n_tokens, GGML_KQ_MASK_PAD), 1, 1);
inp->self_kq_mask = ggml_new_tensor_4d(ctx0, GGML_TYPE_F32, n_tokens, n_tokens, 1, 1);
ggml_set_input(inp->self_kq_mask);

inp->self_kq_mask_cnv = cparams.flash_attn ? ggml_cast(ctx0, inp->self_kq_mask, GGML_TYPE_F16) : inp->self_kq_mask;

if (hparams.swa_type != LLAMA_SWA_TYPE_NONE) {
inp->self_kq_mask_swa = ggml_new_tensor_4d(ctx0, GGML_TYPE_F32, n_tokens, GGML_PAD(n_tokens, GGML_KQ_MASK_PAD), 1, 1);
inp->self_kq_mask_swa = ggml_new_tensor_4d(ctx0, GGML_TYPE_F32, n_tokens, n_tokens, 1, 1);
ggml_set_input(inp->self_kq_mask_swa);

inp->self_kq_mask_swa_cnv = cparams.flash_attn ? ggml_cast(ctx0, inp->self_kq_mask_swa, GGML_TYPE_F16) : inp->self_kq_mask_swa;
Expand Down Expand Up @@ -1558,7 +1558,7 @@ static std::unique_ptr<llm_graph_input_attn_kv> build_attn_inp_kv_impl(
inp->self_k_idxs = mctx_cur->build_input_k_idxs(ctx0, ubatch);
inp->self_v_idxs = mctx_cur->build_input_v_idxs(ctx0, ubatch);

inp->self_kq_mask = ggml_new_tensor_4d(ctx0, GGML_TYPE_F32, n_kv, GGML_PAD(n_tokens/n_stream, GGML_KQ_MASK_PAD), 1, n_stream);
inp->self_kq_mask = ggml_new_tensor_4d(ctx0, GGML_TYPE_F32, n_kv, n_tokens/n_stream, 1, n_stream);
ggml_set_input(inp->self_kq_mask);

inp->self_kq_mask_cnv = cparams.flash_attn ? ggml_cast(ctx0, inp->self_kq_mask, GGML_TYPE_F16) : inp->self_kq_mask;
Expand Down Expand Up @@ -1701,7 +1701,7 @@ llm_graph_input_attn_cross * llm_graph_context::build_attn_inp_cross() const {

const int32_t n_enc = !cross->v_embd.empty() ? cross->n_enc : hparams.n_ctx_train;

inp->cross_kq_mask = ggml_new_tensor_4d(ctx0, GGML_TYPE_F32, n_enc, GGML_PAD(n_tokens, GGML_KQ_MASK_PAD), 1, 1);
inp->cross_kq_mask = ggml_new_tensor_4d(ctx0, GGML_TYPE_F32, n_enc, n_tokens, 1, 1);
ggml_set_input(inp->cross_kq_mask);

inp->cross_kq_mask_cnv = cparams.flash_attn ? ggml_cast(ctx0, inp->cross_kq_mask, GGML_TYPE_F16) : inp->cross_kq_mask;
Expand Down Expand Up @@ -1767,7 +1767,7 @@ llm_graph_input_attn_kv_iswa * llm_graph_context::build_attn_inp_kv_iswa() const
inp->self_k_idxs = mctx_cur->get_base()->build_input_k_idxs(ctx0, ubatch);
inp->self_v_idxs = mctx_cur->get_base()->build_input_v_idxs(ctx0, ubatch);

inp->self_kq_mask = ggml_new_tensor_4d(ctx0, GGML_TYPE_F32, n_kv, GGML_PAD(n_tokens/n_stream, GGML_KQ_MASK_PAD), 1, n_stream);
inp->self_kq_mask = ggml_new_tensor_4d(ctx0, GGML_TYPE_F32, n_kv, n_tokens/n_stream, 1, n_stream);
ggml_set_input(inp->self_kq_mask);

inp->self_kq_mask_cnv = cparams.flash_attn ? ggml_cast(ctx0, inp->self_kq_mask, GGML_TYPE_F16) : inp->self_kq_mask;
Expand All @@ -1781,7 +1781,7 @@ llm_graph_input_attn_kv_iswa * llm_graph_context::build_attn_inp_kv_iswa() const
inp->self_k_idxs_swa = mctx_cur->get_swa()->build_input_k_idxs(ctx0, ubatch);
inp->self_v_idxs_swa = mctx_cur->get_swa()->build_input_v_idxs(ctx0, ubatch);

inp->self_kq_mask_swa = ggml_new_tensor_4d(ctx0, GGML_TYPE_F32, n_kv, GGML_PAD(n_tokens/n_stream, GGML_KQ_MASK_PAD), 1, n_stream);
inp->self_kq_mask_swa = ggml_new_tensor_4d(ctx0, GGML_TYPE_F32, n_kv, n_tokens/n_stream, 1, n_stream);
ggml_set_input(inp->self_kq_mask_swa);

inp->self_kq_mask_swa_cnv = cparams.flash_attn ? ggml_cast(ctx0, inp->self_kq_mask_swa, GGML_TYPE_F16) : inp->self_kq_mask_swa;
Expand Down
5 changes: 2 additions & 3 deletions src/llama-kv-cache.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -1232,8 +1232,7 @@ void llama_kv_cache::set_input_kq_mask(ggml_tensor * dst, const llama_ubatch * u
GGML_ASSERT(n_tokens%n_stream == 0);

// n_tps == n_tokens_per_stream
const int64_t n_tps = n_tokens/n_stream;
const int64_t n_tps_pad = GGML_PAD(n_tps, GGML_KQ_MASK_PAD);
const int64_t n_tps = n_tokens/n_stream;

std::fill(data, data + ggml_nelements(dst), -INFINITY);

Expand Down Expand Up @@ -1266,7 +1265,7 @@ void llama_kv_cache::set_input_kq_mask(ggml_tensor * dst, const llama_ubatch * u
const llama_pos p1_x = is_2d ? ubatch->pos[i + ubatch->n_tokens*2] : 0;
const llama_pos p1_y = is_2d ? ubatch->pos[i + ubatch->n_tokens] : 0;

const uint64_t idst = n_kv*(h*n_stream*n_tps_pad + s*n_tps_pad + ii);
const uint64_t idst = n_kv*(h*n_stream*n_tps + s*n_tps + ii);

for (uint32_t j = 0; j < n_kv; ++j) {
if (cells.is_empty(j)) {
Expand Down
2 changes: 1 addition & 1 deletion tests/test-backend-ops.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -5875,7 +5875,7 @@ struct test_flash_attn_ext : public test_case {

ggml_tensor * m = nullptr;
if (mask) {
m = ggml_new_tensor_4d(ctx, GGML_TYPE_F16, kv, GGML_PAD(nb, GGML_KQ_MASK_PAD), 1, nr23[1]);
m = ggml_new_tensor_4d(ctx, GGML_TYPE_F16, kv, nb, 1, nr23[1]);
ggml_set_name(m, "m");
}

Expand Down
6 changes: 1 addition & 5 deletions tools/mtmd/clip.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -775,10 +775,6 @@ struct clip_graph {

// if flash attn is used, we need to pad the mask and cast to f16
if (ctx->flash_attn_type == CLIP_FLASH_ATTN_TYPE_ENABLED) {
int n_pad = GGML_PAD(window_mask->ne[1], GGML_KQ_MASK_PAD) - window_mask->ne[1];
if (n_pad > 0) {
window_mask = ggml_pad(ctx0, window_mask, 0, n_pad, 0, 0);
}
window_mask = ggml_cast(ctx0, window_mask, GGML_TYPE_F16);
}

Expand All @@ -791,7 +787,7 @@ struct clip_graph {

// loop over layers
for (int il = 0; il < n_layer; il++) {
auto & layer = model.layers[il];
const auto & layer = model.layers[il];
const bool full_attn = use_window_attn ? (il + 1) % n_wa_pattern == 0 : true;

ggml_tensor * cur = inpL; // inpL = residual, cur = hidden_states
Expand Down
Loading