Skip to content

whisper : mask padded K/V in flash-attention encoder and cross-attention #3941

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

Open
gooofy wants to merge 1 commit into
base: master
Choose a base branch
from
Open

whisper : mask padded K/V in flash-attention encoder and cross-attention #3941

Changes from all commits
Commits
Show all changes
1 commit
Select commit
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
72 changes: 70 additions & 2 deletions src/whisper.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -2068,6 +2068,18 @@ static struct ggml_cgraph * whisper_build_graph_encoder(

const float KQscale = 1.0f/sqrtf(float(n_state_head));

// with flash attention the self-attention K/V are viewed zero-padded to
// n_ctx_pad, but only n_ctx rows are written - the padding holds stale
// data from previous calls, so it must be masked out
struct ggml_tensor * KQ_mask_f16 = nullptr;
if (wctx.params.flash_attn) {
struct ggml_tensor * KQ_mask = ggml_new_tensor_3d(ctx0, GGML_TYPE_F32, n_ctx_pad, n_ctx, 1);
ggml_set_name(KQ_mask, "KQ_mask_enc");
ggml_set_input(KQ_mask);

KQ_mask_f16 = ggml_cast(ctx0, KQ_mask, GGML_TYPE_F16);
}

// ===================================================================
// NOTE: experimenting with partial evaluation of the encoder (ignore)
//static int iter = -1;
Expand Down Expand Up @@ -2156,7 +2168,7 @@ static struct ggml_cgraph * whisper_build_graph_encoder(
ggml_element_size(kv_pad.v)*n_state_head,
0);

cur = ggml_flash_attn_ext(ctx0, Q, K, V, nullptr, KQscale, 0.0f, 0.0f);
cur = ggml_flash_attn_ext(ctx0, Q, K, V, KQ_mask_f16, KQscale, 0.0f, 0.0f);

cur = ggml_reshape_2d(ctx0, cur, n_state, n_ctx);
} else {
Expand Down Expand Up @@ -2428,6 +2440,28 @@ static bool whisper_encode_internal(
return false;
}

// mask out the padded part of the flash-attention K/V views
{
struct ggml_tensor * KQ_mask = ggml_graph_get_tensor(gf, "KQ_mask_enc");
if (KQ_mask) {
const int n_ctx = wstate.exp_n_audio_ctx > 0 ? wstate.exp_n_audio_ctx : wctx.model.hparams.n_audio_ctx;
const int n_ctx_pad = GGML_PAD(n_ctx, 256);

wstate.inp_mask.resize(ggml_nelements(KQ_mask));

float * data = wstate.inp_mask.data();
memset(data, 0, ggml_nbytes(KQ_mask));

for (int j = 0; j < n_ctx; ++j) {
for (int i = n_ctx; i < n_ctx_pad; ++i) {
data[j*n_ctx_pad + i] = -INFINITY;
}
}

ggml_backend_tensor_set(KQ_mask, wstate.inp_mask.data(), 0, ggml_nelements(KQ_mask)*sizeof(float));
}
}

if (!ggml_graph_compute_helper(sched, gf, n_threads)) {
return false;
}
Expand Down Expand Up @@ -2511,6 +2545,18 @@ static struct ggml_cgraph * whisper_build_graph_decoder(

struct ggml_tensor * KQ_mask_f16 = ggml_cast(ctx0, KQ_mask, GGML_TYPE_F16);

// with flash attention the cross-attention K/V are viewed zero-padded to
// n_audio_ctx_pad, but only n_audio_ctx rows are written - the padding
// holds stale data from previous calls, so it must be masked out
struct ggml_tensor * KQ_mask_cross_f16 = nullptr;
if (wctx.params.flash_attn) {
struct ggml_tensor * KQ_mask_cross = ggml_new_tensor_3d(ctx0, GGML_TYPE_F32, n_audio_ctx_pad, n_tokens, 1);
ggml_set_name(KQ_mask_cross, "KQ_mask_cross");
ggml_set_input(KQ_mask_cross);

KQ_mask_cross_f16 = ggml_cast(ctx0, KQ_mask_cross, GGML_TYPE_F16);
}

// token encoding + position encoding
struct ggml_tensor * cur =
ggml_add(ctx0,
Expand Down Expand Up @@ -2692,7 +2738,7 @@ static struct ggml_cgraph * whisper_build_graph_decoder(
ggml_element_size(wstate.kv_cross.v)*n_state_head,
ggml_element_size(wstate.kv_cross.v)*n_state*n_audio_ctx_pad*il);

cur = ggml_flash_attn_ext(ctx0, Q, Kcross, Vcross, nullptr, KQscale, 0.0f, 0.0f);
cur = ggml_flash_attn_ext(ctx0, Q, Kcross, Vcross, KQ_mask_cross_f16, KQscale, 0.0f, 0.0f);

cur = ggml_reshape_2d(ctx0, cur, n_state, n_tokens);
} else {
Expand Down Expand Up @@ -2939,6 +2985,28 @@ static bool whisper_decode_internal(
ggml_backend_tensor_set(KQ_mask, wstate.inp_mask.data(), 0, ggml_nelements(KQ_mask)*sizeof(float));
}

// mask out the padded part of the flash-attention cross K/V views
{
struct ggml_tensor * KQ_mask_cross = ggml_graph_get_tensor(gf, "KQ_mask_cross");
if (KQ_mask_cross) {
const int n_audio_ctx = wstate.exp_n_audio_ctx > 0 ? wstate.exp_n_audio_ctx : wctx.model.hparams.n_audio_ctx;
const int n_audio_ctx_pad = GGML_PAD(n_audio_ctx, 256);

wstate.inp_mask.resize(ggml_nelements(KQ_mask_cross));

float * data = wstate.inp_mask.data();
memset(data, 0, ggml_nbytes(KQ_mask_cross));

for (int j = 0; j < n_tokens; ++j) {
for (int i = n_audio_ctx; i < n_audio_ctx_pad; ++i) {
data[j*n_audio_ctx_pad + i] = -INFINITY;
}
}

ggml_backend_tensor_set(KQ_mask_cross, wstate.inp_mask.data(), 0, ggml_nelements(KQ_mask_cross)*sizeof(float));
}
}

logits = ggml_graph_node(gf, -1);

if (!ggml_graph_compute_helper(sched, gf, n_threads)) {
Expand Down