Skip to content

llama : switch KQ multiplication to use F32 precision by default #10015

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
merged 1 commit into from
Oct 27, 2024
Merged

llama : switch KQ multiplication to use F32 precision by default #10015

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
15 changes: 4 additions & 11 deletions src/llama.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -9618,20 +9618,16 @@ static struct ggml_tensor * llm_build_kqv(
cur = ggml_flash_attn_ext(ctx, q, k, v, kq_mask, kq_scale, hparams.f_max_alibi_bias,
hparams.attn_soft_cap ? hparams.f_attn_logit_softcapping : 0.0f);

if (model.arch == LLM_ARCH_PHI2 || model.arch == LLM_ARCH_PHI3 || model.arch == LLM_ARCH_GPTNEOX || model.arch == LLM_ARCH_GEMMA2) {
ggml_flash_attn_ext_set_prec(cur, GGML_PREC_F32);
}
ggml_flash_attn_ext_set_prec(cur, GGML_PREC_F32);

cur = ggml_reshape_2d(ctx, cur, n_embd_head_v*n_head, n_tokens);
} else {
struct ggml_tensor * kq = ggml_mul_mat(ctx, k, q);
cb(kq, "kq", il);

if (model.arch == LLM_ARCH_PHI2 || model.arch == LLM_ARCH_PHI3 || model.arch == LLM_ARCH_GPTNEOX || model.arch == LLM_ARCH_QWEN2 || model.arch == LLM_ARCH_NEMOTRON || model.arch == LLM_ARCH_CHATGLM) {
// for this arch, we need to perform the KQ multiplication with F32 precision, otherwise we get NaNs
// ref: https://github.com/ggerganov/llama.cpp/pull/4490#issuecomment-1859055847
ggml_mul_mat_set_prec(kq, GGML_PREC_F32);
}
// note: this op tends to require high floating point range
// while for some models F16 is enough, for others it is not, so we default to F32 here
ggml_mul_mat_set_prec(kq, GGML_PREC_F32);

if (model.arch == LLM_ARCH_GROK) {
// need to do the following:
Expand All @@ -9640,9 +9636,6 @@ static struct ggml_tensor * llm_build_kqv(
// kq = 30 * tanh(kq / 30)
// before the softmax below

//try from phi2
//ggml_mul_mat_set_prec(kq, GGML_PREC_F32);

kq = ggml_tanh(ctx, ggml_scale(ctx, kq, 0.08838834764831845f/30.0f));
kq = ggml_scale(ctx, kq, 30);
}
Expand Down
Loading