Skip to content

model : apply YaRN truncation config to PLaMo-3 models #30337

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

Draft
tokinasin wants to merge 1 commit into
base: master
Choose a base branch
from
+15 −1
Draft

model : apply YaRN truncation config to PLaMo-3 models #30337

Show file tree
Hide file tree
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
1 change: 1 addition & 0 deletions src/llama-arch.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -340,6 +340,7 @@ static const std::map<llm_kv, const char *> LLM_KV_NAMES = {
{ LLM_KV_ROPE_SCALING_ATTN_FACTOR, "%s.rope.scaling.attn_factor" },
{ LLM_KV_ROPE_SCALING_ORIG_CTX_LEN, "%s.rope.scaling.original_context_length" },
{ LLM_KV_ROPE_SCALING_FINETUNED, "%s.rope.scaling.finetuned" },
{ LLM_KV_ROPE_SCALING_YARN_TRUNCATE, "%s.rope.scaling.yarn_truncate" },
{ LLM_KV_ROPE_SCALING_YARN_LOG_MUL, "%s.rope.scaling.yarn_log_multiplier" },
{ LLM_KV_ROPE_SCALING_YARN_EXT_FACTOR, "%s.rope.scaling.yarn_ext_factor" },
{ LLM_KV_ROPE_SCALING_YARN_ATTN_FACTOR, "%s.rope.scaling.yarn_attn_factor" },
Expand Down
1 change: 1 addition & 0 deletions src/llama-arch.h
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -345,6 +345,7 @@ enum llm_kv {
LLM_KV_ROPE_SCALING_ATTN_FACTOR,
LLM_KV_ROPE_SCALING_ORIG_CTX_LEN,
LLM_KV_ROPE_SCALING_FINETUNED,
LLM_KV_ROPE_SCALING_YARN_TRUNCATE,
LLM_KV_ROPE_SCALING_YARN_LOG_MUL,
LLM_KV_ROPE_SCALING_YARN_EXT_FACTOR,
LLM_KV_ROPE_SCALING_YARN_ATTN_FACTOR,
Expand Down
1 change: 1 addition & 0 deletions src/llama-hparams.h
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -63,6 +63,7 @@ struct llama_hparams {
bool vocab_only;
bool no_alloc;
bool rope_finetuned;
bool yarn_truncate = true;
bool use_par_res;
bool swin_norm;
bool norm_before_residual = false;
Expand Down
9 changes: 8 additions & 1 deletion src/llama-kv-cache.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -1974,16 +1974,23 @@ ggml_tensor * llama_kv_cache::build_rope_shift(
tmp = ggml_rope_ext(ctx, tmp,
shift, factors, n_rot, rope_type, n_ctx_orig, freq_base, freq_scale,
yarn_ext_factor, yarn_attn_factor, yarn_beta_fast, yarn_beta_slow);
ggml_rope_set_truncate(tmp, hparams.yarn_truncate);

// rotate fwd
tmp = llama_mul_mat_hadamard(ctx, tmp, rot);

tmp = ggml_cpy(ctx, tmp, cur);
} else {
} else if (hparams.yarn_truncate) {
// we rotate only the first n_rot dimensions
tmp = ggml_rope_ext_inplace(ctx, cur,
shift, factors, n_rot, rope_type, n_ctx_orig, freq_base, freq_scale,
yarn_ext_factor, yarn_attn_factor, yarn_beta_fast, yarn_beta_slow);
} else {
tmp = ggml_rope_ext(ctx, cur,
shift, factors, n_rot, rope_type, n_ctx_orig, freq_base, freq_scale,
yarn_ext_factor, yarn_attn_factor, yarn_beta_fast, yarn_beta_slow);
ggml_rope_set_truncate(tmp, false);
tmp = ggml_cpy(ctx, tmp, cur);
}

return tmp;
Expand Down
1 change: 1 addition & 0 deletions src/llama-model-saver.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -374,6 +374,7 @@ void llama_model_saver::add_kv_from_model() {
add_kv(LLM_KV_ROPE_SCALING_ATTN_FACTOR, hparams.rope_attn_factor);
add_kv(LLM_KV_ROPE_SCALING_ORIG_CTX_LEN, hparams.n_ctx_orig_yarn);
add_kv(LLM_KV_ROPE_SCALING_FINETUNED, hparams.rope_finetuned);
add_kv(LLM_KV_ROPE_SCALING_YARN_TRUNCATE, hparams.yarn_truncate);
add_kv(LLM_KV_ROPE_SCALING_YARN_LOG_MUL, hparams.rope_yarn_log_mul);
add_kv(LLM_KV_ROPE_SCALING_YARN_EXT_FACTOR, hparams.yarn_ext_factor);
add_kv(LLM_KV_ROPE_SCALING_YARN_ATTN_FACTOR, hparams.yarn_attn_factor);
Expand Down
3 changes: 3 additions & 0 deletions src/models/plamo3.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -2,6 +2,7 @@

void llama_model_plamo3::load_arch_hparams(llama_model_loader & ml) {
ml.get_key(LLM_KV_ATTENTION_LAYERNORM_RMS_EPS, hparams.f_norm_rms_eps);
ml.get_key(LLM_KV_ROPE_SCALING_YARN_TRUNCATE, hparams.yarn_truncate, false);
const bool found_swa = ml.get_key(LLM_KV_ATTENTION_SLIDING_WINDOW, hparams.n_swa, false);
if (found_swa && hparams.n_swa > 0) {
hparams.swa_type = LLAMA_SWA_TYPE_STANDARD;
Expand Down Expand Up @@ -131,9 +132,11 @@ llama_model_plamo3::graph<iswa>::graph(const llama_model & model, const llm_grap
Qcur = ggml_rope_ext(ctx0, Qcur, inp_pos, nullptr,
n_rot, rope_type, n_ctx_orig, freq_base_l, freq_scale_l,
ext_factor, attn_factor, beta_fast, beta_slow);
ggml_rope_set_truncate(Qcur, hparams.yarn_truncate);
Kcur = ggml_rope_ext(ctx0, Kcur, inp_pos, nullptr,
n_rot, rope_type, n_ctx_orig, freq_base_l, freq_scale_l,
ext_factor, attn_factor, beta_fast, beta_slow);
ggml_rope_set_truncate(Kcur, hparams.yarn_truncate);

const float attn_scale = 1.0f / sqrtf(float(head_dim_q));

Expand Down