tqcq/llama.cpp
0
0
Fork 0
mirror of https://github.com/ggml-org/llama.cpp.git synced 2025-06-27 12:05:03 +00:00
Code Issues Packages Projects Releases Wiki Activity

hparams : add SWA rope parameters

Browse Source 
ggml-ci
This commit is contained in:
Georgi Gerganov
2025-03-13 19:26:09 +02:00
parent 84d5475541
commit c4aca65582
5 changed files with 26 additions and 20 deletions
Download Patch File Download Diff File Expand all files Collapse all files

14
src/llama-context.cpp
View File

@ -537,16 +537,12 @@ llm_graph_result_ptr llama_context::build_kv_self_shift(
const int64_t n_head_kv = hparams.n_head_kv(il); const int64_t n_head_kv = hparams.n_head_kv(il);
const int64_t n_embd_k_gqa = hparams.n_embd_k_gqa(il); const int64_t n_embd_k_gqa = hparams.n_embd_k_gqa(il);
float freq_base_l = cparams.rope_freq_base; const bool is_swa = hparams.is_swa(il);
float freq_scale_l = cparams.rope_freq_scale;
// TODO: improve // note: the swa rope params could become part of the cparams in the future
if (model.arch == LLM_ARCH_GEMMA3) { // if we decide to make them configurable, like the non-sliding ones
const bool is_sliding = hparams.is_sliding(il); const float freq_base_l = is_swa ? hparams.rope_freq_base_train_swa : cparams.rope_freq_base;
const float freq_scale_l = is_swa ? hparams.rope_freq_scale_train_swa : cparams.rope_freq_scale;
freq_base_l = is_sliding ? 10000.0f : cparams.rope_freq_base;
freq_scale_l = is_sliding ? 1.0f : cparams.rope_freq_scale;
}
ggml_tensor * rope_factors = kv_self->cbs.get_rope_factors(n_ctx_per_seq(), il); ggml_tensor * rope_factors = kv_self->cbs.get_rope_factors(n_ctx_per_seq(), il);

4
src/llama-graph.cpp
View File

@ -1403,9 +1403,9 @@ ggml_tensor * llm_graph_context::build_attn(
ggml_build_forward_expand(gf, ggml_cpy(ctx0, v_cur, v_cache_view)); ggml_build_forward_expand(gf, ggml_cpy(ctx0, v_cur, v_cache_view));
} }
const bool is_sliding = hparams.is_sliding(il); const bool is_swa = hparams.is_swa(il);
const auto & kq_mask = is_sliding ? inp->get_kq_mask_swa() : inp->get_kq_mask(); const auto & kq_mask = is_swa ? inp->get_kq_mask_swa() : inp->get_kq_mask();
const auto n_kv = kv_self->n; const auto n_kv = kv_self->n;

2
src/llama-hparams.cpp
View File

@ -70,7 +70,7 @@ uint32_t llama_hparams::n_embd_v_s() const {
return ssm_d_state * ssm_d_inner; return ssm_d_state * ssm_d_inner;
} }
bool llama_hparams::is_sliding(uint32_t il) const { bool llama_hparams::is_swa(uint32_t il) const {
if (il < n_layer) { if (il < n_layer) {
return n_swa > 0 && n_swa_pattern > 0 && il % n_swa_pattern < (n_swa_pattern - 1); return n_swa > 0 && n_swa_pattern > 0 && il % n_swa_pattern < (n_swa_pattern - 1);
} }

4
src/llama-hparams.h
View File

@ -79,7 +79,9 @@ struct llama_hparams {
float rope_attn_factor = 1.0f; float rope_attn_factor = 1.0f;
float rope_freq_base_train; float rope_freq_base_train;
float rope_freq_base_train_swa;
float rope_freq_scale_train; float rope_freq_scale_train;
float rope_freq_scale_train_swa;
uint32_t n_ctx_orig_yarn; uint32_t n_ctx_orig_yarn;
float rope_yarn_log_mul; float rope_yarn_log_mul;
@ -135,7 +137,7 @@ struct llama_hparams {
// dimension of the recurrent state embeddings // dimension of the recurrent state embeddings
uint32_t n_embd_v_s() const; uint32_t n_embd_v_s() const;
bool is_sliding(uint32_t il) const; bool is_swa(uint32_t il) const;
}; };
static_assert(std::is_trivially_copyable<llama_hparams>::value, "llama_hparams must be trivially copyable"); static_assert(std::is_trivially_copyable<llama_hparams>::value, "llama_hparams must be trivially copyable");

22
src/llama-model.cpp
View File

@ -475,6 +475,10 @@ void llama_model::load_hparams(llama_model_loader & ml) {
} }
hparams.rope_freq_scale_train = ropescale == 0.0f ? 1.0f : 1.0f/ropescale; hparams.rope_freq_scale_train = ropescale == 0.0f ? 1.0f : 1.0f/ropescale;
// by default assume that the sliding-window layers use the same scaling type as the non-sliding-window layers
hparams.rope_freq_base_train_swa = hparams.rope_freq_base_train;
hparams.rope_freq_scale_train_swa = hparams.rope_freq_scale_train;
ml.get_key(LLM_KV_ROPE_SCALING_ATTN_FACTOR, hparams.rope_attn_factor, false); ml.get_key(LLM_KV_ROPE_SCALING_ATTN_FACTOR, hparams.rope_attn_factor, false);
// non-transformer models do not have attention heads // non-transformer models do not have attention heads
@ -877,6 +881,9 @@ void llama_model::load_hparams(llama_model_loader & ml) {
{ {
hparams.n_swa_pattern = 6; hparams.n_swa_pattern = 6;
hparams.rope_freq_base_train_swa = 10000.0f;
hparams.rope_freq_scale_train_swa = 1.0f;
ml.get_key(LLM_KV_ATTENTION_SLIDING_WINDOW, hparams.n_swa); ml.get_key(LLM_KV_ATTENTION_SLIDING_WINDOW, hparams.n_swa);
ml.get_key(LLM_KV_ATTENTION_LAYERNORM_RMS_EPS, hparams.f_norm_rms_eps); ml.get_key(LLM_KV_ATTENTION_LAYERNORM_RMS_EPS, hparams.f_norm_rms_eps);
@ -1346,13 +1353,14 @@ bool llama_model::load_tensors(llama_model_loader & ml) {
const int i_gpu_start = std::max((int) hparams.n_layer - n_gpu_layers, (int) 0); const int i_gpu_start = std::max((int) hparams.n_layer - n_gpu_layers, (int) 0);
const int act_gpu_layers = devices.empty() ? 0 : std::min(n_gpu_layers, (int)n_layer + 1); const int act_gpu_layers = devices.empty() ? 0 : std::min(n_gpu_layers, (int)n_layer + 1);
auto get_layer_buft_list = [&](int il) -> llama_model::impl::layer_dev { auto get_layer_buft_list = [&](int il) -> llama_model::impl::layer_dev {
const bool is_swa = il < (int) hparams.n_layer && hparams.is_swa(il);
if (il < i_gpu_start || (il - i_gpu_start) >= act_gpu_layers) { if (il < i_gpu_start || (il - i_gpu_start) >= act_gpu_layers) {
LLAMA_LOG_DEBUG("load_tensors: layer %3d assigned to device %s\n", il, ggml_backend_dev_name(cpu_dev)); LLAMA_LOG_DEBUG("load_tensors: layer %3d assigned to device %s, is_swa = %d\n", il, ggml_backend_dev_name(cpu_dev), is_swa);
return {cpu_dev, &pimpl->cpu_buft_list}; return {cpu_dev, &pimpl->cpu_buft_list};
} }
const int layer_gpu = std::upper_bound(splits.begin(), splits.begin() + n_devices(), float(il - i_gpu_start)/act_gpu_layers) - splits.begin(); const int layer_gpu = std::upper_bound(splits.begin(), splits.begin() + n_devices(), float(il - i_gpu_start)/act_gpu_layers) - splits.begin();
auto * dev = devices.at(layer_gpu); auto * dev = devices.at(layer_gpu);
LLAMA_LOG_DEBUG("load_tensors: layer %3d assigned to device %s\n", il, ggml_backend_dev_name(dev)); LLAMA_LOG_DEBUG("load_tensors: layer %3d assigned to device %s, is_swa = %d\n", il, ggml_backend_dev_name(dev), is_swa);
return {dev, &pimpl->gpu_buft_list.at(dev)}; return {dev, &pimpl->gpu_buft_list.at(dev)};
}; };
@ -7381,10 +7389,10 @@ struct llm_build_gemma3 : public llm_graph_context {
auto * inp_attn = build_attn_inp_kv_unified(true, true); auto * inp_attn = build_attn_inp_kv_unified(true, true);
for (int il = 0; il < n_layer; ++il) { for (int il = 0; il < n_layer; ++il) {
const bool is_sliding = hparams.is_sliding(il); const bool is_swa = hparams.is_swa(il);
const float freq_base_l = is_sliding ? 10000.0f : freq_base; const float freq_base_l = is_swa ? hparams.rope_freq_base_train_swa : cparams.rope_freq_base;
const float freq_scale_l = is_sliding ? 1.0f : freq_scale; const float freq_scale_l = is_swa ? hparams.rope_freq_scale_train_swa : cparams.rope_freq_scale;
// norm // norm
cur = build_norm(inpL, model.layers[il].attn_norm, NULL, LLM_NORM_RMS, il); cur = build_norm(inpL, model.layers[il].attn_norm, NULL, LLM_NORM_RMS, il);
@ -7973,7 +7981,7 @@ struct llm_build_cohere2 : public llm_graph_context {
auto * inp_attn = build_attn_inp_kv_unified(true, true); auto * inp_attn = build_attn_inp_kv_unified(true, true);
for (int il = 0; il < n_layer; ++il) { for (int il = 0; il < n_layer; ++il) {
const bool is_sliding = hparams.is_sliding(il); const bool is_swa = hparams.is_swa(il);
// norm // norm
cur = build_norm(inpL, model.layers[il].attn_norm, NULL, LLM_NORM, il); cur = build_norm(inpL, model.layers[il].attn_norm, NULL, LLM_NORM, il);
@ -8007,7 +8015,7 @@ struct llm_build_cohere2 : public llm_graph_context {
cb(Vcur, "Vcur", il); cb(Vcur, "Vcur", il);
} }
if (is_sliding) { if (is_swa) {
Qcur = ggml_rope_ext(ctx0, ggml_reshape_3d(ctx0, Qcur, n_embd_head, n_head, n_tokens), inp_pos, rope_factors, Qcur = ggml_rope_ext(ctx0, ggml_reshape_3d(ctx0, Qcur, n_embd_head, n_head, n_tokens), inp_pos, rope_factors,
n_rot, rope_type, n_ctx_orig, freq_base, freq_scale, ext_factor, attn_factor, n_rot, rope_type, n_ctx_orig, freq_base, freq_scale, ext_factor, attn_factor,
beta_fast, beta_slow); beta_fast, beta_slow);