starred/ollama-ollama
1
0
Fork 0
mirror of https://github.com/ollama/ollama.git synced 2026-04-26 07:45:25 +02:00
Code Issues Packages Projects Releases Wiki Activity

feat: llama.cpp bump (17f7f4) for SSM performance improvements (#13408)

Browse Source 
* feat: Bump llama.cpp to the latest master (17f7f4b)

This brings in significant improvements to prefill performance for all
models using the SSM_CONV and SSM_SCAN ops (granite4, jamba, falcon-h,
nemotron-h, Qwen3 Next) on Apple Metal.

See https://github.com/ggml-org/llama.cpp/pull/17876

Branch: LlamaCPPMetalSSMImprovements

Signed-off-by: Gabe Goodhart <ghart@us.ibm.com>

* feat: Update patches 1-4

Branch: LlamaCPPMetalSSMImprovements

Signed-off-by: Gabe Goodhart <ghart@us.ibm.com>

* fix: Update patches 5-12

Branch: LlamaCPPMetalSSMImprovements

Signed-off-by: Gabe Goodhart <ghart@us.ibm.com>

* feat: Update patches 13-18

Branch: LlamaCPPMetalSSMImprovements

Signed-off-by: Gabe Goodhart <ghart@us.ibm.com>

* feat: Update patch 20

Branch: LlamaCPPMetalSSMImprovements

Signed-off-by: Gabe Goodhart <ghart@us.ibm.com>

* feat: Update patches 21-31

Branch: LlamaCPPMetalSSMImprovements

Signed-off-by: Gabe Goodhart <ghart@us.ibm.com>

* feat: Sync vendored code

The two files I'm not sure about here are the swap from gemma3-iswa.cpp to
gemma3.cpp (I chose to include this because I think it's required), and the
inclusion of `ggml-zendnn.h` which I chose to omit.

Branch: LlamaCPPMetalSSMImprovements

Signed-off-by: Gabe Goodhart <ghart@us.ibm.com>

---------

Signed-off-by: Gabe Goodhart <ghart@us.ibm.com>
This commit is contained in:
Gabe Goodhart
2025-12-10 13:59:27 -07:00
committed by GitHub
parent c34fc64688
commit b95693056c
115 changed files with 5176 additions and 2585 deletions
Download Patch File Download Diff File Expand all files Collapse all files

88
llama/llama.cpp/src/llama-model.cpp vendored
View File

@@ -423,8 +423,8 @@ static buft_list_t make_gpu_buft_list(ggml_backend_dev_t dev, llama_split_mode s
}
struct llama_model::impl {
impl() {}
~impl() {}
impl() = default;
~impl() = default;
uint64_t n_elements = 0;
@@ -461,7 +461,7 @@ llama_model::llama_model(const llama_model_params & params) : params(params), pi
pimpl->has_tensor_overrides = params.tensor_buft_overrides && params.tensor_buft_overrides[0].pattern;
}
llama_model::~llama_model() {}
llama_model::~llama_model() = default;
void llama_model::load_stats(llama_model_loader & ml) {
pimpl->n_elements = ml.n_elements;
@@ -663,8 +663,10 @@ void llama_model::load_hparams(llama_model_loader & ml) {
hparams.swa_type = LLAMA_SWA_TYPE_NONE;
hparams.n_no_rope_layer_step = hparams.n_layer; // always use rope
} else {
hparams.swa_type = LLAMA_SWA_TYPE_CHUNKED;
hparams.n_swa = 8192;
hparams.swa_type = LLAMA_SWA_TYPE_CHUNKED;
hparams.n_swa = 8192;
hparams.n_attn_temp_floor_scale = 8192;
hparams.f_attn_temp_scale = 0.1f;
hparams.set_swa_pattern(4); // pattern: 3 chunked - 1 full
}
@@ -1262,18 +1264,25 @@ void llama_model::load_hparams(llama_model_loader & ml) {
} break;
case LLM_ARCH_GEMMA3:
{
hparams.swa_type = LLAMA_SWA_TYPE_STANDARD;
hparams.set_swa_pattern(6);
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;
hparams.set_swa_pattern(6);
hparams.rope_freq_base_train_swa = 10000.0f;
hparams.rope_freq_scale_train_swa = 1.0f;
hparams.rope_freq_base_train_swa = 10000.0f;
hparams.rope_freq_scale_train_swa = 1.0f;
} else {
hparams.swa_type = LLAMA_SWA_TYPE_NONE;
}
ml.get_key(LLM_KV_ATTENTION_SLIDING_WINDOW, hparams.n_swa);
hparams.f_final_logit_softcapping = 0.0f;
ml.get_key(LLM_KV_FINAL_LOGIT_SOFTCAPPING, hparams.f_final_logit_softcapping, false);
ml.get_key(LLM_KV_ATTENTION_LAYERNORM_RMS_EPS, hparams.f_norm_rms_eps);
switch (hparams.n_layer) {
case 18: type = LLM_TYPE_270M; break;
case 26: type = LLM_TYPE_1B; break;
case 32: type = LLM_TYPE_8B; break; // Rnj-1
case 34: type = LLM_TYPE_4B; break;
case 48: type = LLM_TYPE_12B; break;
case 62: type = LLM_TYPE_27B; break;
@@ -1597,8 +1606,9 @@ void llama_model::load_hparams(llama_model_loader & ml) {
ml.get_key(LLM_KV_EXPERT_SHARED_COUNT, hparams.n_expert_shared);
ml.get_key(LLM_KV_EXPERT_WEIGHTS_SCALE, hparams.expert_weights_scale);
switch (hparams.n_layer) {
case 28: type = LLM_TYPE_20B; break;
switch (hparams.n_ff_exp) {
case 1408: type = LLM_TYPE_16B; break;
case 1792: type = LLM_TYPE_20B; break;
default: type = LLM_TYPE_UNKNOWN;
}
} break;
@@ -1626,6 +1636,10 @@ void llama_model::load_hparams(llama_model_loader & ml) {
}
ml.get_key(LLM_KV_ROPE_SCALING_YARN_LOG_MUL, hparams.rope_yarn_log_mul, false);
// (optional) temperature tuning - used by mistral-large
ml.get_key(LLM_KV_ATTENTION_TEMPERATURE_SCALE, hparams.f_attn_temp_scale, false);
ml.get_key(LLM_KV_ATTENTION_TEMPERATURE_LENGTH, hparams.n_attn_temp_floor_scale, false);
switch (hparams.n_layer) {
case 27: type = LLM_TYPE_16B; break;
case 60: type = LLM_TYPE_236B; break;
@@ -2262,6 +2276,42 @@ void llama_model::load_hparams(llama_model_loader & ml) {
default: type = LLM_TYPE_UNKNOWN;
}
} break;
case LLM_ARCH_MISTRAL3:
{
ml.get_key(LLM_KV_ATTENTION_LAYERNORM_RMS_EPS, hparams.f_norm_rms_eps);
ml.get_key(LLM_KV_ATTENTION_TEMPERATURE_SCALE, hparams.f_attn_temp_scale, false);
ml.get_key(LLM_KV_ROPE_SCALING_YARN_BETA_FAST, hparams.yarn_beta_fast, false);
ml.get_key(LLM_KV_ROPE_SCALING_YARN_BETA_SLOW, hparams.yarn_beta_slow, false);
ml.get_key(LLM_KV_ROPE_SCALING_YARN_LOG_MUL, hparams.rope_yarn_log_mul, false);
// TODO: maybe add n_attn_temp_floor_scale as a separate KV?
if (hparams.f_attn_temp_scale != 0.0f) {
hparams.n_attn_temp_floor_scale = hparams.n_ctx_orig_yarn;
if (hparams.n_attn_temp_floor_scale == 0) {
throw std::runtime_error("invalid n_ctx_orig_yarn for attention temperature scaling");
}
}
// TODO: this seems to be correct with the case of mscale == mscale_all_dims == 1.0f
// but may need further verification with other values
if (hparams.rope_yarn_log_mul != 0.0f) {
float factor = 1.0f / hparams.rope_freq_scale_train;
float mscale = 1.0f;
float mscale_all_dims = hparams.rope_yarn_log_mul;
static auto get_mscale = [](float scale, float mscale) {
return scale <= 1.0f ? 1.0f : (0.1f * mscale * logf(scale) + 1.0f);
};
hparams.yarn_attn_factor = get_mscale(factor, mscale) / get_mscale(factor, mscale_all_dims);
}
switch (hparams.n_layer) {
case 26: type = LLM_TYPE_3B; break;
case 34: type = LLM_TYPE_8B; break;
case 40: type = LLM_TYPE_14B; break;
default: type = LLM_TYPE_UNKNOWN;
}
} break;
default: throw std::runtime_error("unsupported model architecture");
}
@@ -2575,6 +2625,7 @@ bool llama_model::load_tensors(llama_model_loader & ml) {
case LLM_ARCH_MINICPM:
case LLM_ARCH_GRANITE:
case LLM_ARCH_GRANITE_MOE:
case LLM_ARCH_MISTRAL3:
{
tok_embd = create_tensor(tn(LLM_TENSOR_TOKEN_EMBD, "weight"), {n_embd, n_vocab}, 0);
@@ -6530,7 +6581,7 @@ bool llama_model::load_tensors(llama_model_loader & ml) {
layer.ssm_in = create_tensor(tn(LLM_TENSOR_SSM_IN, "weight", i), { n_embd, qkvz_dim }, 0);
layer.ssm_conv1d = create_tensor(tn(LLM_TENSOR_SSM_CONV1D, "weight", i), { hparams.ssm_d_conv, conv_dim }, 0);
layer.ssm_dt = create_tensor(tn(LLM_TENSOR_SSM_DT, "bias", i), { hparams.ssm_dt_rank }, 0);
layer.ssm_a = create_tensor(tn(LLM_TENSOR_SSM_A, i), { hparams.ssm_dt_rank }, 0);
layer.ssm_a = create_tensor(tn(LLM_TENSOR_SSM_A_NOSCAN, i), { hparams.ssm_dt_rank }, 0);
layer.ssm_beta_alpha = create_tensor(tn(LLM_TENSOR_SSM_BETA_ALPHA, "weight", i), { n_embd, ba_dim }, 0);
layer.ssm_norm = create_tensor(tn(LLM_TENSOR_SSM_NORM, "weight", i), { head_v_dim }, 0);
layer.ssm_out = create_tensor(tn(LLM_TENSOR_SSM_OUT, "weight", i), { value_dim, n_embd }, 0);
@@ -7304,7 +7355,11 @@ ggml_cgraph * llama_model::build_graph(const llm_graph_params & params) const {
} break;
case LLM_ARCH_GEMMA3:
{
llm = std::make_unique<llm_build_gemma3_iswa>(*this, params);
if (hparams.swa_type == LLAMA_SWA_TYPE_STANDARD) {
llm = std::make_unique<llm_build_gemma3<true>>(*this, params);
} else {
llm = std::make_unique<llm_build_gemma3<false>>(*this, params);
}
} break;
case LLM_ARCH_GEMMA3N:
{
@@ -7569,6 +7624,10 @@ ggml_cgraph * llama_model::build_graph(const llm_graph_params & params) const {
{
llm = std::make_unique<llm_build_qwen3next>(*this, params);
} break;
case LLM_ARCH_MISTRAL3:
{
llm = std::make_unique<llm_build_mistral3>(*this, params);
} break;
default:
GGML_ABORT("fatal error");
}
@@ -7738,6 +7797,7 @@ llama_rope_type llama_model_rope_type(const llama_model * model) {
case LLM_ARCH_ARCEE:
case LLM_ARCH_ERNIE4_5:
case LLM_ARCH_ERNIE4_5_MOE:
case LLM_ARCH_MISTRAL3:
return LLAMA_ROPE_TYPE_NORM;
// the pairs of head values are offset by n_rot/2