sdgoij/llama.cpp
1
0
Fork 0
mirror of https://github.com/ggml-org/llama.cpp.git synced 2026-05-14 13:04:08 +00:00
Code Issues Packages Projects Releases Wiki Activity

Merge remote-tracking branch 'upstream/master' into backend-sampling

Browse Source
This commit is contained in:
Daniel Bevenius
2025-12-30 07:59:02 +01:00
parent 060c0a585e 51a48720b8
commit ebfe545cf9
56 changed files with 1210 additions and 635 deletions
Download Patch File Download Diff File Expand all files Collapse all files

1
src/CMakeLists.txt
View File

@@ -107,6 +107,7 @@ add_library(llama
models/phi3.cpp
models/plamo.cpp
models/plamo2.cpp
models/plamo3.cpp
models/plm.cpp
models/qwen.cpp
models/qwen2.cpp

17
src/llama-arch.cpp
View File

@@ -42,6 +42,7 @@ static const std::map<llm_arch, const char *> LLM_ARCH_NAMES = {
{ LLM_ARCH_PHIMOE, "phimoe" },
{ LLM_ARCH_PLAMO, "plamo" },
{ LLM_ARCH_PLAMO2, "plamo2" },
{ LLM_ARCH_PLAMO3, "plamo3" },
{ LLM_ARCH_CODESHELL, "codeshell" },
{ LLM_ARCH_ORION, "orion" },
{ LLM_ARCH_INTERNLM2, "internlm2" },
@@ -1077,6 +1078,22 @@ static std::set<llm_tensor> llm_get_tensor_names(llm_arch arch) {
LLM_TENSOR_ATTN_POST_NORM,
LLM_TENSOR_FFN_POST_NORM,
};
case LLM_ARCH_PLAMO3:
return {
LLM_TENSOR_TOKEN_EMBD,
LLM_TENSOR_OUTPUT_NORM,
LLM_TENSOR_OUTPUT,
LLM_TENSOR_ATTN_NORM,
LLM_TENSOR_ATTN_QKV,
LLM_TENSOR_ATTN_Q_NORM,
LLM_TENSOR_ATTN_K_NORM,
LLM_TENSOR_ATTN_OUT,
LLM_TENSOR_ATTN_POST_NORM,
LLM_TENSOR_FFN_NORM,
LLM_TENSOR_FFN_POST_NORM,
LLM_TENSOR_FFN_DOWN,
LLM_TENSOR_FFN_UP,
};
case LLM_ARCH_CODESHELL:
return {
LLM_TENSOR_TOKEN_EMBD,

1
src/llama-arch.h
View File

@@ -46,6 +46,7 @@ enum llm_arch {
LLM_ARCH_PHIMOE,
LLM_ARCH_PLAMO,
LLM_ARCH_PLAMO2,
LLM_ARCH_PLAMO3,
LLM_ARCH_CODESHELL,
LLM_ARCH_ORION,
LLM_ARCH_INTERNLM2,

67
src/llama-model.cpp
View File

@@ -1227,6 +1227,26 @@ void llama_model::load_hparams(llama_model_loader & ml) {
ml.get_key(LLM_KV_ATTENTION_KEY_LENGTH, hparams.n_embd_head_k, false);
ml.get_key(LLM_KV_ATTENTION_VALUE_LENGTH, hparams.n_embd_head_v, false);
} break;
case LLM_ARCH_PLAMO3:
{
ml.get_key(LLM_KV_ATTENTION_LAYERNORM_RMS_EPS, hparams.f_norm_rms_eps);
const bool found_swa = ml.get_key(LLM_KV_ATTENTION_SLIDING_WINDOW, hparams.n_swa, false);
if (found_swa && hparams.n_swa > 0) {
uint32_t swa_period = 8;
hparams.swa_type = LLAMA_SWA_TYPE_STANDARD;
hparams.rope_freq_scale_train_swa = 1.0f;
ml.get_key(LLM_KV_ROPE_FREQ_BASE_SWA, hparams.rope_freq_base_train_swa);
ml.get_key_or_arr(LLM_KV_ATTENTION_SLIDING_WINDOW_PATTERN, swa_period, false);
hparams.set_swa_pattern(swa_period);
} else {
hparams.swa_type = LLAMA_SWA_TYPE_NONE;
}
switch (hparams.n_layer) {
case 24: type = LLM_TYPE_2B; break;
default: type = LLM_TYPE_UNKNOWN;
}
} break;
case LLM_ARCH_GPT2:
{
ml.get_key(LLM_KV_ATTENTION_LAYERNORM_EPS, hparams.f_norm_eps);
@@ -3828,6 +3848,44 @@ bool llama_model::load_tensors(llama_model_loader & ml) {
layer.ffn_post_norm = create_tensor(tn(LLM_TENSOR_FFN_POST_NORM, i), {n_embd}, 0);
}
} break;
case LLM_ARCH_PLAMO3:
{
const int64_t head_dim_q = hparams.n_embd_head_k;
const int64_t head_dim_v = hparams.n_embd_head_v;
tok_embd = create_tensor(tn(LLM_TENSOR_TOKEN_EMBD, "weight"), {n_embd, n_vocab}, 0);
output_norm = create_tensor(tn(LLM_TENSOR_OUTPUT_NORM, "weight"), {n_embd}, 0);
output = create_tensor(tn(LLM_TENSOR_OUTPUT, "weight"), {n_embd, n_vocab}, TENSOR_NOT_REQUIRED);
if (output == NULL) {
output = create_tensor(tn(LLM_TENSOR_TOKEN_EMBD, "weight"), {n_embd, n_vocab}, TENSOR_DUPLICATED);
}
for (int i = 0; i < n_layer; ++i) {
auto & layer = layers[i];
const int64_t num_attention_heads = hparams.n_head(i);
const int64_t num_key_value_heads = hparams.n_head_kv(i);
const int64_t q_proj_dim = num_attention_heads * head_dim_q;
const int64_t k_proj_dim = num_key_value_heads * head_dim_q;
const int64_t v_proj_dim = num_key_value_heads * head_dim_v;
const int64_t n_ff_cur = hparams.n_ff(i);
layer.attn_norm = create_tensor(tn(LLM_TENSOR_ATTN_NORM, "weight", i), {n_embd}, 0);
layer.wqkv = create_tensor(tn(LLM_TENSOR_ATTN_QKV, "weight", i),
{n_embd,q_proj_dim + k_proj_dim + v_proj_dim}, 0);
layer.attn_q_norm = create_tensor(tn(LLM_TENSOR_ATTN_Q_NORM, "weight", i), {head_dim_q}, 0);
layer.attn_k_norm = create_tensor(tn(LLM_TENSOR_ATTN_K_NORM, "weight", i), {head_dim_q}, 0);
layer.wo = create_tensor(tn(LLM_TENSOR_ATTN_OUT, "weight", i), {num_attention_heads * head_dim_v, n_embd}, 0);
layer.attn_post_norm = create_tensor(tn(LLM_TENSOR_ATTN_POST_NORM, i), {n_embd}, 0);
layer.ffn_norm = create_tensor(tn(LLM_TENSOR_FFN_NORM, "weight", i), {n_embd}, 0);
layer.ffn_post_norm = create_tensor(tn(LLM_TENSOR_FFN_POST_NORM, i), {n_embd}, 0);
layer.ffn_up = create_tensor(tn(LLM_TENSOR_FFN_UP, "weight", i), {n_embd, n_ff_cur * 2}, 0);
layer.ffn_down = create_tensor(tn(LLM_TENSOR_FFN_DOWN, "weight", i), {n_ff_cur, n_embd}, 0);
}
} break;
case LLM_ARCH_GPT2:
{
tok_embd = create_tensor(tn(LLM_TENSOR_TOKEN_EMBD, "weight"), {n_embd, n_vocab}, 0);
@@ -7473,6 +7531,14 @@ ggml_cgraph * llama_model::build_graph(const llm_graph_params & params) const {
{
llm = std::make_unique<llm_build_plamo2>(*this, params);
} break;
case LLM_ARCH_PLAMO3:
{
if (hparams.swa_type != LLAMA_SWA_TYPE_NONE) {
llm = std::make_unique<llm_build_plamo3<true>> (*this, params);
} else {
llm = std::make_unique<llm_build_plamo3<false>>(*this, params);
}
} break;
case LLM_ARCH_GPT2:
{
llm = std::make_unique<llm_build_gpt2>(*this, params);
@@ -7982,6 +8048,7 @@ llama_rope_type llama_model_rope_type(const llama_model * model) {
case LLM_ARCH_PHIMOE:
case LLM_ARCH_PLAMO:
case LLM_ARCH_PLAMO2:
case LLM_ARCH_PLAMO3:
case LLM_ARCH_GEMMA:
case LLM_ARCH_GEMMA2:
case LLM_ARCH_GEMMA3:

3
src/llama.cpp
View File

@@ -512,6 +512,9 @@ static void llama_params_fit_impl(
if (mem_high[id] > targets[id]) {
assert(ngl_per_device_high[id].n_layer > ngl_per_device[id].n_layer);
uint32_t delta = ngl_per_device_high[id].n_layer - ngl_per_device[id].n_layer;
if (hp_nex > 0 && size_t(id) == nd - 1) {
delta--;
}
LLAMA_LOG_DEBUG("%s: start filling device %" PRIu32 ", delta=%" PRIu32 "\n", __func__, id, delta);
while (delta > 1) {
uint32_t step_size = int64_t(delta) * (targets[id] - mem[id]) / (mem_high[id] - mem[id]);

5
src/models/models.h
View File

@@ -406,6 +406,11 @@ struct llm_build_plamo : public llm_graph_context {
llm_build_plamo(const llama_model & model, const llm_graph_params & params);
};
template <bool iswa>
struct llm_build_plamo3 : public llm_graph_context {
llm_build_plamo3(const llama_model & model, const llm_graph_params & params);
};
struct llm_build_plm : public llm_graph_context {
llm_build_plm(const llama_model & model, const llm_graph_params & params);
};

128
src/models/plamo3.cpp Normal file
View File

@@ -0,0 +1,128 @@
#include "models.h"
template <bool iswa>
llm_build_plamo3<iswa>::llm_build_plamo3(const llama_model & model, const llm_graph_params & params) :
llm_graph_context(params) {
const int64_t head_dim_q = hparams.n_embd_head_k;
const int64_t head_dim_v = hparams.n_embd_head_v;
ggml_tensor * cur;
ggml_tensor * inpL = build_inp_embd(model.tok_embd);
ggml_tensor * inp_pos = build_inp_pos();
using inp_attn_type = std::conditional_t<iswa, llm_graph_input_attn_kv_iswa, llm_graph_input_attn_kv>;
inp_attn_type * inp_attn = nullptr;
if constexpr (iswa) {
inp_attn = build_attn_inp_kv_iswa();
} else {
inp_attn = build_attn_inp_kv();
}
ggml_tensor * inp_out_ids = build_inp_out_ids();
for (int il = 0; il < n_layer; ++il) {
ggml_tensor * residual = inpL;
float freq_base_l = 0.0f;
float freq_scale_l = 0.0f;
if constexpr (iswa) {
freq_base_l = model.get_rope_freq_base (cparams, il);
freq_scale_l = model.get_rope_freq_scale(cparams, il);
} else {
freq_base_l = freq_base;
freq_scale_l = freq_scale;
}
cur = build_norm(inpL, model.layers[il].attn_norm, NULL, LLM_NORM_RMS, il);
cb(cur, "attn_norm", il);
ggml_tensor * qkv = build_lora_mm(model.layers[il].wqkv, cur);
cb(cur, "wqkv", il);
const int32_t n_head = hparams.n_head(il);
const int32_t n_head_kv = hparams.n_head_kv(il);
const int64_t q_offset = 0;
const int64_t k_offset = head_dim_q * n_head;
const int64_t v_offset = k_offset + head_dim_q * n_head_kv;
ggml_tensor * Qcur = ggml_view_3d(ctx0, qkv, head_dim_q, n_head, n_tokens,
head_dim_q * sizeof(float), qkv->nb[1], q_offset * ggml_element_size(qkv));
ggml_tensor * Kcur = ggml_view_3d(ctx0, qkv, head_dim_q, n_head_kv, n_tokens,
head_dim_q * sizeof(float), qkv->nb[1], k_offset * ggml_element_size(qkv));
ggml_tensor * Vcur = ggml_view_3d(ctx0, qkv, head_dim_v, n_head_kv, n_tokens,
head_dim_v * sizeof(float), qkv->nb[1], v_offset * ggml_element_size(qkv));
cb(Qcur, "Qcur", il);
cb(Kcur, "Kcur", il);
cb(Vcur, "Vcur", il);
Qcur = build_norm(Qcur, model.layers[il].attn_q_norm, NULL, LLM_NORM_RMS, il);
cb(Qcur, "attn_q_norm", il);
Kcur = build_norm(Kcur, model.layers[il].attn_k_norm, NULL, LLM_NORM_RMS, il);
cb(Kcur, "attn_k_norm", il);
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);
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);
const float attn_scale = 1.0f / sqrtf(float(head_dim_q));
cur = build_attn(inp_attn,
model.layers[il].wo, NULL,
Qcur, Kcur, Vcur, nullptr, nullptr, nullptr, attn_scale, il);
cb(cur, "attn_out", il);
if (il == n_layer - 1 && inp_out_ids) {
cur = ggml_get_rows(ctx0, cur, inp_out_ids);
residual = ggml_get_rows(ctx0, residual, inp_out_ids);
}
cur = build_norm(cur, model.layers[il].attn_post_norm, NULL, LLM_NORM_RMS, il);
cb(cur, "attn_post_norm", il);
cur = ggml_add(ctx0, cur, residual);
cb(cur, "attn_residual", il);
residual = cur;
cur = build_norm(cur, model.layers[il].ffn_norm, NULL, LLM_NORM_RMS, il);
cb(cur, "ffn_norm", il);
cur = build_ffn(cur,
model.layers[il].ffn_up, NULL, NULL,
NULL, NULL, NULL,
model.layers[il].ffn_down, NULL, NULL,
NULL,
LLM_FFN_SWIGLU, LLM_FFN_SEQ, il);
cb(cur, "ffn_out", il);
cur = build_norm(cur, model.layers[il].ffn_post_norm, NULL, LLM_NORM_RMS, il);
cb(cur, "ffn_post_norm", il);
cur = ggml_add(ctx0, cur, residual);
cb(cur, "ffn_residual", il);
cur = build_cvec(cur, il);
cb(cur, "l_out", il);
inpL = cur;
}
cur = inpL;
cur = build_norm(cur, model.output_norm, NULL, LLM_NORM_RMS, -1);
res->t_embd = cur;
cur = build_lora_mm(model.output, cur);
res->t_logits = cur;
ggml_build_forward_expand(gf, cur);
}
// Explicit template instantiations
template struct llm_build_plamo3<false>;
template struct llm_build_plamo3<true>;