mirrors/ollama-for-amd
1
0
Fork 0
mirror of https://github.com/likelovewant/ollama-for-amd.git synced 2025-12-21 22:33:56 +00:00
Code Issues Packages Projects Releases Wiki Activity

ggml update to b6840 (#12791)

Browse Source
This commit is contained in:
Daniel Hiltgen
2025-11-06 10:19:22 -08:00
committed by GitHub
parent c4ba257c64
commit 544b6739dd
103 changed files with 3644 additions and 1215 deletions
Download Patch File Download Diff File Expand all files Collapse all files

42
llama/patches/0004-solar-pro.patch
View File

@@ -15,10 +15,10 @@ adds support for the Solar Pro architecture
7 files changed, 248 insertions(+), 1 deletion(-)
diff --git a/src/llama-arch.cpp b/src/llama-arch.cpp
index 869e4dcc..9f6b6ad2 100644
index 8ca769c5f..ab262ec0c 100644
--- a/src/llama-arch.cpp
+++ b/src/llama-arch.cpp
@@ -81,6 +81,7 @@ static const std::map<llm_arch, const char *> LLM_ARCH_NAMES = {
@@ -82,6 +82,7 @@ static const std::map<llm_arch, const char *> LLM_ARCH_NAMES = {
{ LLM_ARCH_GRANITE_MOE, "granitemoe" },
{ LLM_ARCH_GRANITE_HYBRID, "granitehybrid" },
{ LLM_ARCH_CHAMELEON, "chameleon" },
@@ -26,7 +26,7 @@ index 869e4dcc..9f6b6ad2 100644
{ LLM_ARCH_WAVTOKENIZER_DEC, "wavtokenizer-dec" },
{ LLM_ARCH_PLM, "plm" },
{ LLM_ARCH_BAILINGMOE, "bailingmoe" },
@@ -179,6 +180,7 @@ static const std::map<llm_kv, const char *> LLM_KV_NAMES = {
@@ -183,6 +184,7 @@ static const std::map<llm_kv, const char *> LLM_KV_NAMES = {
{ LLM_KV_ATTENTION_SCALE, "%s.attention.scale" },
{ LLM_KV_ATTENTION_OUTPUT_SCALE, "%s.attention.output_scale" },
{ LLM_KV_ATTENTION_TEMPERATURE_LENGTH, "%s.attention.temperature_length" },
@@ -34,7 +34,7 @@ index 869e4dcc..9f6b6ad2 100644
{ LLM_KV_ATTENTION_KEY_LENGTH_MLA, "%s.attention.key_length_mla" },
{ LLM_KV_ATTENTION_VALUE_LENGTH_MLA, "%s.attention.value_length_mla" },
@@ -1893,6 +1895,24 @@ static const std::map<llm_arch, std::map<llm_tensor, const char *>> LLM_TENSOR_N
@@ -1901,6 +1903,24 @@ static const std::map<llm_arch, std::map<llm_tensor, const char *>> LLM_TENSOR_N
{ LLM_TENSOR_ATTN_K_NORM, "blk.%d.attn_k_norm" },
},
},
@@ -59,7 +59,7 @@ index 869e4dcc..9f6b6ad2 100644
{
LLM_ARCH_WAVTOKENIZER_DEC,
{
@@ -2429,6 +2449,7 @@ static const std::map<llm_tensor, llm_tensor_info> LLM_TENSOR_INFOS = {
@@ -2469,6 +2489,7 @@ static const std::map<llm_tensor, llm_tensor_info> LLM_TENSOR_INFOS = {
{LLM_TENSOR_LAUREL_POST_NORM, {LLM_TENSOR_LAYER_REPEATING, GGML_OP_MUL}},
// this tensor is loaded for T5, but never used
{LLM_TENSOR_DEC_CROSS_ATTN_REL_B, {LLM_TENSOR_LAYER_REPEATING, GGML_OP_NONE}},
@@ -68,10 +68,10 @@ index 869e4dcc..9f6b6ad2 100644
{LLM_TENSOR_POS_NET_NORM, {LLM_TENSOR_LAYER_REPEATING, GGML_OP_MUL}},
{LLM_TENSOR_POS_NET_NORM1, {LLM_TENSOR_LAYER_REPEATING, GGML_OP_MUL}},
diff --git a/src/llama-arch.h b/src/llama-arch.h
index c3ae7165..dc7a362a 100644
index dea725c1a..ea2b4ffb9 100644
--- a/src/llama-arch.h
+++ b/src/llama-arch.h
@@ -85,6 +85,7 @@ enum llm_arch {
@@ -86,6 +86,7 @@ enum llm_arch {
LLM_ARCH_GRANITE_MOE,
LLM_ARCH_GRANITE_HYBRID,
LLM_ARCH_CHAMELEON,
@@ -79,7 +79,7 @@ index c3ae7165..dc7a362a 100644
LLM_ARCH_WAVTOKENIZER_DEC,
LLM_ARCH_PLM,
LLM_ARCH_BAILINGMOE,
@@ -183,6 +184,7 @@ enum llm_kv {
@@ -187,6 +188,7 @@ enum llm_kv {
LLM_KV_ATTENTION_SCALE,
LLM_KV_ATTENTION_OUTPUT_SCALE,
LLM_KV_ATTENTION_TEMPERATURE_LENGTH,
@@ -87,7 +87,7 @@ index c3ae7165..dc7a362a 100644
LLM_KV_ATTENTION_KEY_LENGTH_MLA,
LLM_KV_ATTENTION_VALUE_LENGTH_MLA,
@@ -432,6 +434,7 @@ enum llm_tensor {
@@ -436,6 +438,7 @@ enum llm_tensor {
LLM_TENSOR_ENC_OUTPUT_NORM,
LLM_TENSOR_CLS,
LLM_TENSOR_CLS_OUT,
@@ -96,7 +96,7 @@ index c3ae7165..dc7a362a 100644
LLM_TENSOR_CONVNEXT_DW,
LLM_TENSOR_CONVNEXT_NORM,
diff --git a/src/llama-hparams.cpp b/src/llama-hparams.cpp
index db65d69e..b6bf6bbf 100644
index db65d69ea..b6bf6bbf2 100644
--- a/src/llama-hparams.cpp
+++ b/src/llama-hparams.cpp
@@ -151,6 +151,14 @@ uint32_t llama_hparams::n_pos_per_embd() const {
@@ -115,7 +115,7 @@ index db65d69e..b6bf6bbf 100644
if (il < n_layer) {
return swa_layers[il];
diff --git a/src/llama-hparams.h b/src/llama-hparams.h
index 4e7f73ec..80582728 100644
index 6fcf91b7d..24569a258 100644
--- a/src/llama-hparams.h
+++ b/src/llama-hparams.h
@@ -64,6 +64,8 @@ struct llama_hparams {
@@ -127,7 +127,7 @@ index 4e7f73ec..80582728 100644
uint32_t n_layer_dense_lead = 0;
uint32_t n_lora_q = 0;
uint32_t n_lora_kv = 0;
@@ -248,6 +250,9 @@ struct llama_hparams {
@@ -250,6 +252,9 @@ struct llama_hparams {
uint32_t n_pos_per_embd() const;
@@ -138,7 +138,7 @@ index 4e7f73ec..80582728 100644
bool has_kv(uint32_t il) const;
diff --git a/src/llama-model-loader.cpp b/src/llama-model-loader.cpp
index aa3a65f8..ee303bd5 100644
index aa3a65f87..ee303bd58 100644
--- a/src/llama-model-loader.cpp
+++ b/src/llama-model-loader.cpp
@@ -466,7 +466,7 @@ namespace GGUFMeta {
@@ -151,10 +151,10 @@ index aa3a65f8..ee303bd5 100644
llama_model_loader::llama_model_loader(
const std::string & fname,
diff --git a/src/llama-model.cpp b/src/llama-model.cpp
index 36d495d6..74e1d162 100644
index 2a83d6627..54621ea39 100644
--- a/src/llama-model.cpp
+++ b/src/llama-model.cpp
@@ -1865,6 +1865,21 @@ void llama_model::load_hparams(llama_model_loader & ml) {
@@ -1890,6 +1890,21 @@ void llama_model::load_hparams(llama_model_loader & ml) {
default: type = LLM_TYPE_UNKNOWN;
}
} break;
@@ -176,7 +176,7 @@ index 36d495d6..74e1d162 100644
case LLM_ARCH_WAVTOKENIZER_DEC:
{
ml.get_key(LLM_KV_ATTENTION_LAYERNORM_EPS, hparams.f_norm_eps);
@@ -5170,6 +5185,34 @@ bool llama_model::load_tensors(llama_model_loader & ml) {
@@ -5224,6 +5239,34 @@ bool llama_model::load_tensors(llama_model_loader & ml) {
layer.ffn_norm = create_tensor(tn(LLM_TENSOR_FFN_NORM, "weight", i), {n_embd}, 0);
@@ -211,7 +211,7 @@ index 36d495d6..74e1d162 100644
layer.ffn_gate = create_tensor(tn(LLM_TENSOR_FFN_GATE, "weight", i), {n_embd, n_ff}, 0);
layer.ffn_down = create_tensor(tn(LLM_TENSOR_FFN_DOWN, "weight", i), { n_ff, n_embd}, 0);
layer.ffn_up = create_tensor(tn(LLM_TENSOR_FFN_UP, "weight", i), {n_embd, n_ff}, 0);
@@ -16392,6 +16435,165 @@ struct llm_build_granite_hybrid : public llm_graph_context_mamba {
@@ -16515,6 +16558,165 @@ struct llm_build_granite_hybrid : public llm_graph_context_mamba {
}
};
@@ -377,7 +377,7 @@ index 36d495d6..74e1d162 100644
// ref: https://github.com/facebookresearch/chameleon
// based on the original build_llama() function, changes:
// * qk-norm
@@ -19827,6 +20029,10 @@ ggml_cgraph * llama_model::build_graph(const llm_graph_params & params) const {
@@ -20096,6 +20298,10 @@ ggml_cgraph * llama_model::build_graph(const llm_graph_params & params) const {
{
llm = std::make_unique<llm_build_chameleon>(*this, params);
} break;
@@ -388,7 +388,7 @@ index 36d495d6..74e1d162 100644
case LLM_ARCH_WAVTOKENIZER_DEC:
{
llm = std::make_unique<llm_build_wavtokenizer_dec>(*this, params);
@@ -20057,6 +20263,7 @@ llama_rope_type llama_model_rope_type(const llama_model * model) {
@@ -20331,6 +20537,7 @@ llama_rope_type llama_model_rope_type(const llama_model * model) {
case LLM_ARCH_GRANITE_MOE:
case LLM_ARCH_GRANITE_HYBRID:
case LLM_ARCH_CHAMELEON:
@@ -397,7 +397,7 @@ index 36d495d6..74e1d162 100644
case LLM_ARCH_NEO_BERT:
case LLM_ARCH_SMOLLM3:
diff --git a/src/llama-model.h b/src/llama-model.h
index 7f48662f..ec3fbd33 100644
index 248f85410..4a7924aaa 100644
--- a/src/llama-model.h
+++ b/src/llama-model.h
@@ -76,6 +76,7 @@ enum llm_type {
@@ -408,7 +408,7 @@ index 7f48662f..ec3fbd33 100644
LLM_TYPE_27B,
LLM_TYPE_30B,
LLM_TYPE_32B,
@@ -387,6 +388,8 @@ struct llama_layer {
@@ -390,6 +391,8 @@ struct llama_layer {
struct ggml_tensor * ffn_act_beta = nullptr;
struct ggml_tensor * ffn_act_eps = nullptr;