mirrors/ollama-for-amd
1
0
Fork 0
mirror of https://github.com/likelovewant/ollama-for-amd.git synced 2025-12-25 07:58:01 +00:00
Code Issues Packages Projects Releases Wiki Activity

GGML update to ec98e2002 (#13451)

Browse Source 
* Revert "add support for NVIDIA Nemotron 3 Nano"

This reverts commit e7d2ae9d69421012e9a8765c06a3fdf0e45b12f3.

* GGML update to 380b4c984

Remove MaskBatchPadding as GGML_KQ_MASK_PAD is no longer present (no
padding required)

* update to c45f89d55

* ec98e2002

solar pro needed more adjusting - needs verification

* review comments
This commit is contained in:
Daniel Hiltgen
2025-12-17 13:13:55 -08:00
committed by GitHub
parent 1c094038bc
commit 49a9c9ba6a
127 changed files with 8128 additions and 6710 deletions
Download Patch File Download Diff File Expand all files Collapse all files

104
llama/patches/0004-solar-pro.patch
View File

@@ -6,7 +6,7 @@ Subject: [PATCH] solar-pro
adds support for the Solar Pro architecture
---
src/CMakeLists.txt | 1 +
src/llama-arch.cpp | 21 +++++
src/llama-arch.cpp | 20 +++++
src/llama-arch.h | 3 +
src/llama-hparams.cpp | 8 ++
src/llama-hparams.h | 5 ++
@@ -15,7 +15,7 @@ adds support for the Solar Pro architecture
src/llama-model.h | 3 +
src/models/models.h | 5 ++
src/models/solar.cpp | 158 +++++++++++++++++++++++++++++++++++++
10 files changed, 253 insertions(+), 1 deletion(-)
10 files changed, 252 insertions(+), 1 deletion(-)
create mode 100644 src/models/solar.cpp
diff --git a/src/CMakeLists.txt b/src/CMakeLists.txt
@@ -31,10 +31,10 @@ index 4192af7c0..bd44d73e7 100644
models/starcoder.cpp
models/starcoder2.cpp
diff --git a/src/llama-arch.cpp b/src/llama-arch.cpp
index 64ad1b776..a5fe4f66c 100644
index 8caf80afc..2ce8ffec0 100644
--- a/src/llama-arch.cpp
+++ b/src/llama-arch.cpp
@@ -85,6 +85,7 @@ static const std::map<llm_arch, const char *> LLM_ARCH_NAMES = {
@@ -87,6 +87,7 @@ 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" },
@@ -42,7 +42,7 @@ index 64ad1b776..a5fe4f66c 100644
{ LLM_ARCH_WAVTOKENIZER_DEC, "wavtokenizer-dec" },
{ LLM_ARCH_PLM, "plm" },
{ LLM_ARCH_BAILINGMOE, "bailingmoe" },
@@ -206,6 +207,7 @@ static const std::map<llm_kv, const char *> LLM_KV_NAMES = {
@@ -208,6 +209,7 @@ static const std::map<llm_kv, const char *> LLM_KV_NAMES = {
{ LLM_KV_ATTENTION_OUTPUT_SCALE, "%s.attention.output_scale" },
{ LLM_KV_ATTENTION_TEMPERATURE_LENGTH, "%s.attention.temperature_length" },
{ LLM_KV_ATTENTION_TEMPERATURE_SCALE, "%s.attention.temperature_scale" },
@@ -50,32 +50,38 @@ index 64ad1b776..a5fe4f66c 100644
{ LLM_KV_ATTENTION_KEY_LENGTH_MLA, "%s.attention.key_length_mla" },
{ LLM_KV_ATTENTION_VALUE_LENGTH_MLA, "%s.attention.value_length_mla" },
@@ -2025,6 +2027,24 @@ 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" },
},
},
+ {
+ LLM_ARCH_SOLAR,
+ {
+ { LLM_TENSOR_TOKEN_EMBD, "token_embd" },
+ { LLM_TENSOR_OUTPUT_NORM, "output_norm" },
+ { LLM_TENSOR_OUTPUT, "output" },
+ { LLM_TENSOR_ATTN_NORM, "blk.%d.attn_norm" },
+ { LLM_TENSOR_ATTN_Q, "blk.%d.attn_q" },
+ { LLM_TENSOR_ATTN_K, "blk.%d.attn_k" },
+ { LLM_TENSOR_ATTN_V, "blk.%d.attn_v" },
+ { LLM_TENSOR_ATTN_OUT, "blk.%d.attn_output" },
+ { LLM_TENSOR_FFN_NORM, "blk.%d.ffn_norm" },
+ { LLM_TENSOR_FFN_GATE, "blk.%d.ffn_gate" },
+ { LLM_TENSOR_FFN_DOWN, "blk.%d.ffn_down" },
+ { LLM_TENSOR_FFN_UP, "blk.%d.ffn_up" },
+ { LLM_TENSOR_BSKCN_TV, "bskcn_tv" },
+ },
+ },
{
LLM_ARCH_WAVTOKENIZER_DEC,
{
@@ -2710,6 +2730,7 @@ static const std::map<llm_tensor, llm_tensor_info> LLM_TENSOR_INFOS = {
@@ -339,6 +341,7 @@ static const std::map<llm_tensor, const char *> LLM_TENSOR_NAMES = {
{ LLM_TENSOR_ATTN_QKV, "blk.%d.attn_qkv" },
{ LLM_TENSOR_LAYER_OUT_NORM, "blk.%d.layer_output_norm" },
{ LLM_TENSOR_ATTN_OUT_NORM, "blk.%d.attn_output_norm" },
+ { LLM_TENSOR_BSKCN_TV, "bskcn_tv" },
{ LLM_TENSOR_POS_EMBD, "position_embd" },
{ LLM_TENSOR_FFN_ACT, "blk.%d.ffn.act" },
{ LLM_TENSOR_TOKEN_EMBD_NORM, "token_embd_norm" },
@@ -2176,6 +2179,22 @@ static std::set<llm_tensor> llm_get_tensor_names(llm_arch arch) {
return {
LLM_TENSOR_TOKEN_EMBD,
};
+ case LLM_ARCH_SOLAR:
+ return {
+ LLM_TENSOR_TOKEN_EMBD,
+ LLM_TENSOR_OUTPUT_NORM,
+ LLM_TENSOR_OUTPUT,
+ LLM_TENSOR_ATTN_NORM,
+ LLM_TENSOR_ATTN_Q,
+ LLM_TENSOR_ATTN_K,
+ LLM_TENSOR_ATTN_V,
+ LLM_TENSOR_ATTN_OUT,
+ LLM_TENSOR_FFN_NORM,
+ LLM_TENSOR_FFN_GATE,
+ LLM_TENSOR_FFN_DOWN,
+ LLM_TENSOR_FFN_UP,
+ LLM_TENSOR_BSKCN_TV,
+ };
default:
GGML_ABORT("unknown architecture for tensor mapping");
}
@@ -2344,6 +2363,7 @@ 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}},
@@ -84,10 +90,10 @@ index 64ad1b776..a5fe4f66c 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 e11318002..ec9e3a6df 100644
index 6cbf9b1f8..14d461c76 100644
--- a/src/llama-arch.h
+++ b/src/llama-arch.h
@@ -89,6 +89,7 @@ enum llm_arch {
@@ -91,6 +91,7 @@ enum llm_arch {
LLM_ARCH_GRANITE_MOE,
LLM_ARCH_GRANITE_HYBRID,
LLM_ARCH_CHAMELEON,
@@ -95,7 +101,7 @@ index e11318002..ec9e3a6df 100644
LLM_ARCH_WAVTOKENIZER_DEC,
LLM_ARCH_PLM,
LLM_ARCH_BAILINGMOE,
@@ -210,6 +211,7 @@ enum llm_kv {
@@ -212,6 +213,7 @@ enum llm_kv {
LLM_KV_ATTENTION_OUTPUT_SCALE,
LLM_KV_ATTENTION_TEMPERATURE_LENGTH,
LLM_KV_ATTENTION_TEMPERATURE_SCALE,
@@ -103,7 +109,7 @@ index e11318002..ec9e3a6df 100644
LLM_KV_ATTENTION_KEY_LENGTH_MLA,
LLM_KV_ATTENTION_VALUE_LENGTH_MLA,
@@ -462,6 +464,7 @@ enum llm_tensor {
@@ -465,6 +467,7 @@ enum llm_tensor {
LLM_TENSOR_ENC_OUTPUT_NORM,
LLM_TENSOR_CLS,
LLM_TENSOR_CLS_OUT,
@@ -112,10 +118,10 @@ index e11318002..ec9e3a6df 100644
LLM_TENSOR_CONVNEXT_DW,
LLM_TENSOR_CONVNEXT_NORM,
diff --git a/src/llama-hparams.cpp b/src/llama-hparams.cpp
index 8cdbaf69f..41127bf91 100644
index fe1fa4341..aabff2f06 100644
--- a/src/llama-hparams.cpp
+++ b/src/llama-hparams.cpp
@@ -161,6 +161,14 @@ uint32_t llama_hparams::n_pos_per_embd() const {
@@ -163,6 +163,14 @@ uint32_t llama_hparams::n_pos_per_embd() const {
return rope_type == LLAMA_ROPE_TYPE_MROPE || rope_type == LLAMA_ROPE_TYPE_IMROPE ? 4 : 1;
}
@@ -131,10 +137,10 @@ index 8cdbaf69f..41127bf91 100644
if (il < n_layer) {
return swa_layers[il];
diff --git a/src/llama-hparams.h b/src/llama-hparams.h
index 6eff334a5..a778fc3cf 100644
index f6e95b5d2..c6e673276 100644
--- a/src/llama-hparams.h
+++ b/src/llama-hparams.h
@@ -64,6 +64,8 @@ struct llama_hparams {
@@ -65,6 +65,8 @@ struct llama_hparams {
std::array<uint32_t, LLAMA_MAX_LAYERS> n_head_kv_arr;
std::array<uint32_t, LLAMA_MAX_LAYERS> n_ff_arr;
@@ -143,7 +149,7 @@ index 6eff334a5..a778fc3cf 100644
uint32_t n_layer_dense_lead = 0;
uint32_t n_lora_q = 0;
uint32_t n_lora_kv = 0;
@@ -256,6 +258,9 @@ struct llama_hparams {
@@ -259,6 +261,9 @@ struct llama_hparams {
uint32_t n_pos_per_embd() const;
@@ -154,7 +160,7 @@ index 6eff334a5..a778fc3cf 100644
bool has_kv(uint32_t il) const;
diff --git a/src/llama-model-loader.cpp b/src/llama-model-loader.cpp
index aa3a65f87..ee303bd58 100644
index ca2ea2461..8916a6242 100644
--- a/src/llama-model-loader.cpp
+++ b/src/llama-model-loader.cpp
@@ -466,7 +466,7 @@ namespace GGUFMeta {
@@ -167,10 +173,10 @@ index aa3a65f87..ee303bd58 100644
llama_model_loader::llama_model_loader(
const std::string & fname,
diff --git a/src/llama-model.cpp b/src/llama-model.cpp
index 04fccc979..3c503b424 100644
index ae8207ee1..00cd579e0 100644
--- a/src/llama-model.cpp
+++ b/src/llama-model.cpp
@@ -1975,6 +1975,21 @@ void llama_model::load_hparams(llama_model_loader & ml) {
@@ -1995,6 +1995,21 @@ void llama_model::load_hparams(llama_model_loader & ml) {
default: type = LLM_TYPE_UNKNOWN;
}
} break;
@@ -192,7 +198,7 @@ index 04fccc979..3c503b424 100644
case LLM_ARCH_WAVTOKENIZER_DEC:
{
ml.get_key(LLM_KV_ATTENTION_LAYERNORM_EPS, hparams.f_norm_eps);
@@ -5401,6 +5416,34 @@ bool llama_model::load_tensors(llama_model_loader & ml) {
@@ -5429,6 +5444,34 @@ bool llama_model::load_tensors(llama_model_loader & ml) {
layer.ffn_norm = create_tensor(tn(LLM_TENSOR_FFN_NORM, "weight", i), {n_embd}, 0);
@@ -227,7 +233,7 @@ index 04fccc979..3c503b424 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);
@@ -7480,6 +7523,10 @@ ggml_cgraph * llama_model::build_graph(const llm_graph_params & params) const {
@@ -7534,6 +7577,10 @@ ggml_cgraph * llama_model::build_graph(const llm_graph_params & params) const {
{
llm = std::make_unique<llm_build_chameleon>(*this, params);
} break;
@@ -238,7 +244,7 @@ index 04fccc979..3c503b424 100644
case LLM_ARCH_WAVTOKENIZER_DEC:
{
llm = std::make_unique<llm_build_wavtokenizer_dec>(*this, params);
@@ -7743,6 +7790,7 @@ llama_rope_type llama_model_rope_type(const llama_model * model) {
@@ -7798,6 +7845,7 @@ 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:
@@ -247,7 +253,7 @@ index 04fccc979..3c503b424 100644
case LLM_ARCH_NEO_BERT:
case LLM_ARCH_SMOLLM3:
diff --git a/src/llama-model.h b/src/llama-model.h
index f8342cf2c..cbf4e1bfa 100644
index c6eb95318..b378b23ec 100644
--- a/src/llama-model.h
+++ b/src/llama-model.h
@@ -76,6 +76,7 @@ enum llm_type {
@@ -258,7 +264,7 @@ index f8342cf2c..cbf4e1bfa 100644
LLM_TYPE_26B,
LLM_TYPE_27B,
LLM_TYPE_30B,
@@ -404,6 +405,8 @@ struct llama_layer {
@@ -405,6 +406,8 @@ struct llama_layer {
struct ggml_tensor * ffn_act_beta = nullptr;
struct ggml_tensor * ffn_act_eps = nullptr;
@@ -268,7 +274,7 @@ index f8342cf2c..cbf4e1bfa 100644
struct llama_layer_convnext convnext;
diff --git a/src/models/models.h b/src/models/models.h
index 6494f5450..e0aec822c 100644
index ffb36acc6..6d84a185d 100644
--- a/src/models/models.h
+++ b/src/models/models.h
@@ -515,6 +515,11 @@ struct llm_build_smollm3 : public llm_graph_context {