mirror of
https://github.com/likelovewant/ollama-for-amd.git
synced 2025-12-21 22:33:56 +00:00
Llama cpp bump (df1b612): granite docling / mamba2 optimizations / multimodal encoding fixes (#12552)
* feat: Bump llama.cpp to df1b612 Branch: LlamaCPPBump-GraniteDocling Signed-off-by: Gabe Goodhart <ghart@us.ibm.com> * fix(mtmd): Correctly encode text chunks during mtmd tokenization There can be text chunks that appear interspersed with the image embeddings that contain template delimiter tokens for some models. These need to be correctly translated to text tokens. Branch: LlamaCPPBump-GraniteDocling Signed-off-by: Gabe Goodhart <ghart@us.ibm.com> * tests: Use MtmdChunk in image_test Branch: LlamaCPPBump-GraniteDocling Signed-off-by: Gabe Goodhart <ghart@us.ibm.com> * style: Fix unnecessary conversion linting Branch: LlamaCPPBump-GraniteDocling Signed-off-by: Gabe Goodhart <ghart@us.ibm.com> * fix(ggml): Revert changes to ggml_hip.cpp These changes were done largely by our code assistant and are likely wrong Branch: LlamaCPPBump-GraniteDocling Signed-off-by: Gabe Goodhart <ghart@us.ibm.com> * fix: Revert changes in mem_nvml.cpp Branch: LlamaCPPBump-GraniteDocling Signed-off-by: Gabe Goodhart <ghart@us.ibm.com> * feat: Update sync point to 1deee0 This brings in several more optimization commits and model support for EmbeddingGemma Branch: LlamaCPPBump-GraniteDocling Signed-off-by: Gabe Goodhart <ghart@us.ibm.com> * feat: Update patches for 1deee0 Branch: LlamaCPPBump-GraniteDocling Signed-off-by: Gabe Goodhart <ghart@us.ibm.com> * feat: sync for bump to 1deee0 Branch: LlamaCPPBump-GraniteDocling Signed-off-by: Gabe Goodhart <ghart@us.ibm.com> * fix: Bad patch updates with errant `+` Branch: LlamaCPPBump-GraniteDocling Signed-off-by: Gabe Goodhart <ghart@us.ibm.com> * feat: Bump llama.cpp/ggml to 7049736 Branch: LlamaCPPBump-GraniteDocling Signed-off-by: Gabe Goodhart <ghart@us.ibm.com> * fix: format-patches after latest bump Branch: LlamaCPPBump-GraniteDocling 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
@@ -9,13 +9,13 @@ adds support for the Solar Pro architecture
|
||||
src/llama-arch.h | 3 +
|
||||
src/llama-hparams.cpp | 8 ++
|
||||
src/llama-hparams.h | 5 +
|
||||
src/llama-model-loader.cpp | 1 +
|
||||
src/llama-model-loader.cpp | 2 +-
|
||||
src/llama-model.cpp | 207 +++++++++++++++++++++++++++++++++++++
|
||||
src/llama-model.h | 3 +
|
||||
7 files changed, 248 insertions(+)
|
||||
7 files changed, 248 insertions(+), 1 deletion(-)
|
||||
|
||||
diff --git a/src/llama-arch.cpp b/src/llama-arch.cpp
|
||||
index 4e8d54c4..f98a3574 100644
|
||||
index 869e4dcc..9f6b6ad2 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 = {
|
||||
@@ -26,7 +26,7 @@ index 4e8d54c4..f98a3574 100644
|
||||
{ LLM_ARCH_WAVTOKENIZER_DEC, "wavtokenizer-dec" },
|
||||
{ LLM_ARCH_PLM, "plm" },
|
||||
{ LLM_ARCH_BAILINGMOE, "bailingmoe" },
|
||||
@@ -177,6 +178,7 @@ static const std::map<llm_kv, const char *> LLM_KV_NAMES = {
|
||||
@@ -179,6 +180,7 @@ 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 4e8d54c4..f98a3574 100644
|
||||
{ LLM_KV_ATTENTION_KEY_LENGTH_MLA, "%s.attention.key_length_mla" },
|
||||
{ LLM_KV_ATTENTION_VALUE_LENGTH_MLA, "%s.attention.value_length_mla" },
|
||||
|
||||
@@ -1879,6 +1881,24 @@ static const std::map<llm_arch, std::map<llm_tensor, const char *>> LLM_TENSOR_N
|
||||
@@ -1893,6 +1895,24 @@ 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 4e8d54c4..f98a3574 100644
|
||||
{
|
||||
LLM_ARCH_WAVTOKENIZER_DEC,
|
||||
{
|
||||
@@ -2368,6 +2388,7 @@ static const std::map<llm_tensor, llm_tensor_info> LLM_TENSOR_INFOS = {
|
||||
@@ -2429,6 +2449,7 @@ 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}},
|
||||
@@ -68,7 +68,7 @@ index 4e8d54c4..f98a3574 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 b5c6f3d7..aa8e0e7b 100644
|
||||
index c3ae7165..dc7a362a 100644
|
||||
--- a/src/llama-arch.h
|
||||
+++ b/src/llama-arch.h
|
||||
@@ -85,6 +85,7 @@ enum llm_arch {
|
||||
@@ -79,7 +79,7 @@ index b5c6f3d7..aa8e0e7b 100644
|
||||
LLM_ARCH_WAVTOKENIZER_DEC,
|
||||
LLM_ARCH_PLM,
|
||||
LLM_ARCH_BAILINGMOE,
|
||||
@@ -181,6 +182,7 @@ enum llm_kv {
|
||||
@@ -183,6 +184,7 @@ enum llm_kv {
|
||||
LLM_KV_ATTENTION_SCALE,
|
||||
LLM_KV_ATTENTION_OUTPUT_SCALE,
|
||||
LLM_KV_ATTENTION_TEMPERATURE_LENGTH,
|
||||
@@ -87,7 +87,7 @@ index b5c6f3d7..aa8e0e7b 100644
|
||||
LLM_KV_ATTENTION_KEY_LENGTH_MLA,
|
||||
LLM_KV_ATTENTION_VALUE_LENGTH_MLA,
|
||||
|
||||
@@ -417,6 +419,7 @@ enum llm_tensor {
|
||||
@@ -432,6 +434,7 @@ enum llm_tensor {
|
||||
LLM_TENSOR_ENC_OUTPUT_NORM,
|
||||
LLM_TENSOR_CLS,
|
||||
LLM_TENSOR_CLS_OUT,
|
||||
@@ -96,10 +96,10 @@ index b5c6f3d7..aa8e0e7b 100644
|
||||
LLM_TENSOR_CONVNEXT_DW,
|
||||
LLM_TENSOR_CONVNEXT_NORM,
|
||||
diff --git a/src/llama-hparams.cpp b/src/llama-hparams.cpp
|
||||
index c04ac58f..24a515a0 100644
|
||||
index db65d69e..b6bf6bbf 100644
|
||||
--- a/src/llama-hparams.cpp
|
||||
+++ b/src/llama-hparams.cpp
|
||||
@@ -147,6 +147,14 @@ uint32_t llama_hparams::n_pos_per_embd() const {
|
||||
@@ -151,6 +151,14 @@ uint32_t llama_hparams::n_pos_per_embd() const {
|
||||
return rope_type == LLAMA_ROPE_TYPE_MROPE ? 4 : 1;
|
||||
}
|
||||
|
||||
@@ -115,7 +115,7 @@ index c04ac58f..24a515a0 100644
|
||||
if (il < n_layer) {
|
||||
return swa_layers[il];
|
||||
diff --git a/src/llama-hparams.h b/src/llama-hparams.h
|
||||
index 0fe4b569..eb13709f 100644
|
||||
index 4e7f73ec..80582728 100644
|
||||
--- a/src/llama-hparams.h
|
||||
+++ b/src/llama-hparams.h
|
||||
@@ -64,6 +64,8 @@ struct llama_hparams {
|
||||
@@ -127,7 +127,7 @@ index 0fe4b569..eb13709f 100644
|
||||
uint32_t n_layer_dense_lead = 0;
|
||||
uint32_t n_lora_q = 0;
|
||||
uint32_t n_lora_kv = 0;
|
||||
@@ -236,6 +238,9 @@ struct llama_hparams {
|
||||
@@ -248,6 +250,9 @@ struct llama_hparams {
|
||||
|
||||
uint32_t n_pos_per_embd() const;
|
||||
|
||||
@@ -138,22 +138,23 @@ index 0fe4b569..eb13709f 100644
|
||||
|
||||
bool has_kv(uint32_t il) const;
|
||||
diff --git a/src/llama-model-loader.cpp b/src/llama-model-loader.cpp
|
||||
index 8182a9ad..daef900c 100644
|
||||
index aa3a65f8..ee303bd5 100644
|
||||
--- a/src/llama-model-loader.cpp
|
||||
+++ b/src/llama-model-loader.cpp
|
||||
@@ -465,6 +465,7 @@ namespace GGUFMeta {
|
||||
// TODO: this is not very clever - figure out something better
|
||||
@@ -466,7 +466,7 @@ namespace GGUFMeta {
|
||||
template bool llama_model_loader::get_key_or_arr<std::array<int, 4>>(enum llm_kv kid, std::array<int, 4> & result, uint32_t n, bool required);
|
||||
template bool llama_model_loader::get_key_or_arr<std::array<uint32_t, 512>>(enum llm_kv kid, std::array<uint32_t, 512> & result, uint32_t n, bool required);
|
||||
template bool llama_model_loader::get_key_or_arr<std::array<float, 512>>(enum llm_kv kid, std::array<float, 512> & result, uint32_t n, bool required);
|
||||
-
|
||||
+ template bool llama_model_loader::get_key_or_arr<uint32_t>(const std::string & key, std::array<uint32_t, 512> & result, uint32_t n, bool required);
|
||||
|
||||
llama_model_loader::llama_model_loader(
|
||||
const std::string & fname,
|
||||
diff --git a/src/llama-model.cpp b/src/llama-model.cpp
|
||||
index 2470f878..0398b553 100644
|
||||
index 36d495d6..74e1d162 100644
|
||||
--- a/src/llama-model.cpp
|
||||
+++ b/src/llama-model.cpp
|
||||
@@ -1845,6 +1845,21 @@ void llama_model::load_hparams(llama_model_loader & ml) {
|
||||
@@ -1865,6 +1865,21 @@ void llama_model::load_hparams(llama_model_loader & ml) {
|
||||
default: type = LLM_TYPE_UNKNOWN;
|
||||
}
|
||||
} break;
|
||||
@@ -175,7 +176,7 @@ index 2470f878..0398b553 100644
|
||||
case LLM_ARCH_WAVTOKENIZER_DEC:
|
||||
{
|
||||
ml.get_key(LLM_KV_ATTENTION_LAYERNORM_EPS, hparams.f_norm_eps);
|
||||
@@ -5113,6 +5128,34 @@ bool llama_model::load_tensors(llama_model_loader & ml) {
|
||||
@@ -5170,6 +5185,34 @@ bool llama_model::load_tensors(llama_model_loader & ml) {
|
||||
|
||||
layer.ffn_norm = create_tensor(tn(LLM_TENSOR_FFN_NORM, "weight", i), {n_embd}, 0);
|
||||
|
||||
@@ -210,7 +211,7 @@ index 2470f878..0398b553 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);
|
||||
@@ -16273,6 +16316,165 @@ struct llm_build_granite_hybrid : public llm_graph_context_mamba {
|
||||
@@ -16392,6 +16435,165 @@ struct llm_build_granite_hybrid : public llm_graph_context_mamba {
|
||||
}
|
||||
};
|
||||
|
||||
@@ -376,7 +377,7 @@ index 2470f878..0398b553 100644
|
||||
// ref: https://github.com/facebookresearch/chameleon
|
||||
// based on the original build_llama() function, changes:
|
||||
// * qk-norm
|
||||
@@ -19552,6 +19754,10 @@ ggml_cgraph * llama_model::build_graph(const llm_graph_params & params) const {
|
||||
@@ -19827,6 +20029,10 @@ ggml_cgraph * llama_model::build_graph(const llm_graph_params & params) const {
|
||||
{
|
||||
llm = std::make_unique<llm_build_chameleon>(*this, params);
|
||||
} break;
|
||||
@@ -387,7 +388,7 @@ index 2470f878..0398b553 100644
|
||||
case LLM_ARCH_WAVTOKENIZER_DEC:
|
||||
{
|
||||
llm = std::make_unique<llm_build_wavtokenizer_dec>(*this, params);
|
||||
@@ -19770,6 +19976,7 @@ llama_rope_type llama_model_rope_type(const llama_model * model) {
|
||||
@@ -20057,6 +20263,7 @@ 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:
|
||||
@@ -396,7 +397,7 @@ index 2470f878..0398b553 100644
|
||||
case LLM_ARCH_NEO_BERT:
|
||||
case LLM_ARCH_SMOLLM3:
|
||||
diff --git a/src/llama-model.h b/src/llama-model.h
|
||||
index d73ce969..c086f94e 100644
|
||||
index 7f48662f..ec3fbd33 100644
|
||||
--- a/src/llama-model.h
|
||||
+++ b/src/llama-model.h
|
||||
@@ -76,6 +76,7 @@ enum llm_type {
|
||||
@@ -407,9 +408,9 @@ index d73ce969..c086f94e 100644
|
||||
LLM_TYPE_27B,
|
||||
LLM_TYPE_30B,
|
||||
LLM_TYPE_32B,
|
||||
@@ -380,6 +381,8 @@ struct llama_layer {
|
||||
// openai-moe
|
||||
struct ggml_tensor * attn_sinks = nullptr;
|
||||
@@ -387,6 +388,8 @@ struct llama_layer {
|
||||
struct ggml_tensor * ffn_act_beta = nullptr;
|
||||
struct ggml_tensor * ffn_act_eps = nullptr;
|
||||
|
||||
+ struct ggml_tensor * bskcn_tv = nullptr;
|
||||
+
|
||||
|
||||
Reference in New Issue
Block a user