matt/ollama37
1
0
Fork 0
mirror of https://github.com/dogkeeper886/ollama37.git synced 2025-12-11 00:07:07 +00:00
Code Issues Packages Projects Releases Wiki Activity

llama: update to commit 2016f07b (#10352)

Browse Source
This commit is contained in:
Jeffrey Morgan
2025-04-25 09:26:02 +09:00
committed by GitHub
parent 11dde41824
commit e9e5f61c45
46 changed files with 1967 additions and 1753 deletions
Download Patch File Download Diff File Expand all files Collapse all files

46
llama/patches/0005-solar-pro.patch
View File

@@ -1,6 +1,6 @@
From 0000000000000000000000000000000000000000 Mon Sep 17 00:00:00 2001
From: jmorganca <jmorganca@gmail.com>
Date: Tue, 8 Apr 2025 16:03:51 -0700
Date: Sun, 20 Apr 2025 16:11:09 -0700
Subject: [PATCH] solar-pro
adds support for the Solar Pro architecture
@@ -15,7 +15,7 @@ adds support for the Solar Pro architecture
7 files changed, 248 insertions(+)
diff --git a/src/llama-arch.cpp b/src/llama-arch.cpp
index a6fddc7f..0b0fedcd 100644
index 62e1480b..f754bc8f 100644
--- a/src/llama-arch.cpp
+++ b/src/llama-arch.cpp
@@ -68,6 +68,7 @@ static const std::map<llm_arch, const char *> LLM_ARCH_NAMES = {
@@ -31,10 +31,10 @@ index a6fddc7f..0b0fedcd 100644
{ LLM_KV_ATTENTION_SLIDING_WINDOW, "%s.attention.sliding_window" },
{ LLM_KV_ATTENTION_SCALE, "%s.attention.scale" },
+ { LLM_KV_ATTENTION_BLOCK_SKIP_CONNECTION, "%s.attention.block_skip_connection" },
{ LLM_KV_ATTENTION_KEY_LENGTH_MLA, "%s.attention.key_length_mla" },
{ LLM_KV_ATTENTION_VALUE_LENGTH_MLA, "%s.attention.value_length_mla" },
{ LLM_KV_ROPE_DIMENSION_COUNT, "%s.rope.dimension_count" },
{ LLM_KV_ROPE_DIMENSION_SECTIONS, "%s.rope.dimension_sections" },
@@ -1478,6 +1480,24 @@ static const std::map<llm_arch, std::map<llm_tensor, const char *>> LLM_TENSOR_N
@@ -1482,6 +1484,24 @@ 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 a6fddc7f..0b0fedcd 100644
{
LLM_ARCH_WAVTOKENIZER_DEC,
{
@@ -1671,6 +1691,7 @@ static const std::map<llm_tensor, llm_tensor_info> LLM_TENSOR_INFOS = {
@@ -1660,6 +1680,7 @@ static const std::map<llm_tensor, llm_tensor_info> LLM_TENSOR_INFOS = {
{LLM_TENSOR_FFN_EXP_PROBS_B, {LLM_TENSOR_LAYER_REPEATING, GGML_OP_ADD}},
// 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 a6fddc7f..0b0fedcd 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 2c2099b3..74aa3dd0 100644
index 98ca00a1..439aaeab 100644
--- a/src/llama-arch.h
+++ b/src/llama-arch.h
@@ -72,6 +72,7 @@ enum llm_arch {
@@ -84,10 +84,10 @@ index 2c2099b3..74aa3dd0 100644
LLM_KV_ATTENTION_SLIDING_WINDOW,
LLM_KV_ATTENTION_SCALE,
+ LLM_KV_ATTENTION_BLOCK_SKIP_CONNECTION,
LLM_KV_ATTENTION_KEY_LENGTH_MLA,
LLM_KV_ATTENTION_VALUE_LENGTH_MLA,
LLM_KV_ROPE_DIMENSION_COUNT,
LLM_KV_ROPE_DIMENSION_SECTIONS,
@@ -340,6 +342,7 @@ enum llm_tensor {
@@ -344,6 +346,7 @@ enum llm_tensor {
LLM_TENSOR_ENC_OUTPUT_NORM,
LLM_TENSOR_CLS,
LLM_TENSOR_CLS_OUT,
@@ -115,10 +115,10 @@ index 90dfe7a7..8a667960 100644
if (il < n_layer) {
return n_swa > 0 && n_swa_pattern > 0 && il % n_swa_pattern < (n_swa_pattern - 1);
diff --git a/src/llama-hparams.h b/src/llama-hparams.h
index 4e0b5719..c3147cbc 100644
index 80fcd65d..6e278945 100644
--- a/src/llama-hparams.h
+++ b/src/llama-hparams.h
@@ -51,6 +51,8 @@ struct llama_hparams {
@@ -55,6 +55,8 @@ struct llama_hparams {
std::array<uint32_t, LLAMA_MAX_LAYERS> n_head_kv_arr;
std::array<uint32_t, LLAMA_MAX_LAYERS> n_ff_arr;
@@ -127,7 +127,7 @@ index 4e0b5719..c3147cbc 100644
uint32_t n_layer_dense_lead = 0;
uint32_t n_lora_q = 0;
uint32_t n_lora_kv = 0;
@@ -149,6 +151,9 @@ struct llama_hparams {
@@ -153,6 +155,9 @@ struct llama_hparams {
// dimension of the recurrent state embeddings
uint32_t n_embd_v_s() const;
@@ -150,10 +150,10 @@ index ea73a8a7..a012aeae 100644
llama_model_loader::llama_model_loader(
const std::string & fname,
diff --git a/src/llama-model.cpp b/src/llama-model.cpp
index b74dd72c..5fbd0055 100644
index 6b7bfecf..aba42819 100644
--- a/src/llama-model.cpp
+++ b/src/llama-model.cpp
@@ -1372,6 +1372,21 @@ void llama_model::load_hparams(llama_model_loader & ml) {
@@ -1374,6 +1374,21 @@ void llama_model::load_hparams(llama_model_loader & ml) {
default: type = LLM_TYPE_UNKNOWN;
}
} break;
@@ -175,7 +175,7 @@ index b74dd72c..5fbd0055 100644
case LLM_ARCH_WAVTOKENIZER_DEC:
{
ml.get_key(LLM_KV_ATTENTION_LAYERNORM_EPS, hparams.f_norm_eps);
@@ -3701,6 +3716,34 @@ bool llama_model::load_tensors(llama_model_loader & ml) {
@@ -3717,6 +3732,34 @@ 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 +210,7 @@ index b74dd72c..5fbd0055 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);
@@ -12244,6 +12287,165 @@ struct llm_build_chameleon : public llm_graph_context {
@@ -12296,6 +12339,165 @@ struct llm_build_chameleon : public llm_graph_context {
}
};
@@ -309,14 +309,14 @@ index b74dd72c..5fbd0055 100644
+ n_rot, rope_type, n_ctx_orig, freq_base, freq_scale,
+ ext_factor, attn_factor, beta_fast, beta_slow
+ );
+
+
+ cb(Qcur, "Qcur", il);
+ cb(Kcur, "Kcur", il);
+ cb(Vcur, "Vcur", il);
+
+ cur = build_attn(inp_attn, gf,
+ model.layers[il].wo, model.layers[il].bo,
+ Qcur, Kcur, Vcur, nullptr, kq_scale, il);
+ Qcur, Kcur, Vcur, nullptr, nullptr, kq_scale, il);
+ cb(cur, "attn_out", il);
+ }
+
@@ -376,7 +376,7 @@ index b74dd72c..5fbd0055 100644
struct llm_build_wavtokenizer_dec : public llm_graph_context {
llm_build_wavtokenizer_dec(const llama_model & model, const llm_graph_params & params, ggml_cgraph * gf) : llm_graph_context(params) {
ggml_tensor * cur;
@@ -12993,6 +13195,10 @@ llm_graph_result_ptr llama_model::build_graph(
@@ -13045,6 +13247,10 @@ llm_graph_result_ptr llama_model::build_graph(
{
llm = std::make_unique<llm_build_chameleon>(*this, params, gf);
} break;
@@ -387,7 +387,7 @@ index b74dd72c..5fbd0055 100644
case LLM_ARCH_WAVTOKENIZER_DEC:
{
llm = std::make_unique<llm_build_wavtokenizer_dec>(*this, params, gf);
@@ -13139,6 +13345,7 @@ llama_rope_type llama_model_rope_type(const llama_model * model) {
@@ -13191,6 +13397,7 @@ llama_rope_type llama_model_rope_type(const llama_model * model) {
case LLM_ARCH_GRANITE:
case LLM_ARCH_GRANITE_MOE:
case LLM_ARCH_CHAMELEON:
@@ -396,7 +396,7 @@ index b74dd72c..5fbd0055 100644
return LLAMA_ROPE_TYPE_NORM;
diff --git a/src/llama-model.h b/src/llama-model.h
index 0f18dac1..e08d4ae4 100644
index fd82d106..5865d5e9 100644
--- a/src/llama-model.h
+++ b/src/llama-model.h
@@ -62,6 +62,7 @@ enum llm_type {
@@ -407,7 +407,7 @@ index 0f18dac1..e08d4ae4 100644
LLM_TYPE_30B,
LLM_TYPE_32B,
LLM_TYPE_34B,
@@ -305,6 +306,8 @@ struct llama_layer {
@@ -307,6 +308,8 @@ struct llama_layer {
struct ggml_tensor * ffn_up_scale = nullptr;
struct ggml_tensor * ffn_down_scale = nullptr;