Public/koboldcpp
1
0
Fork 0
mirror of https://github.com/LostRuins/koboldcpp.git synced 2026-09-19 09:15:18 +02:00
Code Issues Packages Projects Releases Wiki Activity

Merge branch 'upstream' into concedo_experimental

Browse Source 
# Conflicts:
#	.github/workflows/build.yml
#	.github/workflows/close-issue.yml
#	.github/workflows/server.yml
#	AUTHORS
#	CMakeLists.txt
#	Makefile
#	README.md
#	cmake/llama.pc.in
#	common/CMakeLists.txt
#	docs/build.md
#	examples/batched.swift/Sources/main.swift
#	examples/llama.swiftui/llama.cpp.swift/LibLlama.swift
#	examples/llava/CMakeLists.txt
#	examples/llava/clip.h
#	examples/run/run.cpp
#	examples/server/README.md
#	ggml/CMakeLists.txt
#	ggml/src/ggml-cuda/CMakeLists.txt
#	ggml/src/ggml-hip/CMakeLists.txt
#	ggml/src/ggml-musa/CMakeLists.txt
#	scripts/sync-ggml.last
#	tests/CMakeLists.txt
#	tests/test-backend-ops.cpp
#	tests/test-chat-template.cpp
#	tests/test-grammar-integration.cpp
#	tests/test-json-schema-to-grammar.cpp
This commit is contained in:
Concedo
2025-02-07 00:52:31 +08:00
parent 8fef9f3fb5 2fb3c32a16
commit db6db9dff9
142 changed files with 74261 additions and 10014 deletions
Download Patch File Download Diff File Expand all files Collapse all files
src/llama-arch.cpp
+3
View File
@@ -1024,6 +1024,9 @@ static const std::map<llm_arch, std::map<llm_tensor, const char *>> LLM_TENSOR_N
{ LLM_TENSOR_OUTPUT, "output" },
{ LLM_TENSOR_ATTN_NORM, "blk.%d.attn_norm" },
{ LLM_TENSOR_ATTN_QKV, "blk.%d.attn_qkv" },
{ 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_UP, "blk.%d.ffn_up" },
src/llama-chat.cpp
+10 -1
View File
@@ -51,6 +51,7 @@ static const std::map<std::string, llm_chat_template> LLM_CHAT_TEMPLATES = {
{ "llama3", LLM_CHAT_TEMPLATE_LLAMA_3 },
{ "chatglm3", LLM_CHAT_TEMPLATE_CHATGML_3 },
{ "chatglm4", LLM_CHAT_TEMPLATE_CHATGML_4 },
{ "glmedge", LLM_CHAT_TEMPLATE_GLMEDGE },
{ "minicpm", LLM_CHAT_TEMPLATE_MINICPM },
{ "exaone3", LLM_CHAT_TEMPLATE_EXAONE_3 },
{ "rwkv-world", LLM_CHAT_TEMPLATE_RWKV_WORLD },
@@ -115,7 +116,7 @@ llm_chat_template llm_chat_detect_template(const std::string & tmpl) {
} else if (tmpl_contains("<|assistant|>") && tmpl_contains("<|end|>")) {
return LLM_CHAT_TEMPLATE_PHI_3;
} else if (tmpl_contains("<|assistant|>") && tmpl_contains("<|user|>")) {
return LLM_CHAT_TEMPLATE_FALCON_3;
return tmpl_contains("</s>") ? LLM_CHAT_TEMPLATE_FALCON_3 : LLM_CHAT_TEMPLATE_GLMEDGE;
} else if (tmpl_contains("<|user|>") && tmpl_contains("<|endoftext|>")) {
return LLM_CHAT_TEMPLATE_ZEPHYR;
} else if (tmpl_contains("bos_token + message['role']")) {
@@ -440,6 +441,14 @@ int32_t llm_chat_apply_template(
if (add_ass) {
ss << "<|assistant|>";
}
} else if (tmpl == LLM_CHAT_TEMPLATE_GLMEDGE) {
for (auto message : chat) {
std::string role(message->role);
ss << "<|" << role << "|>" << "\n" << message->content;
}
if (add_ass) {
ss << "<|assistant|>";
}
} else if (tmpl == LLM_CHAT_TEMPLATE_MINICPM) {
// MiniCPM-3B-OpenHermes-2.5-v2-GGUF
for (auto message : chat) {
src/llama-chat.h
+1
View File
@@ -31,6 +31,7 @@ enum llm_chat_template {
LLM_CHAT_TEMPLATE_LLAMA_3,
LLM_CHAT_TEMPLATE_CHATGML_3,
LLM_CHAT_TEMPLATE_CHATGML_4,
LLM_CHAT_TEMPLATE_GLMEDGE,
LLM_CHAT_TEMPLATE_MINICPM,
LLM_CHAT_TEMPLATE_EXAONE_3,
LLM_CHAT_TEMPLATE_RWKV_WORLD,
src/llama-grammar.cpp
+3 -1
View File
@@ -1213,5 +1213,7 @@ void llama_grammar_accept_str(struct llama_grammar & grammar, const std::string
}
grammar.partial_utf8 = decoded.second;
GGML_ASSERT(!grammar.stacks.empty());
if (grammar.stacks.empty()) {
throw std::runtime_error("Unexpected empty grammar stack after accepting piece: " + piece);
}
}
src/llama-model.cpp
+26 -4
View File
@@ -1098,8 +1098,20 @@ void llama_model::load_hparams(llama_model_loader & ml) {
{
ml.get_key(LLM_KV_ATTENTION_LAYERNORM_RMS_EPS, hparams.f_norm_rms_eps);
switch (hparams.n_layer) {
case 28: type = LLM_TYPE_6B; break;
case 40: type = LLM_TYPE_9B; break;
case 28: {
if (hparams.n_head(0) == 16) {
type = LLM_TYPE_1_5B;
} else {
type = LLM_TYPE_6B;
}
} break;
case 40: {
if (hparams.n_head(0) == 24) {
type = LLM_TYPE_4B;
} else {
type = LLM_TYPE_9B;
}
} break;
default: type = LLM_TYPE_UNKNOWN;
}
} break;
@@ -1268,6 +1280,8 @@ bool llama_model::load_tensors(llama_model_loader & ml) {
bool use_mmap_buffer = true;
LLAMA_LOG_INFO("%s: loading model tensors, this can take a while... (mmap = %s)\n", __func__, use_mmap_buffer ? "true" : "false");
// build a list of buffer types for the CPU and GPU devices
pimpl->cpu_buft_list = make_cpu_buft_list(devices);
for (auto * dev : devices) {
@@ -3163,9 +3177,17 @@ bool llama_model::load_tensors(llama_model_loader & ml) {
auto & layer = layers[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, n_embd + 2*n_embd_gqa}, llama_model_loader::TENSOR_NOT_REQUIRED);
layer.bqkv = create_tensor(tn(LLM_TENSOR_ATTN_QKV, "bias", i), {n_embd + 2*n_embd_gqa}, llama_model_loader::TENSOR_NOT_REQUIRED);
layer.wqkv = create_tensor(tn(LLM_TENSOR_ATTN_QKV, "weight", i), {n_embd, n_embd + 2*n_embd_gqa}, 0);
layer.bqkv = create_tensor(tn(LLM_TENSOR_ATTN_QKV, "bias", i), {n_embd + 2*n_embd_gqa}, 0);
if (layer.wqkv == nullptr) {
layer.wq = create_tensor(tn(LLM_TENSOR_ATTN_Q, "weight", i), {n_embd, n_embd_head_k * n_head}, 0);
layer.wk = create_tensor(tn(LLM_TENSOR_ATTN_K, "weight", i), {n_embd, n_embd_k_gqa}, 0);
layer.wv = create_tensor(tn(LLM_TENSOR_ATTN_V, "weight", i), {n_embd, n_embd_v_gqa}, 0);
layer.bq = create_tensor(tn(LLM_TENSOR_ATTN_Q, "bias", i), {n_embd}, llama_model_loader::TENSOR_NOT_REQUIRED);
layer.bk = create_tensor(tn(LLM_TENSOR_ATTN_K, "bias", i), {n_embd_gqa}, llama_model_loader::TENSOR_NOT_REQUIRED);
layer.bv = create_tensor(tn(LLM_TENSOR_ATTN_V, "bias", i), {n_embd_gqa}, llama_model_loader::TENSOR_NOT_REQUIRED);
}
layer.wo = create_tensor(tn(LLM_TENSOR_ATTN_OUT, "weight", i), {n_embd, n_embd}, 0);
src/llama.cpp
+28 -13
View File
@@ -4646,7 +4646,8 @@ struct llm_build_context {
ggml_row_size(kv_pe_compresseed->type, kv_lora_rank));
cb(k_pe, "k_pe", il);
kv_compressed = ggml_cont(ctx0, kv_compressed); // TODO: the CUDA backend does not support non-contiguous norm
// TODO: the CUDA backend used to not support non-cont. (RMS) norm, investigate removing ggml_cont
kv_compressed = ggml_cont(ctx0, kv_compressed);
kv_compressed = llm_build_norm(ctx0, kv_compressed, hparams,
model.layers[il].attn_kv_a_norm, NULL,
LLM_NORM_RMS, cb, il);
@@ -6500,7 +6501,8 @@ struct llm_build_context {
ggml_row_size(kv_pe_compresseed->type, kv_lora_rank));
cb(k_pe, "k_pe", il);
kv_compressed = ggml_cont(ctx0, kv_compressed); // TODO: the CUDA backend does not support non-contiguous norm
// TODO: the CUDA backend used to not support non-cont. (RMS) norm, investigate removing ggml_cont
kv_compressed = ggml_cont(ctx0, kv_compressed);
kv_compressed = llm_build_norm(ctx0, kv_compressed, hparams,
model.layers[il].attn_kv_a_norm, NULL,
LLM_NORM_RMS, cb, il);
@@ -7251,17 +7253,30 @@ struct llm_build_context {
struct ggml_tensor * Qcur = nullptr;
struct ggml_tensor * Kcur = nullptr;
struct ggml_tensor * Vcur = nullptr;
cur = llm_build_lora_mm(lctx, ctx0, model.layers[il].wqkv, cur);
cb(cur, "wqkv", il);
cur = ggml_add(ctx0, cur, model.layers[il].bqkv);
cb(cur, "bqkv", il);
Qcur = ggml_cont(ctx0, ggml_view_2d(ctx0, cur, n_embd, n_tokens, cur->nb[1], 0*sizeof(float)*(n_embd)));
Kcur = ggml_cont(ctx0, ggml_view_2d(ctx0, cur, n_embd_gqa, n_tokens, cur->nb[1], 1*sizeof(float)*(n_embd)));
Vcur = ggml_cont(ctx0, ggml_view_2d(ctx0, cur, n_embd_gqa, n_tokens, cur->nb[1], 1*sizeof(float)*(n_embd + n_embd_gqa)));
if (model.type == LLM_TYPE_1_5B || model.type == LLM_TYPE_4B || model.type == LLM_TYPE_9B) {
Qcur = llm_build_lora_mm(lctx, ctx0, model.layers[il].wq, cur);
if (model.layers[il].bq) {
Qcur = ggml_add(ctx0, Qcur, model.layers[il].bq);
}
Kcur = llm_build_lora_mm(lctx, ctx0, model.layers[il].wk, cur);
if (model.layers[il].bk) {
Kcur = ggml_add(ctx0, Kcur, model.layers[il].bk);
}
Vcur = llm_build_lora_mm(lctx, ctx0, model.layers[il].wv, cur);
if (model.layers[il].bv) {
Vcur = ggml_add(ctx0, Vcur, model.layers[il].bv);
}
} else {
cur = llm_build_lora_mm(lctx, ctx0, model.layers[il].wqkv, cur);
cb(cur, "wqkv", il);
if (model.layers[il].bqkv) {
cur = ggml_add(ctx0, cur, model.layers[il].bqkv);
cb(cur, "bqkv", il);
}
Qcur = ggml_cont(ctx0, ggml_view_2d(ctx0, cur, n_embd, n_tokens, cur->nb[1], 0*sizeof(float)*(n_embd)));
Kcur = ggml_cont(ctx0, ggml_view_2d(ctx0, cur, n_embd_gqa, n_tokens, cur->nb[1], 1*sizeof(float)*(n_embd)));
Vcur = ggml_cont(ctx0, ggml_view_2d(ctx0, cur, n_embd_gqa, n_tokens, cur->nb[1], 1*sizeof(float)*(n_embd + n_embd_gqa)));
}
cb(Qcur, "Qcur", il);
cb(Kcur, "Kcur", il);
cb(Vcur, "Vcur", il);