Public/koboldcpp
1
0
Fork 0
mirror of https://github.com/LostRuins/koboldcpp.git synced 2026-09-20 01:31:42 +02:00
Code Issues Packages Projects Releases Wiki Activity

Merge commit '8f8f2274ee3601fecf6e2d57b52f701c81bede21' into concedo_experimental

Browse Source 
# Conflicts:
#	.devops/rocm.Dockerfile
#	.github/workflows/build.yml
#	.github/workflows/release.yml
#	CMakeLists.txt
#	examples/simple/simple.cpp
#	ggml/src/ggml-cann/common.h
#	ggml/src/ggml-cann/ggml-cann.cpp
#	ggml/src/ggml-opencl/kernels/tsembd.cl
#	ggml/src/ggml-sycl/binbcast.cpp
#	ggml/src/ggml-sycl/binbcast.hpp
#	ggml/src/ggml-sycl/ggml-sycl.cpp
#	ggml/src/ggml-sycl/tsembd.cpp
#	ggml/src/ggml-zdnn/ggml-zdnn.cpp
#	src/llama-model.cpp
#	tools/batched-bench/CMakeLists.txt
#	tools/cvector-generator/CMakeLists.txt
#	tools/export-lora/CMakeLists.txt
#	tools/gguf-split/CMakeLists.txt
#	tools/imatrix/CMakeLists.txt
#	tools/llama-bench/CMakeLists.txt
#	tools/llama-bench/llama-bench.cpp
#	tools/main/CMakeLists.txt
#	tools/main/README.md
#	tools/mtmd/CMakeLists.txt
#	tools/perplexity/CMakeLists.txt
#	tools/perplexity/perplexity.cpp
#	tools/quantize/CMakeLists.txt
#	tools/rpc/rpc-server.cpp
#	tools/run/CMakeLists.txt
#	tools/run/run.cpp
#	tools/tokenize/CMakeLists.txt
#	tools/tts/CMakeLists.txt
This commit is contained in:
Concedo
2025-09-21 08:58:23 +08:00
parent 04498a345a 8f8f2274ee
commit 3e72aaff5b
41 changed files with 1998 additions and 1463 deletions
Download Patch File Download Diff File Expand all files Collapse all files
common/arg.cpp
+5 -5
View File
@@ -1706,7 +1706,7 @@ common_params_context common_params_parser_init(common_params & params, llama_ex
[](common_params & params, const std::string & value) {
params.system_prompt = value;
}
).set_examples({LLAMA_EXAMPLE_MAIN}));
).set_examples({LLAMA_EXAMPLE_MAIN, LLAMA_EXAMPLE_DIFFUSION}));
add_opt(common_arg(
{"--no-perf"},
string_format("disable internal libllama performance timings (default: %s)", params.no_perf ? "true" : "false"),
@@ -2550,7 +2550,7 @@ common_params_context common_params_parser_init(common_params & params, llama_ex
{"--cpu-moe", "-cmoe"},
"keep all Mixture of Experts (MoE) weights in the CPU",
[](common_params & params) {
params.tensor_buft_overrides.push_back({"\\.ffn_(up|down|gate)_exps", ggml_backend_cpu_buffer_type()});
params.tensor_buft_overrides.push_back(llm_ffn_exps_cpu_override());
}
).set_env("LLAMA_ARG_CPU_MOE"));
add_opt(common_arg(
@@ -2563,7 +2563,7 @@ common_params_context common_params_parser_init(common_params & params, llama_ex
for (int i = 0; i < value; ++i) {
// keep strings alive and avoid leaking memory by storing them in a static vector
static std::list<std::string> buft_overrides;
buft_overrides.push_back(string_format("blk\\.%d\\.ffn_(up|down|gate)_exps", i));
buft_overrides.push_back(llm_ffn_exps_block_regex(i));
params.tensor_buft_overrides.push_back({buft_overrides.back().c_str(), ggml_backend_cpu_buffer_type()});
}
}
@@ -2572,7 +2572,7 @@ common_params_context common_params_parser_init(common_params & params, llama_ex
{"--cpu-moe-draft", "-cmoed"},
"keep all Mixture of Experts (MoE) weights in the CPU for the draft model",
[](common_params & params) {
params.speculative.tensor_buft_overrides.push_back({"\\.ffn_(up|down|gate)_exps", ggml_backend_cpu_buffer_type()});
params.speculative.tensor_buft_overrides.push_back(llm_ffn_exps_cpu_override());
}
).set_examples({LLAMA_EXAMPLE_SPECULATIVE, LLAMA_EXAMPLE_SERVER}).set_env("LLAMA_ARG_CPU_MOE_DRAFT"));
add_opt(common_arg(
@@ -2584,7 +2584,7 @@ common_params_context common_params_parser_init(common_params & params, llama_ex
}
for (int i = 0; i < value; ++i) {
static std::list<std::string> buft_overrides_draft;
buft_overrides_draft.push_back(string_format("blk\\.%d\\.ffn_(up|down|gate)_exps", i));
buft_overrides_draft.push_back(llm_ffn_exps_block_regex(i));
params.speculative.tensor_buft_overrides.push_back({buft_overrides_draft.back().c_str(), ggml_backend_cpu_buffer_type()});
}
}
common/common.h
+17 -3
View File
@@ -284,9 +284,9 @@ struct common_params {
float rope_freq_base = 0.0f; // RoPE base frequency
float rope_freq_scale = 0.0f; // RoPE frequency scaling factor
float yarn_ext_factor = -1.0f; // YaRN extrapolation mix factor
float yarn_attn_factor = 1.0f; // YaRN magnitude scaling factor
float yarn_beta_fast = 32.0f; // YaRN low correction dim
float yarn_beta_slow = 1.0f; // YaRN high correction dim
float yarn_attn_factor = -1.0f; // YaRN magnitude scaling factor
float yarn_beta_fast = -1.0f; // YaRN low correction dim
float yarn_beta_slow = -1.0f; // YaRN high correction dim
int32_t yarn_orig_ctx = 0; // YaRN original context length
// offload params
@@ -730,6 +730,20 @@ const char * const LLM_KV_SPLIT_TENSORS_COUNT = "split.tensors.count";
}
//
// MoE utils
//
const char * const LLM_FFN_EXPS_REGEX = "\\.ffn_(up|down|gate)_exps";
static std::string llm_ffn_exps_block_regex(int idx) {
return string_format("blk\\.%d%s", idx, LLM_FFN_EXPS_REGEX);
}
static llama_model_tensor_buft_override llm_ffn_exps_cpu_override() {
return { LLM_FFN_EXPS_REGEX, ggml_backend_cpu_buffer_type() };
}
//
// training utils
//
common/json-schema-to-grammar.cpp
+7 -6
View File
@@ -257,12 +257,13 @@ std::unordered_map<std::string, BuiltinRule> STRING_FORMAT_RULES = {
};
static bool is_reserved_name(const std::string & name) {
static std::unordered_set<std::string> RESERVED_NAMES;
if (RESERVED_NAMES.empty()) {
RESERVED_NAMES.insert("root");
for (const auto &p : PRIMITIVE_RULES) RESERVED_NAMES.insert(p.first);
for (const auto &p : STRING_FORMAT_RULES) RESERVED_NAMES.insert(p.first);
}
static const std::unordered_set<std::string> RESERVED_NAMES = [] {
std::unordered_set<std::string> s;
s.insert("root");
for (const auto & p : PRIMITIVE_RULES) s.insert(p.first);
for (const auto & p : STRING_FORMAT_RULES) s.insert(p.first);
return s;
}();
return RESERVED_NAMES.find(name) != RESERVED_NAMES.end();
}