ExternalVendorCode/LocalAI
1
0
Fork 0
mirror of https://github.com/mudler/LocalAI.git synced 2024-12-18 20:27:57 +00:00
Code Issues Actions 12 Packages Projects Releases Wiki Activity

feat(llama.cpp): expose cache_type_k and cache_type_v for quant of kv cache (#4329)

Browse Source 
Signed-off-by: Ettore Di Giacinto <mudler@localai.io>
This commit is contained in:
Ettore Di Giacinto 2024-12-06 10:23:59 +01:00 committed by GitHub
parent 88737e1d76
commit d4c1746c7d
No known key found for this signature in database
GPG Key ID: B5690EEEBB952194
4 changed files with 15 additions and 2 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

3
backend/backend.proto
View File

@ -242,6 +242,9 @@ message ModelOptions {
repeated float LoraScales = 61; repeated float LoraScales = 61;
repeated string Options = 62; repeated string Options = 62;
string CacheTypeKey = 63;
string CacheTypeValue = 64;
} }
message Result { message Result {

6
backend/cpp/llama/grpc-server.cpp
View File

@ -2241,6 +2241,12 @@ static void params_parse(const backend::ModelOptions* request,
} }
// params.model_alias ?? // params.model_alias ??
params.model_alias = request->modelfile(); params.model_alias = request->modelfile();
if (!request->cachetypekey().empty()) {
params.cache_type_k = request->cachetypekey();
}
if (!request->cachetypevalue().empty()) {
params.cache_type_v = request->cachetypevalue();
}
params.n_ctx = request->contextsize(); params.n_ctx = request->contextsize();
//params.memory_f16 = request->f16memory(); //params.memory_f16 = request->f16memory();
params.cpuparams.n_threads = request->threads(); params.cpuparams.n_threads = request->threads();

2
core/backend/options.go
View File

@ -151,6 +151,8 @@ func grpcModelOpts(c config.BackendConfig) *pb.ModelOptions {
TensorParallelSize: int32(c.TensorParallelSize), TensorParallelSize: int32(c.TensorParallelSize),
MMProj: c.MMProj, MMProj: c.MMProj,
FlashAttention: c.FlashAttention, FlashAttention: c.FlashAttention,
CacheTypeKey: c.CacheTypeK,
CacheTypeValue: c.CacheTypeV,
NoKVOffload: c.NoKVOffloading, NoKVOffload: c.NoKVOffloading,
YarnExtFactor: c.YarnExtFactor, YarnExtFactor: c.YarnExtFactor,
YarnAttnFactor: c.YarnAttnFactor, YarnAttnFactor: c.YarnAttnFactor,

6
core/config/backend_config.go
View File

@ -155,8 +155,10 @@ type LLMConfig struct {
TensorParallelSize int `yaml:"tensor_parallel_size"` // vLLM TensorParallelSize int `yaml:"tensor_parallel_size"` // vLLM
MMProj string `yaml:"mmproj"` MMProj string `yaml:"mmproj"`
FlashAttention bool `yaml:"flash_attention"` FlashAttention bool `yaml:"flash_attention"`
NoKVOffloading bool `yaml:"no_kv_offloading"` NoKVOffloading bool `yaml:"no_kv_offloading"`
CacheTypeK string `yaml:"cache_type_k"`
CacheTypeV string `yaml:"cache_type_v"`
RopeScaling string `yaml:"rope_scaling"` RopeScaling string `yaml:"rope_scaling"`
ModelType string `yaml:"type"` ModelType string `yaml:"type"`