tqcq/llama.cpp
0
0
Fork 0
mirror of https://github.com/ggml-org/llama.cpp.git synced 2025-06-26 19:55:04 +00:00
Code Issues Packages Projects Releases Wiki Activity

kv-cache : add SWA support (#13194)

Browse Source 
* kv-cache : prepare for SWA

ggml-ci

* kv-cache : initial iSWA implementation

ggml-ci

* kv-cache : rework error recovery logic

ggml-ci

* models : fix Phi-3 SWA parameters

ggml-ci

* model : adjust Granite to rope factor changes

ggml-ci

* server : check if context can do shifts

ggml-ci

* iswa : for now, always enable shifts (experiment)

ggml-ci

* kv-cache : simplify SWA logic

ggml-ci

* kv-cache : apply defrag when we fail to find slots for the batch

ggml-ci

* llama : update docs about llama_decode

ggml-ci

* kv-cache : update warning logs when no space for the batch is available

ggml-ci

* llama : add llama_kv_self_seq_pos_min()

* kv-cache : keep track of partial SWA computes and print warnings

* server : disallow use cases involving partial SWA context

ggml-ci

* llama : add param to control SWA cache size

ggml-ci

* minor : clean-up

ggml-ci
This commit is contained in:
Georgi Gerganov
2025-05-20 08:05:46 +03:00
committed by GitHub
parent f0adb80bf7
commit e298d2fbd0
15 changed files with 1426 additions and 650 deletions
Download Patch File Download Diff File Expand all files Collapse all files

8
common/arg.cpp
View File

@ -1445,6 +1445,14 @@ common_params_context common_params_parser_init(common_params & params, llama_ex
params.n_keep = value;
}
));
add_opt(common_arg(
{"--swa-full"},
string_format("use full-size SWA cache (default: %s)\n"
"[(more info)](https://github.com/ggml-org/llama.cpp/pull/13194#issuecomment-2868343055)", params.swa_full ? "true" : "false"),
[](common_params & params) {
params.swa_full = true;
}
));
add_opt(common_arg(
{"--no-context-shift"},
string_format("disables context shift on infinite text generation (default: %s)", params.ctx_shift ? "disabled" : "enabled"),

1
common/common.cpp
View File

@ -1136,6 +1136,7 @@ struct llama_context_params common_context_params_to_llama(const common_params &
cparams.flash_attn = params.flash_attn;
cparams.no_perf = params.no_perf;
cparams.op_offload = !params.no_op_offload;
cparams.swa_full = params.swa_full;
if (params.reranking) {
cparams.embeddings = true;

1
common/common.h
View File

@ -323,6 +323,7 @@ struct common_params {
bool flash_attn = false; // flash attention
bool no_perf = false; // disable performance metrics
bool ctx_shift = true; // context shift on inifinite text generation
bool swa_full = false; // use full-size SWA cache (https://github.com/ggml-org/llama.cpp/pull/13194#issuecomment-2868343055)
bool input_prefix_bos = false; // prefix BOS to user inputs, preceding input_prefix
bool use_mmap = true; // use mmap for faster loads