thek0tyara/llama-cpp-turboquant
1
0
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity Actions 6

llama : remove KV cache defragmentation logic (#15473)

Browse source 
ggml-ci
This commit is contained in:
Georgi Gerganov 2025-08-22 12:22:13 +03:00 • committed by GitHub
parent ad5c975c2d
commit 9ebebef62f
No known key found for this signature in database
GPG key ID: B5690EEEBB952194
16 changed files with 32 additions and 440 deletions
Download patch file Download diff file Expand all files Collapse all files

6
common/arg.cpp
View file

@ -2254,9 +2254,11 @@ common_params_context common_params_parser_init(common_params & params, llama_ex
).set_examples({LLAMA_EXAMPLE_PERPLEXITY}));
add_opt(common_arg(
{"-dt", "--defrag-thold"}, "N",
string_format("KV cache defragmentation threshold (default: %.1f, < 0 - disabled)", (double)params.defrag_thold),
string_format("KV cache defragmentation threshold (DEPRECATED)"),
[](common_params & params, const std::string & value) {
params.defrag_thold = std::stof(value);
GGML_UNUSED(params);
GGML_UNUSED(value);
LOG_WRN("DEPRECATED: --defrag-thold is deprecated and no longer necessary to specify\n");
}
).set_env("LLAMA_ARG_DEFRAG_THOLD"));
add_opt(common_arg(

1
common/common.cpp
View file

@ -1152,7 +1152,6 @@ struct llama_context_params common_context_params_to_llama(const common_params &
cparams.yarn_orig_ctx = params.yarn_orig_ctx;
cparams.pooling_type = params.pooling_type;
cparams.attention_type = params.attention_type;
cparams.defrag_thold = params.defrag_thold;
cparams.cb_eval = params.cb_eval;
cparams.cb_eval_user_data = params.cb_eval_user_data;
cparams.offload_kqv = !params.no_kv_offload;

1
common/common.h
View file

@ -288,7 +288,6 @@ struct common_params {
float yarn_beta_fast = 32.0f; // YaRN low correction dim
float yarn_beta_slow = 1.0f; // YaRN high correction dim
int32_t yarn_orig_ctx = 0; // YaRN original context length
float defrag_thold = 0.1f; // KV cache defragmentation threshold
// offload params
std::vector<ggml_backend_dev_t> devices; // devices to use for offloading