vrr/koboldcpp
2
0
Fork 0
mirror of https://github.com/LostRuins/koboldcpp.git synced 2026-05-06 08:01:27 +00:00
Code Issues Projects Releases Packages Wiki Activity Actions

server : support unified cache across slots (#16736)

Browse source 
* server : support unified context across slots

* cont : fix speculative decoding initialization

* context : fix n_ctx_per_seq computation

* server : purge slots one by one

* tests : add unified cache server tests

* llama : update per-seq context computation

* test-thread-safety : handle tiny training context of the input model

* server : fix server_tokens clear()

* server : use 4 slots + unified KV by default

* llama : add note about context size queries

* cont : update todos [no ci]

* context : do not cap the size of the context

* tests : adjust parameters to be CI friendlier

* context : add warning
This commit is contained in:
Georgi Gerganov 2025-11-02 18:14:04 +02:00 • committed by GitHub
parent 87c9efc3b2
commit cd5e3b5754
No known key found for this signature in database
GPG key ID: B5690EEEBB952194
12 changed files with 163 additions and 48 deletions
Download patch file Download diff file Expand all files Collapse all files

14
src/llama-model.cpp
View file

@ -6712,14 +6712,14 @@ float llama_model::get_rope_freq_scale(const llama_cparams & cparams, int il) co
}
ggml_tensor * llama_model::get_rope_factors(const llama_cparams & cparams, int il) const {
const uint32_t n_ctx_per_seq = cparams.n_ctx / cparams.n_seq_max;
const uint32_t n_ctx_seq = cparams.n_ctx_seq;
// choose long/short freq factors based on the context size
if (layers[il].rope_freqs != nullptr) {
return layers[il].rope_freqs;
}
if (n_ctx_per_seq > hparams.n_ctx_orig_yarn) {
if (n_ctx_seq > hparams.n_ctx_orig_yarn) {
return layers[il].rope_long;
}
@ -6795,12 +6795,6 @@ llama_memory_i * llama_model::create_memory(const llama_memory_params & params,
/* filter_attn */ std::move(filter_attn),
/* filter_recr */ std::move(filter_recr));
} else {
uint32_t n_ctx_per_stream = cparams.n_ctx;
if (!cparams.kv_unified) {
n_ctx_per_stream = (cparams.n_ctx + cparams.n_seq_max - 1)/cparams.n_seq_max;
}
llama_memory_i::layer_reuse_cb reuse = nullptr;
if (arch == LLM_ARCH_GEMMA3N) {
@ -6824,7 +6818,7 @@ llama_memory_i * llama_model::create_memory(const llama_memory_params & params,
cparams.offload_kqv,
params.swa_full,
cparams.kv_unified,
n_ctx_per_stream,
cparams.n_ctx_seq,
cparams.n_seq_max,
cparams.n_ubatch,
1,
@ -6840,7 +6834,7 @@ llama_memory_i * llama_model::create_memory(const llama_memory_params & params,
!cparams.flash_attn,
cparams.offload_kqv,
cparams.kv_unified,
n_ctx_per_stream,
cparams.n_ctx_seq,
cparams.n_seq_max,
1,
hparams.n_swa,