tqcq/llama.cpp
0
0
Fork 0
mirror of https://github.com/ggml-org/llama.cpp.git synced 2025-07-14 22:58:10 +00:00
Code Issues Packages Projects Releases Wiki Activity

context : fix recurrent reserve

Browse Source 
ggml-ci
This commit is contained in:
Georgi Gerganov
2025-02-24 08:59:12 +02:00
parent 0699a44c83
commit a5a85a3bc0
2 changed files with 7 additions and 0 deletions
Download Patch File Download Diff File Expand all files Collapse all files

5
src/llama-context.cpp
View File

@ -3883,6 +3883,11 @@ llama_context_recurrent::llama_context_recurrent(
llama_context_recurrent::~llama_context_recurrent() = default;
void llama_context_recurrent::reserve() {
// simulate full KV cache
kv_self.n = kv_self.size;
LLAMA_LOG_DEBUG("%s: kv_self.n = %u\n", __func__, kv_self.n);
// TODO: implement recurrent-specific reserve logic
llama_context::reserve();
}

2
src/llama-context.h
View File

@ -447,6 +447,7 @@ private:
ggml_tensor * self_k_shift; // I32 [kv_size]
} inp;
protected:
//
// graph
//
@ -570,6 +571,7 @@ private:
ggml_tensor * s_mask; // F32 [1, n_kv]
} inp;
protected:
//
// graph
//