CS348Project/llama.cpp
5
0
Fork 0
mirror of https://github.com/ggml-org/llama.cpp.git synced 2025-10-27 08:21:30 +00:00
Code Issues Packages Projects Releases Wiki Activity

model : avoid ggml_cont_3d for fused QKV weights (#15662)

Browse Source 
* model : avoid ggml_cont_3d for fused QKV weights

ggml-ci

* kv-cache : make cpy_k and cpy_v implementation more readable

ggml-ci

* cont : add comments

ggml-ci

* cont : minor fix [no ci]

* cont : one more fix

* cont : clarity

ggml-ci

* kv-cache : require contiguous heads of k_cur and v_cur

ggml-ci
This commit is contained in:
Georgi Gerganov
2025-09-08 10:25:33 +03:00
committed by GitHub
parent d413dca003
commit cf0e3ba150
3 changed files with 100 additions and 108 deletions
Download Patch File Download Diff File Expand all files Collapse all files

8
src/llama-kv-cache.h
View File

@@ -317,9 +317,17 @@ public:
ggml_tensor * get_v(ggml_context * ctx, int32_t il) const;
// store k_cur and v_cur in the cache based on the provided head location
// note: the heads in k_cur and v_cur should be layed out contiguously in memory
// - k_cur [n_embd_head_k, n_head_k, n_tokens]
// - k_idxs [n_tokens]
// - v_cur [n_embd_head_v, n_head_v, n_tokens]
// - v_idxs [n_tokens] or [n_tokens*n_embd_v_gqa] depending if V cache is transposed
ggml_tensor * cpy_k(ggml_context * ctx, ggml_tensor * k_cur, ggml_tensor * k_idxs, int32_t il) const;
ggml_tensor * cpy_v(ggml_context * ctx, ggml_tensor * v_cur, ggml_tensor * v_idxs, int32_t il) const;
// create destination indices for each head of the current batch for where it would be written in the KV cache
// the indices address the global KV cache (not per stream) - this is not relevant for the user of this API, but
// helps understand the implementation logic of cpy_k and cpy_v
ggml_tensor * build_input_k_idxs(ggml_context * ctx, const llama_ubatch & ubatch) const;
ggml_tensor * build_input_v_idxs(ggml_context * ctx, const llama_ubatch & ubatch) const;