happyz
/
llama.cpp
mirror of https://github.com/ggerganov/llama.cpp.git
1
0
Fork 0
Code Issues Packages Projects Releases Wiki Activity

fix: Use per-layer n_embd_k/v_s calls for mamba (1) layers 

Branch: HybridRecurrentCache

Signed-off-by: Gabe Goodhart <ghart@us.ibm.com>
This commit is contained in:
Gabe Goodhart 2025-06-10 16:30:49 -06:00
parent f6d5f055c6
commit 833dfb54ae
1 changed files with 2 additions and 2 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

4
src/llama-model.cpp
View File

@ -9194,11 +9194,11 @@ struct llm_build_mamba : public llm_graph_context {
// (ab)using the KV cache to store the states
ggml_tensor * conv = build_recurrent_state(
gf, conv_states_all, state_copy,
hparams.n_embd_k_s(), n_seqs);
hparams.n_embd_k_s(il), n_seqs);
conv = ggml_reshape_3d(ctx0, conv, d_conv - 1, d_inner, n_seqs);
ggml_tensor * ssm = build_recurrent_state(
gf, ssm_states_all, state_copy,
hparams.n_embd_v_s(), n_seqs);
hparams.n_embd_v_s(il), n_seqs);
ssm = ggml_reshape_3d(ctx0, ssm, d_state, d_inner, n_seqs);
// {n_embd, n_tokens} => {n_embd, n_seq_tokens, n_seqs}