vbatts/llama.cpp
1
0
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity Actions

llama : fix command-r inference when omitting outputs (#6367)

Browse source
This commit is contained in:
compilade 2024-03-28 08:05:54 -04:00 • committed by GitHub
parent 28cb9a09c4
commit 0308f5e3d7
No known key found for this signature in database
GPG key ID: B5690EEEBB952194
1 changed files with 3 additions and 2 deletions
Download patch file Download diff file Expand all files Collapse all files

1
llama.cpp
View file

@ -9154,6 +9154,7 @@ struct llm_build_context {
struct ggml_tensor * inp_out_ids = build_inp_out_ids(); struct ggml_tensor * inp_out_ids = build_inp_out_ids();
cur = ggml_get_rows(ctx0, cur, inp_out_ids); cur = ggml_get_rows(ctx0, cur, inp_out_ids);
inpL = ggml_get_rows(ctx0, inpL, inp_out_ids); inpL = ggml_get_rows(ctx0, inpL, inp_out_ids);
ffn_inp = ggml_get_rows(ctx0, ffn_inp, inp_out_ids);
} }
struct ggml_tensor * attn_out = cur; struct ggml_tensor * attn_out = cur;