git/ik_llama.cpp
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

Mark some multi-prediction tensors as not required. (#814)

Browse Source
This commit is contained in:
Downtown-Case 2025-10-01 13:37:31 -05:00 committed by GitHub
parent e94d1a92a5
commit 6051ba25ee
No known key found for this signature in database
GPG Key ID: B5690EEEBB952194
1 changed files with 3 additions and 3 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

6
src/llama.cpp
View File

@ -6741,7 +6741,7 @@ static bool llm_load_tensors(
layer.nextn.embed_tokens = create_tensor(ctx_for_layer(final_layer),
tn(LLM_TENSOR_NEXTN_EMBED_TOKENS, "weight", final_layer),
{ n_embd, n_vocab },
flags);
flags | TENSOR_NOT_REQUIRED);
// ENORM, HNORM: [embd]
layer.nextn.enorm = create_tensor(ctx_for_layer(final_layer),
tn(LLM_TENSOR_NEXTN_ENORM, "weight", final_layer),
@ -6755,12 +6755,12 @@ static bool llm_load_tensors(
layer.nextn.shared_head_head = create_tensor(ctx_for_layer(final_layer),
tn(LLM_TENSOR_NEXTN_SHARED_HEAD_HEAD, "weight", final_layer),
{ n_embd, n_vocab },
flags);
flags | TENSOR_NOT_REQUIRED);
// SHARED_HEAD_NORM: [embd]
layer.nextn.shared_head_norm = create_tensor(ctx_for_layer(final_layer),
tn(LLM_TENSOR_NEXTN_SHARED_HEAD_NORM, "weight", final_layer),
{ n_embd },
flags);
flags | TENSOR_NOT_REQUIRED);
}
}
}