Lumpiasty/llama.cpp
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

models : appropriately flag noscan ssm_a tensors (#28121)

Browse Source
This commit is contained in:
Sigbjørn Skjæret
2026-09-01 19:59:15 +03:00
committed by GitHub
parent 73159c3039
commit f28493c783
3 changed files with 4 additions and 4 deletions
Download Patch File Download Diff File Expand all files Collapse all files
src/models/kimi-linear.cpp
+2 -2
View File
@@ -84,9 +84,9 @@ void llama_model_kimi_linear::load_arch_tensors(llama_model_loader &) {
layer.ssm_beta = create_tensor(tn(LLM_TENSOR_SSM_BETA, "weight", i), {n_embd, n_head}, 0);
// A_log - Shape in GGUF: [1, num_heads, 1, 1] (4D) or [1, num_heads] (2D after quantization) Note: -exp(A_log) is applied in convert_hf_to_gguf.py
layer.ssm_a = create_tensor(tn(LLM_TENSOR_SSM_A, i), {1, n_head, 1, 1}, TENSOR_NOT_REQUIRED);
layer.ssm_a = create_tensor(tn(LLM_TENSOR_SSM_A_NOSCAN, i), {1, n_head, 1, 1}, TENSOR_NOT_REQUIRED);
if (!layer.ssm_a) {
layer.ssm_a = create_tensor(tn(LLM_TENSOR_SSM_A, i), {1, n_head}, 0);
layer.ssm_a = create_tensor(tn(LLM_TENSOR_SSM_A_NOSCAN, i), {1, n_head}, 0);
}
// dt_bias - shape [n_embd_head_k_kda * n_head] = [4096]