|
1 | 1 | diff --git a/src/models/mamba-base.cpp b/src/models/mamba-base.cpp |
2 | | -index 03ee3805b..9deff6629 100644 |
| 2 | +index 03ee3805b..370de0f01 100644 |
3 | 3 | --- a/src/models/mamba-base.cpp |
4 | 4 | +++ b/src/models/mamba-base.cpp |
5 | | -@@ -103,7 +103,7 @@ ggml_tensor * llm_build_mamba_base::build_mamba_layer(llm_graph_input_rs * inp, |
| 5 | +@@ -100,6 +100,8 @@ ggml_tensor * llm_build_mamba_base::build_mamba_layer(llm_graph_input_rs * inp, |
| 6 | + dt = build_norm(dt, layer.ssm_dt_norm, NULL, LLM_NORM_RMS, il); |
| 7 | + B = build_norm(B, layer.ssm_b_norm, NULL, LLM_NORM_RMS, il); |
| 8 | + C = build_norm(C, layer.ssm_c_norm, NULL, LLM_NORM_RMS, il); |
| 9 | ++ } else { |
| 10 | ++ dt = ggml_cont(ctx0, dt); |
6 | 11 | } |
7 | 12 |
|
8 | 13 | // {dt_rank, d_inner} @ {dt_rank, n_seq_tokens, n_seqs} => {d_inner, n_seq_tokens, n_seqs} |
9 | | -- dt = build_lora_mm(layer.ssm_dt, dt); |
10 | | -+ dt = build_lora_mm(layer.ssm_dt, ggml_cont(ctx0, dt)); |
11 | | - dt = ggml_add(ctx0, dt, layer.ssm_dt_b); |
12 | | - |
13 | | - cur = x; |
0 commit comments