diff options
| author | Georgi Gerganov <ggerganov@gmail.com> | 2024-05-23 10:00:44 +0300 |
|---|---|---|
| committer | GitHub <noreply@github.com> | 2024-05-23 10:00:44 +0300 |
| commit | d48c88cbd563b6cf0ce972e2f56796896e240736 (patch) | |
| tree | 14574b8ddc14145b093f6ab531f1e2968bbd6d85 /examples/finetune/finetune.cpp | |
| parent | e84b71c2c6da6e69c8f815168ea836f9716a325e (diff) | |
ggml : remove ggml_flash_attn and ggml_flash_ff (#7463)
ggml-ci
Diffstat (limited to 'examples/finetune/finetune.cpp')
| -rw-r--r-- | examples/finetune/finetune.cpp | 3 |
1 files changed, 2 insertions, 1 deletions
diff --git a/examples/finetune/finetune.cpp b/examples/finetune/finetune.cpp index 992426c1..22425730 100644 --- a/examples/finetune/finetune.cpp +++ b/examples/finetune/finetune.cpp @@ -643,7 +643,8 @@ static struct ggml_tensor * llama_build_lora_finetune_graphs( struct ggml_tensor * t15 = ggml_permute (ctx, t12, 0, 3, 1, 2); set_name(t15, "t15"); assert_shape_4d(t15, N, n_embd_head, n_head_kv, n_batch); struct ggml_tensor * t16; if (enable_flash_attn) { - t16 = ggml_flash_attn(ctx, t13, t14, t15, true); set_name(t16, "t16"); assert_shape_4d(t16, n_embd_head, N, n_head, n_batch); + GGML_ASSERT(false && "TODO: ggml_flash_attn_ext() not yet supported"); + //t16 = ggml_flash_attn(ctx, t13, t14, t15, true); set_name(t16, "t16"); assert_shape_4d(t16, n_embd_head, N, n_head, n_batch); } else { struct ggml_tensor * t16_0 = ggml_mul_mat (ctx, t14, t13); set_name(t16_0, "t16_0"); assert_shape_4d(t16_0, N, N, n_head, n_batch); struct ggml_tensor * t16_1 = ggml_scale_inplace (ctx, t16_0, kv_scale); set_name(t16_1, "t16_1"); assert_shape_4d(t16_1, N, N, n_head, n_batch); |