From: Johannes Gäßler <redacted>
Date: Wed, 22 May 2024 08:24:29 +0000 (+0200)
Subject: CUDA: remove incorrect precision check (llama/7454)
X-Git-Tag: upstream/0.0.1642~657
X-Git-Url: https://git.djapps.eu/?a=commitdiff_plain;h=843f90a9e2a3bf0cdd983fc0761e8bf2664ef59a;p=pkg%2Fggml%2Fsources%2Fggml

CUDA: remove incorrect precision check (llama/7454)
---

diff --git a/src/ggml-cuda/fattn-tile-f32.cu b/src/ggml-cuda/fattn-tile-f32.cu
index 130e7cbd..54db765e 100644
--- a/src/ggml-cuda/fattn-tile-f32.cu
+++ b/src/ggml-cuda/fattn-tile-f32.cu
@@ -286,9 +286,6 @@ void ggml_cuda_flash_attn_ext_tile_f32(ggml_backend_cuda_context & ctx, ggml_ten
     const ggml_tensor * KQV = dst;
     const ggml_tensor * Q   = dst->src[0];
 
-    const int32_t precision = KQV->op_params[2];
-    GGML_ASSERT(precision == GGML_PREC_DEFAULT);
-
     if (Q->ne[1] <= 16) {
         constexpr int cols_per_block = 16;
         constexpr int parallel_blocks = 4;