From 024335665074ddfa7c6975e26c6587494cf22a37 Mon Sep 17 00:00:00 2001 From: Kawrakow Date: Mon, 24 Nov 2025 09:13:46 +0100 Subject: [PATCH] Fix q6_0 dequantize (#1005) Co-authored-by: Iwan Kawrakow --- ggml/src/ggml-cuda/dequantize.cuh | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/ggml/src/ggml-cuda/dequantize.cuh b/ggml/src/ggml-cuda/dequantize.cuh index bd76dc2f..52649f27 100644 --- a/ggml/src/ggml-cuda/dequantize.cuh +++ b/ggml/src/ggml-cuda/dequantize.cuh @@ -91,7 +91,7 @@ static __device__ __forceinline__ void dequantize_q6_0(const void * vx, const in const dfloat d = x[ib].d; - const uint8_t h = x[ib].qh[iqs%8] >> 2*(iqs/8); + const uint8_t h = x[ib].qh[iqs%8] >> 4*(iqs/8); v.x = ((x[ib].qs[iqs] & 0xf) | ((h & 0x3) << 4)); v.y = ((x[ib].qs[iqs] >> 4) | ((h & 0xc) << 2));