From d59bd97065cd7ded6c4ecab54b1d5e0b1b11e318 Mon Sep 17 00:00:00 2001 From: Kawrakow <48489457+ikawrakow@users.noreply.github.com> Date: Tue, 5 Sep 2023 09:55:33 +0200 Subject: [PATCH] Guard against all weights in a super-block being zero (#3010) * Guard against all weights in a super-block being zero * Also guard against extremely small weights Closes #2982 --------- Co-authored-by: Iwan Kawrakow --- k_quants.c | 8 +++++++- 1 file changed, 7 insertions(+), 1 deletion(-) diff --git a/k_quants.c b/k_quants.c index 4accd2480..8742d4aee 100644 --- a/k_quants.c +++ b/k_quants.c @@ -83,7 +83,7 @@ static float make_qx_quants(int n, int nmax, const float * restrict x, int8_t * float ax = fabsf(x[i]); if (ax > amax) { amax = ax; max = x[i]; } } - if (!amax) { // all zero + if (amax < 1e-30f) { // all zero for (int i = 0; i < n; ++i) { L[i] = 0; } @@ -1086,6 +1086,12 @@ void quantize_row_q6_K_reference(const float * restrict x, block_q6_K * restrict } + if (!max_abs_scale) { + memset(&y[i], 0, sizeof(block_q6_K)); + y[i].d = ggml_fp32_to_fp16(0.f); + continue; + } + float iscale = -128.f/max_scale; y[i].d = ggml_fp32_to_fp16(1/iscale); for (int ib = 0; ib < QK_K/16; ++ib) {