From 55ac3b7aeaf52f19786ed96e885d89521fc0f6c8 Mon Sep 17 00:00:00 2001 From: Georgi Gerganov Date: Thu, 23 May 2024 15:28:14 +0300 Subject: ci : use Pythia models instead of OpenLlama (#7470) * ci : start using Pythia models over OpenLlama ggml-ci * ci : disable q2_k ppl tests * ci : use convert-hf-to-gguf.py * ci : update gg_get_model * ci : fix convert outfile name ggml-ci * llama : gptneox arch use F32 attn prec ggml-ci --- llama.cpp | 4 ++-- 1 file changed, 2 insertions(+), 2 deletions(-) (limited to 'llama.cpp') diff --git a/llama.cpp b/llama.cpp index 5ff186a5..1f9e10ee 100644 --- a/llama.cpp +++ b/llama.cpp @@ -6718,7 +6718,7 @@ static struct ggml_tensor * llm_build_kqv( cur = ggml_flash_attn_ext(ctx, q, k, v, kq_mask, kq_scale, hparams.f_max_alibi_bias); - if (model.arch == LLM_ARCH_PHI2 || model.arch == LLM_ARCH_PHI3) { + if (model.arch == LLM_ARCH_PHI2 || model.arch == LLM_ARCH_PHI3 || model.arch == LLM_ARCH_GPTNEOX) { ggml_flash_attn_ext_set_prec(cur, GGML_PREC_F32); } @@ -6727,7 +6727,7 @@ static struct ggml_tensor * llm_build_kqv( struct ggml_tensor * kq = ggml_mul_mat(ctx, k, q); cb(kq, "kq", il); - if (model.arch == LLM_ARCH_PHI2 || model.arch == LLM_ARCH_PHI3) { + if (model.arch == LLM_ARCH_PHI2 || model.arch == LLM_ARCH_PHI3 || model.arch == LLM_ARCH_GPTNEOX) { // for this arch, we need to perform the KQ multiplication with F32 precision, otherwise we get NaNs // ref: https://github.com/ggerganov/llama.cpp/pull/4490#issuecomment-1859055847 ggml_mul_mat_set_prec(kq, GGML_PREC_F32); -- cgit v1.2.3