diff options
Diffstat (limited to 'src')
| -rw-r--r-- | src/llama.cpp | 6 |
1 files changed, 5 insertions, 1 deletions
diff --git a/src/llama.cpp b/src/llama.cpp index 5ac44055..e246dec9 100644 --- a/src/llama.cpp +++ b/src/llama.cpp @@ -8777,7 +8777,11 @@ 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, hparams.attn_soft_cap ? hparams.f_attn_logit_softcapping : 0.0f); - if (model.arch == LLM_ARCH_PHI2 || model.arch == LLM_ARCH_PHI3 || model.arch == LLM_ARCH_GPTNEOX) { + // Some models produced NaNs/gibberish when FA is computed with f16 precision on CUDA + // For DeepSeek-2, it is perfectly fine with fp16 for PP, but I get gibberish when uding fp16 for TG. + // Not sure if it is really a matter of insufficient precision, or I have made a mistake in the fattn-vec-f16 kernel. + if (model.arch == LLM_ARCH_PHI2 || model.arch == LLM_ARCH_PHI3 || model.arch == LLM_ARCH_GPTNEOX || + (model.arch == LLM_ARCH_DEEPSEEK2 && q->ne[1] <= 8)) { ggml_flash_attn_ext_set_prec(cur, GGML_PREC_F32); } //ggml_flash_attn_ext_set_prec(cur, GGML_PREC_F32); |