From: anon998 <redacted>
Date: Sun, 25 Jun 2023 08:48:36 +0000 (+0000)
Subject: fix server sampling: top k sampler first (#1977)
X-Git-Tag: gguf-v0.4.0~568
X-Git-Url: https://git.djapps.eu/?a=commitdiff_plain;h=c2a08f87b8d180115d04b8688f383d1b2761b16d;p=pkg%2Fggml%2Fsources%2Fllama.cpp

fix server sampling: top k sampler first (#1977)

Co-authored-by: anon <redacted>
---

diff --git a/examples/server/server.cpp b/examples/server/server.cpp
index de22d301..79df5e84 100644
--- a/examples/server/server.cpp
+++ b/examples/server/server.cpp
@@ -325,10 +325,10 @@ struct llama_server_context {
                     id = llama_sample_token_mirostat_v2(ctx, &candidates_p, mirostat_tau, mirostat_eta, &mirostat_mu);
                 } else {
                     // Temperature sampling
+                    llama_sample_top_k(ctx, &candidates_p, top_k, 1);
                     llama_sample_tail_free(ctx, &candidates_p, tfs_z, 1);
                     llama_sample_typical(ctx, &candidates_p, typical_p, 1);
                     llama_sample_top_p(ctx, &candidates_p, top_p, 1);
-                    llama_sample_top_k(ctx, &candidates_p, top_k, 1);
                     llama_sample_temperature(ctx, &candidates_p, temp);
                     id = llama_sample_token(ctx, &candidates_p);
                 }