From: Georgi Gerganov <redacted>
Date: Sun, 27 Aug 2023 15:55:41 +0000 (+0300)
Subject: llama : fix MPI threads (close #2827)
X-Git-Tag: gguf-v0.4.0~211
X-Git-Url: https://git.djapps.eu/?a=commitdiff_plain;h=c10704d01e21e3dbe4d6ca1026ebff85349dd239;p=pkg%2Fggml%2Fsources%2Fllama.cpp

llama : fix MPI threads (close #2827)
---

diff --git a/llama.cpp b/llama.cpp
index 0bb8fcd6..72d2d1de 100644
--- a/llama.cpp
+++ b/llama.cpp
@@ -2845,7 +2845,6 @@ static bool llama_eval_internal(
 
     GGML_ASSERT(n_tokens > 0);
     GGML_ASSERT(n_past >= 0);
-    GGML_ASSERT(n_threads > 0);
     // TODO: keep the values of n_batch and n_ctx
     // GGML_ASSERT(n_tokens <= n_batch);
     // GGML_ASSERT(n_past + n_tokens <= n_ctx);
@@ -2856,6 +2855,8 @@ static bool llama_eval_internal(
     ggml_mpi_eval_init(lctx.ctx_mpi, &n_tokens, &n_past, &n_threads);
 #endif
 
+    GGML_ASSERT(n_threads > 0);
+
     const int N = n_tokens;
 
     const auto & model   = lctx.model;