GNU bug report logs - #69990
[PATCH] gnu: llama-cpp: mark as tunable per CPU architecture

Previous Next

Package: guix-patches;

Reported by: John Fremlin <john <at> fremlin.org>

Date: Sun, 24 Mar 2024 21:08:02 UTC

Severity: normal

Tags: patch

Done: Ludovic Courtès <ludo <at> gnu.org>

Bug is archived. No further changes may be made.

Full log

View this message in rfc822 format

From: John Fremlin <john <at> fremlin.org>
To: 69990 <at> debbugs.gnu.org
Cc: John Fremlin <john <at> fremlin.org>
Subject: [bug#69990] [PATCH] gnu: llama-cpp: mark as tunable per CPU architecture
Date: Sun, 24 Mar 2024 17:05:49 -0400

* gnu/packages/machine-learning.scm (llama-cpp): add tunable property.

Otherwise llama-cpp will use FMA4 instructions leading to SIGILL on even
recent AMD chips, that do not support them.

Change-Id: Ie69086cb9c3ed609d6bf73b0abf0c6f9df7220b1
---
 gnu/packages/machine-learning.scm | 1 +
 1 file changed, 1 insertion(+)

diff --git a/gnu/packages/machine-learning.scm b/gnu/packages/machine-learning.scm
index 003be7e3c6..79fa425de3 100644
--- a/gnu/packages/machine-learning.scm
+++ b/gnu/packages/machine-learning.scm
@@ -538,6 +538,7 @@ (define-public llama-cpp
          (file-name (git-file-name name version))
          (sha256
           (base32 "1ag1jash84hasz10h0piw72a8ginm8kzvhihbzzljz96gq2kjm88"))))
+      (properties '((tunable? . #true)))
       (build-system cmake-build-system)
       (arguments
        (list

base-commit: 28bc0e870b4d48b8e3e773382bb0e999df2e3611
-- 
2.41.0
This bug report was last modified 1 year and 48 days ago.

Previous Next

GNU bug tracking system
Copyright (C) 1999 Darren O. Benham, 1997,2003 nCipher Corporation Ltd, 1994-97 Ian Jackson.