From 81d55d3e4d3e18404414900dd341438aad329656 Mon Sep 17 00:00:00 2001
From: Daniel Hiltgen <dhiltgen@users.noreply.github.com>
Date: Mon, 18 Nov 2024 11:48:13 -0800
Subject: [PATCH] fix index out of range on zero layer metal load (#7696)

If the model doesn't fit any layers on metal, and we load zero layers
we would panic trying to look up the GPU size during scheduling ops
---
 llm/server.go | 4 +++-
 1 file changed, 3 insertions(+), 1 deletion(-)

diff --git a/llm/server.go b/llm/server.go
index 96815826..624acbf8 100644
--- a/llm/server.go
+++ b/llm/server.go
@@ -1092,7 +1092,9 @@ func (s *llmServer) EstimatedTotal() uint64 {
 func (s *llmServer) EstimatedVRAMByGPU(gpuID string) uint64 {
 	for i, gpu := range s.gpus {
 		if gpu.ID == gpuID {
-			return s.estimate.GPUSizes[i]
+			if i < len(s.estimate.GPUSizes) {
+				return s.estimate.GPUSizes[i]
+			}
 		}
 	}
 	return 0