From be61a81758ecd7990e0bd4cdd77093ffc4ecb1ca Mon Sep 17 00:00:00 2001
From: Purinda Gunasekara <purinda@gmail.com>
Date: Tue, 21 Nov 2023 02:52:52 +1100
Subject: [PATCH] main-gpu argument is not getting passed to llamacpp, fixed.
 (#1192)

---
 llm/llama.go | 2 ++
 1 file changed, 2 insertions(+)

diff --git a/llm/llama.go b/llm/llama.go
index 4481e97de..4eab751d6 100644
--- a/llm/llama.go
+++ b/llm/llama.go
@@ -339,6 +339,7 @@ func newLlama(model string, adapters []string, runners []ModelRunner, numLayers
 		"--model", model,
 		"--ctx-size", fmt.Sprintf("%d", opts.NumCtx),
 		"--batch-size", fmt.Sprintf("%d", opts.NumBatch),
+		"--main-gpu", fmt.Sprintf("%d", opts.MainGPU),
 		"--n-gpu-layers", fmt.Sprintf("%d", numGPU),
 		"--embedding",
 	}
@@ -544,6 +545,7 @@ func (llm *llama) Predict(ctx context.Context, prevContext []int, prompt string,
 		"stream":            true,
 		"n_predict":         llm.NumPredict,
 		"n_keep":            llm.NumKeep,
+		"main_gpu":          llm.MainGPU,
 		"temperature":       llm.Temperature,
 		"top_k":             llm.TopK,
 		"top_p":             llm.TopP,