#!/bin/bash
#grab container image name
MODEL_NAME="Gemma-4-26B-A4B-noThink"
CONT_NAME="llama-$MODEL_NAME"

MODEL_FOLDER="/home/jared/.cache/huggingface/hub/models--unsloth--gemma-4-26B-A4B-it-GGUF"
MODEL="snapshots/3365c68df1a83799b846d05324ebfadbb8cc70b3/gemma-4-26B-A4B-it-MXFP4_MOE.gguf"
MODEL_MM="snapshots/3365c68df1a83799b846d05324ebfadbb8cc70b3/mmproj-F16.gguf"

source ../env/llama_env.sh
echo $IMAGkE
docker container create --name $CONT_NAME --network llms -p 10987:10987 -v $MODEL_FOLDER:/model --user $(id -u):$(id -g) --gpus=all --restart on-failure:3 \
	--health-cmd "curl -f http://$CONT_NAME:10987/v1/models || exit 1" \
	--health-interval 5s \
	--health-timeout 5s \
	--health-start-period 20s \
	$IMAGE \
	/llama.cpp/build/bin/llama-server \
	--model /model/$MODEL \
	--mm /model/$MODEL_MM \
	--alias $MODEL_NAME \
	-ctk q8_0 -ctv q8_0 \
	--reasoning-budget 0 \
	--host $CONT_NAME \
	--port 10987 --no-mmap -dio