New base prompt

Author: Lorenzo Toniazzi

tests/test_lora_conversion_and_inference.sh

@@ -18,7 +18,6 @@ results=()
 run_conversion_and_inference_lora() {
     local model_name=$1
     local size_matrix=$2
-    local bos_token=$3
 
     # Convert safetensors to gguf
     echo "Running convert_hf_to_gguf.py for $model_name with size $size_matrix..."
@@ -40,7 +39,7 @@ run_conversion_and_inference_lora() {
     # Run inference
     echo "Running llama-cli without lora for $model_name with size $size_matrix..."
     OUTPUT_BASE=$(llama-cli -m $MODELS_REPO/$model_name/size=$size_matrix/base/Base-F32.gguf \
-    -p "When forty winters shall besiege" -n 50 --seed 42)
+    -p "Look again at that dot." -n 50 --seed 42)
 
     echo "Running llama-cli with lora for $model_name with size $size_matrix..."
     OUTPUT_LORA_HOT=$(llama-cli -m $MODELS_REPO/$model_name/size=$size_matrix/base/Base-F32.gguf \
@@ -65,8 +64,8 @@ run_conversion_and_inference_lora() {
 
 # Array of parameters to iterate over
 declare -a params=(
-    "Gemma2ForCausalLM 64 <bos>"
-    "LlamaForCausalLM 64 <|begin_of_text|>"
+    "Gemma2ForCausalLM 64"
+    "LlamaForCausalLM 64"
 )
 
 # Loop through each set of parameters