Cleanup

Author: Lorenzo Toniazzi

tests/test_lora_conversion_and_inference.sh

@@ -19,6 +19,7 @@ run_conversion_and_inference_lora() {
     local size_matrix=$2
     local model_size_mb=$3
 
+    # Convert safetensors to gguf
     echo "Running convert_hf_to_gguf.py for $model_name with size $size_matrix..."
     python convert_hf_to_gguf.py reduce-llms-for-testing/$model_name/size=$size_matrix/base \
     --outtype f32
@@ -28,6 +29,13 @@ run_conversion_and_inference_lora() {
     --base reduce-llms-for-testing/$model_name/size=$size_matrix/base \
     --outtype f32
 
+    echo "Running llama-export-lora with lora for $model_name with size $size_matrix and model size $model_size_mb..."
+    llama-export-lora \
+    -m reduce-llms-for-testing/$model_name/size=$size_matrix/base/Base-$model_size_mb-F32.gguf \
+    -o reduce-llms-for-testing/$model_name/size=$size_matrix/base/Base-$model_size_mb-F32-lora-merged.gguf \
+    --lora reduce-llms-for-testing/$model_name/size=$size_matrix/lora/Lora-F32-LoRA.gguf
+
+    # Run inference
     echo "Running llama-cli without lora for $model_name with size $size_matrix and model size $model_size_mb..."
     OUTPUT_BASE=$(llama-cli -m reduce-llms-for-testing/$model_name/size=$size_matrix/base/Base-$model_size_mb-F32.gguf \
     -p "<bos>When forty winters shall besiege" -n 50 --seed 42)
@@ -37,24 +45,17 @@ run_conversion_and_inference_lora() {
     --lora reduce-llms-for-testing/$model_name/size=$size_matrix/lora/Lora-F32-LoRA.gguf \
     -p "<bos>I see a little silhouetto" -n 50 --seed 42)
 
-    # TODO add merge lora with lora-export and check
-    echo "Running llama-export-lora with lora for $model_name with size $size_matrix and model size $model_size_mb..."
-    llama-export-lora \
-    -m reduce-llms-for-testing/$model_name/size=$size_matrix/base/Base-$model_size_mb-F32.gguf \
-    -o reduce-llms-for-testing/$model_name/size=$size_matrix/base/Base-$model_size_mb-F32-lora-merged.gguf \
-    --lora reduce-llms-for-testing/$model_name/size=$size_matrix/lora/Lora-F32-LoRA.gguf \
-
     echo "Running llama-cli with exported lora for $model_name with size $size_matrix and model size $model_size_mb..."
     OUTPUT_LORA_MERGED=$(llama-cli -m reduce-llms-for-testing/$model_name/size=$size_matrix/base/Base-$model_size_mb-F32-lora-merged.gguf \
     -p "<bos>I see a little silhouetto" -n 50 --seed 42)
 
     # Store the results in the regular array
     results+=("
-    \n\n\n\033[1mResults for $model_name with size $size_matrix and model size $model_size_mb:\033[0m
-    \n  • \033[32mBase:\n $OUTPUT_BASE
-    \n  • \033[34mLora hot:\n $OUTPUT_LORA_HOT
-    \n  • \033[36mLora merged:\n $OUTPUT_LORA_MERGED
-    \n\n\n \033[0m
+    \n\n\033[1mResults for $model_name with size $size_matrix and model size $model_size_mb:\033[0m
+    \n  • \033[32mBase:\n$OUTPUT_BASE
+    \n  • \033[34mLora hot:\n$OUTPUT_LORA_HOT
+    \n  • \033[36mLora merged:\n$OUTPUT_LORA_MERGED
+    \n\n \033[0m
     ")
 
     echo "All steps completed for $model_name with size $size_matrix and model size $model_size_mb!"