Initial commit

Author: shintaro-ozaki

.gitignore

@@ -160,3 +160,6 @@ cython_debug/
 #  and can be added to the global gitignore or merged into this file.  For a more nuclear
 #  option (not recommended) you can uncomment the following to ignore the entire idea folder.
 #.idea/
+
+data/
+images/

scripts/102424_execute_inference.sh

@@ -1,51 +1,16 @@
 #!/bin/bash
-#SBATCH -p gpu_long
-#SBATCH -c 4
-#SBATCH -t 100:00:00
-#SBATCH --gres=gpu:a6000:1
-#SBATCH --account=is-nlp
-#SBATCH --job-name=inference-llama3.1-instruct
-#SBATCH -o logs/slurm-%x-%j.log
-#SBATCH --nodelist=elm66
+set -eu
 
-project=/cl/home2/shintaro/rag-notebook/shintaro
+project=$(pwd)
 source $project/.venv/bin/activate
 
-<< COMMENTOUT
-MODELS=("microsoft/Phi-3.5-mini-instruct" "axiong/PMC_LLaMA_13B" epfl-llm/meditron-70b meta-llama/Llama-3.1-70B meta-llama/Llama-2-70b-chat-hf meta-llama/Llama-3.1-8B
-task=("medqa" "medmcqa" "mmlu" "pubmed")
-COMMENTOUT
-
 inference_model=meta-llama/Llama-3.1-70B-Instruct
 quantize_type=none
 inference_max_length=2048
-
 task=medmcqa
-time python $project/src/102424_execute_inference.py \
-    --inference_model $inference_model \
-    --task $task \
-    --quantize_type $quantize_type \
-    --inference_max_length $inference_max_length
-
-task=mmlu
-time python $project/src/102424_execute_inference.py \
-    --inference_model $inference_model \
-    --task $task \
-    --quantize_type $quantize_type \
-    --inference_max_length $inference_max_length
-
-task=pubmedqa
-time python $project/src/102424_execute_inference.py \
-    --inference_model $inference_model \
-    --task $task \
-    --quantize_type $quantize_type \
-    --inference_max_length $inference_max_length
 
-task=medqa
 time python $project/src/102424_execute_inference.py \
     --inference_model $inference_model \
     --task $task \
     --quantize_type $quantize_type \
     --inference_max_length $inference_max_length
-
-echo "Done"

scripts/102424_execute_inference_llama2.sh

@@ -1,28 +1,13 @@
 #!/bin/bash
-#SBATCH -p gpu_long
-#SBATCH -c 4
-#SBATCH -t 100:00:00
-#SBATCH --gres=gpu:6000:1
-#SBATCH --account=is-nlp
-#SBATCH --job-name=inference-with-bm25
-#SBATCH -o logs/slurm-%x-%j.log
+set -eu
 
-project=/cl/home2/shintaro/rag-notebook/shintaro
+project=$(pwd)
 source $project/.venv/bin/activate
 
-<< COMMENTOUT
-models: microsoft/Phi-3.5-mini-instruct, axiong/PMC_LLaMA_13B, meta-llama/Llama-3.1-70B, epfl-llm/meditron-70b
-RAG_MODELS=("ncbi/MedCPT-Query-Encoder" "intfloat/e5-base-v2" "facebook/contriever" "bm25")
-TASKS=("medqa" "medmcqa" "mmlu" "pubmedqa")
-RAGDB=("statpearls" "texatbooks" "pubmed" "wikipedia")
-COMMENTOUT
-
 inference_model=meta-llama/Llama-3.1-70B
-
 rag_db=statpearls
 inference_max_length=2048
 quantize_type=4bit
-
 task=mmlu
 rag_max_length=512
 top_k=1