fix: update docs and scripts for the Llama2 experiment

docs/Llama2FineTuning.md

@@ -30,6 +30,9 @@ cd Llama-2-7b-hf
 git lfs fetch
 ```
 
+6. Copy the content of [`requirements.txt`](https://github.com/facebookresearch/llama/blob/main/requirements.txt)
+for setting up the Llama2 models into a new file named `requirements-llama2.txt` in the "llama" directory.
+
 ## Use the Llama2 model
 
 In the [`jobs/original_use`](../jobs/Llama2/original_use) directory, there are two scripts:

jobs/Llama2/original_use/job_original_use_7b_hf.sh

@@ -8,13 +8,22 @@
 #SBATCH --cpus-per-task=4
 #SBATCH --account=def-whkchun
 #SBATCH --output=%x.o%j
- 
-pip install --no-index --upgrade pip
-module load python/3.8.2
+
+pip install --upgrade pip
+module load python/3.11.5
 python -V
 
-source ~/llama2/.venv/bin/activate
+virtualenv --no-download $SLURM_TMPDIR/env
+source $SLURM_TMPDIR/.env/bin/activate
+
+pip install --upgrade pip
 
+module load StdEnv/2023
+module load rust/1.70.0
+pip install --no-index transformers==4.36.2
+pip install -r /home/cindyli/llama2/requirements-llama2.txt
+
+python -V
 pip list
 
 echo "Llama2 original use from job $SLURM_JOB_ID on nodes $SLURM_JOB_NODELIST."

jobs/Llama2/original_use/original_use_7b_hf.py

@@ -22,7 +22,7 @@
 # write results into the result file
 file.write("2. Word prediction: \n")
 for prediction in predictions:
-    file.write(f"- {prediction['generated_text']}: {prediction['score']}\n")
+    file.write(f"- {prediction['generated_text']}\n")
 file.write("\n")
 
 # 3. Test prompt for inference

jobs/Llama2/requirements-llama2.txt

@@ -0,0 +1,4 @@
+torch
+fairscale
+fire
+sentencepiece