diff --git a/.gitignore b/.gitignore
index bf560c6..03bcc6a 100644
--- a/.gitignore
+++ b/.gitignore
@@ -164,3 +164,4 @@ cython_debug/
 metrics.txt
 metrics.png
 gdrive-oauth.txt
+/eval
diff --git a/data/.gitignore b/data/.gitignore
index 09fbf7e..addcca2 100644
--- a/data/.gitignore
+++ b/data/.gitignore
@@ -13,3 +13,4 @@
 /supporting-docs.json
 /metrics.json
 /eval.png
+/eidc_rag_testset.csv
diff --git a/dvc.yaml b/dvc.yaml
index 2473eec..903cb5a 100644
--- a/dvc.yaml
+++ b/dvc.yaml
@@ -1,3 +1,5 @@
+metrics: 
+- data/metrics.json
 stages:
   fetch-metadata:
     cmd: python scripts/fetch_eidc_metadata.py ${files.metadata} -s ${sub-sample}
@@ -20,7 +22,7 @@ stages:
     outs:
     - ${files.extracted}
   chunk-data:
-    cmd: python scripts/chunk_data.py -o ${files.chunked} -c ${hp.chunk-size} -ol ${hp.overlap} ${files.extracted} ${files.supporting-docs}
+    cmd: python scripts/chunk_data.py -o ${files.chunked} -c ${hp.chunk-size} -ol ${hp.overlap} ${files.extracted} ${files.supporting-docs} -m ${max-length}
     deps:
     - ${files.extracted}
     - ${files.supporting-docs}
@@ -42,7 +44,7 @@ stages:
     outs:
     - ${files.doc-store}
   generate-testset:
-    cmd: cp data/synthetic-datasets/eidc_rag_test_sample.csv data/
+    cmd: head -n 2 data/synthetic-datasets/eidc_rag_test_sample.csv > ${files.test-set}
     outs: 
     - ${files.test-set}
   run-rag-pipeline:
@@ -61,5 +63,3 @@ stages:
     outs:
     - ${files.metrics}
     - ${files.eval-plot}
-metrics: 
-- ${files.metrics}
\ No newline at end of file
diff --git a/params.yaml b/params.yaml
index edf0085..bfc46af 100644
--- a/params.yaml
+++ b/params.yaml
@@ -12,11 +12,12 @@ files:
   chunked: data/chunked_data.json
   embeddings: data/embeddings.json
   doc-store: data/chroma-data
-  test-set: data/eidc_rag_test_sample.csv
+  test-set: data/eidc_rag_testset.csv
   eval-set: data/evaluation_data.csv
   metrics: data/metrics.json
   eval-plot: data/eval.png
-sub-sample: 3 # sample size of 0 will process all data
+sub-sample: 0 # sample n datasets for testing (0 will use all datasets)
+max-length: 0 # truncate longer texts for testing (0 will use all data)
 rag:
   model: llama3.1
   prompt: >-
diff --git a/pyproject.toml b/pyproject.toml
index 3dda280..a56e3f0 100644
--- a/pyproject.toml
+++ b/pyproject.toml
@@ -17,6 +17,7 @@ dependencies = [
     "ragas == 0.1.10",
     "nltk == 3.9.1",
     "nbformat == 4.2.0",
+    "pygit2 == 1.14.1",
 ]
 
 [project.optional-dependencies]
diff --git a/run-experiments.sh b/run-experiments.sh
new file mode 100755
index 0000000..43a8d67
--- /dev/null
+++ b/run-experiments.sh
@@ -0,0 +1,7 @@
+#!/bin/bash
+dvc queue remove --all
+dvc exp run --queue -S hp.chunk-size=400 -S sub-sample=1 -S max-length=500
+dvc exp run --queue -S hp.chunk-size=600 -S sub-sample=1 -S max-length=500
+dvc queue start
+dvc queue status
+echo "Run `dvc queue status` to check the state of the experiments"
diff --git a/scripts/chunk_data.py b/scripts/chunk_data.py
index 7fe672b..673651d 100644
--- a/scripts/chunk_data.py
+++ b/scripts/chunk_data.py
@@ -3,19 +3,20 @@
 from typing import Any, Dict, List
 
 
-def chunk_value(value: str, chunk_size: int, overlap: int) -> List[str]:
+def chunk_value(value: str, chunk_size: int, overlap: int, max_length: int) -> List[str]:
     chunks = []
     start = 0
-    while start < len(value):
+    end = max_length if len(value) > max_length > 0 else len(value)
+    while start < end:
         chunks.append(value[start : (start + chunk_size)])
         start += chunk_size - overlap
     return chunks
 
 
 def chunk_metadata_value(
-    metada_value: str, chunk_size: int, overlap: int
+    metada_value: str, chunk_size: int, overlap: int, max_length: int
 ) -> List[Dict[str, Any]]:
-    chunks = chunk_value(metada_value["value"], chunk_size, overlap)
+    chunks = chunk_value(metada_value["value"], chunk_size, overlap, max_length)
     return [
         {
             "chunk": chunks[i],
@@ -28,20 +29,20 @@ def chunk_metadata_value(
 
 
 def chunk_metadata_file(
-    file: str, chunk_size: int, overlap: int
+    file: str, chunk_size: int, overlap: int, max_length: int
 ) -> List[Dict[str, str]]:
     chunked_metadata = []
     with open(file) as f:
         json_data = json.load(f)
         for metadata in json_data:
-            chunked_metadata.extend(chunk_metadata_value(metadata, chunk_size, overlap))
+            chunked_metadata.extend(chunk_metadata_value(metadata, chunk_size, overlap, max_length))
     return chunked_metadata
 
 
-def main(files: List[str], ouput_file: str, chunk_size: int, overlap: int) -> None:
+def main(files: List[str], ouput_file: str, chunk_size: int, overlap: int, max_length: int) -> None:
     all_chunked_metadata = []
     for file in files:
-        all_chunked_metadata.extend(chunk_metadata_file(file, chunk_size, overlap))
+        all_chunked_metadata.extend(chunk_metadata_file(file, chunk_size, overlap, max_length))
     with open(ouput_file, "w") as f:
         json.dump(all_chunked_metadata, f, indent=4)
 
@@ -73,6 +74,15 @@ def main(files: List[str], ouput_file: str, chunk_size: int, overlap: int) -> No
         nargs="?",
         const=100,
     )
+    parser.add_argument(
+        "-m",
+        "--max_length",
+        help="""Maximum length of data in characters - meant for truncating large 
+        strings in testing. 0 defaults to all data""",
+        type=int,
+        nargs="?",
+        const=0,
+    )
     args = parser.parse_args()
     assert args.chunk > args.overlap
-    main(args.input_files, args.output, args.chunk, args.overlap)
+    main(args.input_files, args.output, args.chunk, args.overlap, args.max_length)
diff --git a/scripts/create_embeddings.py b/scripts/create_embeddings.py
index 7aa507c..220eed0 100644
--- a/scripts/create_embeddings.py
+++ b/scripts/create_embeddings.py
@@ -1,6 +1,7 @@
 import json
 from argparse import ArgumentParser
-
+import gc
+import torch
 from sentence_transformers import SentenceTransformer
 from torch import Tensor
 from tqdm import tqdm
@@ -16,6 +17,8 @@ def main(input_file: str, output_file: str) -> None:
         data = json.load(input)
         for chunk in tqdm(data):
             chunk["embedding"] = create_embedding(chunk["chunk"]).tolist()
+            gc.collect()
+            torch.cuda.empty_cache()
         json.dump(data, output)
 
 
diff --git a/scripts/evaluate.py b/scripts/evaluate.py
index c130e96..fc2ffb8 100644
--- a/scripts/evaluate.py
+++ b/scripts/evaluate.py
@@ -1,5 +1,6 @@
 import json
 from argparse import ArgumentParser
+from pathlib import Path
 
 import nest_asyncio
 import pandas as pd
@@ -44,10 +45,9 @@ def main(eval_dataset: str, metric_output: str, image_output: str) -> None:
         run_config=RunConfig(max_workers=1),
     )
     result_df = result.to_pandas()
-    pio.templates.default = "gridon"
-    fig = go.Figure()
 
-    with open(metric_output, "w") as f:
+    Path(metric_output).parent.mkdir(parents=True, exist_ok=True)
+    with open(metric_output, "w+") as f:
         json.dump(result, f)
     metrics = [
         metric
@@ -55,6 +55,10 @@ def main(eval_dataset: str, metric_output: str, image_output: str) -> None:
         if metric not in ["question", "ground_truth", "answer", "contexts"]
     ]
 
+
+    pio.templates.default = "gridon"
+    fig = go.Figure()
+
     for metric in metrics:
         fig.add_trace(
             go.Violin(
@@ -66,7 +70,7 @@ def main(eval_dataset: str, metric_output: str, image_output: str) -> None:
             )
         )
     fig.update_yaxes(range=[-0.02, 1.02])
-    with open(image_output, "wb") as f:
+    with open(image_output, "wb+") as f:
         f.write(fig.to_image(format="png"))