diff --git a/src/pypi_data/cli.py b/src/pypi_data/cli.py
index b4146fec..3fbecd2a 100644
--- a/src/pypi_data/cli.py
+++ b/src/pypi_data/cli.py
@@ -158,11 +158,13 @@ def merge_datasets(
     repo_path: Path,
     output: Path,
     max_buffer_size: Annotated[str, typer.Option()] = "10GB",
+    target_size: Annotated[str, typer.Option()] = "1.8GB",
 ):
     with open_path(repo_path, mode="rb") as fd:
         repos = Repos.model_validate_json(fd.read()).root
     max_buffer_size = pydantic.RootModel[ByteSize].model_validate(max_buffer_size).root
-    asyncio.run(combine_parquet(repos, output, max_buffer_size))
+    target_size = pydantic.RootModel[ByteSize].model_validate(target_size).root
+    asyncio.run(combine_parquet(repos, output, max_buffer_size, target_size))
 
 
 async def resolve_dataset_redirects(
diff --git a/src/pypi_data/combine_parquet.py b/src/pypi_data/combine_parquet.py
index 00e40947..11cfb10f 100644
--- a/src/pypi_data/combine_parquet.py
+++ b/src/pypi_data/combine_parquet.py
@@ -18,7 +18,6 @@
 
 log = structlog.get_logger()
 
-TARGET_SIZE = 1024 * 1024 * 1024 * 1.8  # 1.8 GB
 IO_BUFFER_SIZE = 1024 * 1024 * 50  # 50 MB
 
 
@@ -27,6 +26,7 @@ def append_buffer(
     writer: pq.ParquetWriter,
     batch: RecordBatch,
     roll_up_path: Path,
+    target_size: int,
 ) -> bool:
     initial_size = roll_up_path.stat().st_size
     writer.write_batch(batch)
@@ -35,12 +35,12 @@ def append_buffer(
     written_size = end_size - initial_size
     log.info(
         f"Wrote {batch.num_rows} rows "
-        f"Batch Size: {ByteSize(batch.nbytes).human_readable(decimal=True)} MB "
-        f"Initial Size: {ByteSize(initial_size).human_readable(decimal=True)} MB "
-        f"End Size: {ByteSize(end_size).human_readable(decimal=True)} MB "
-        f"Written: {ByteSize(written_size).human_readable(decimal=True)} MB"
+        f"Batch Size: {ByteSize(batch.nbytes).human_readable(decimal=True)} "
+        f"Initial Size: {ByteSize(initial_size).human_readable(decimal=True)} "
+        f"End Size: {ByteSize(end_size).human_readable(decimal=True)} "
+        f"Written: {ByteSize(written_size).human_readable(decimal=True)}"
     )
-    return end_size >= TARGET_SIZE
+    return end_size >= target_size
 
 
 def buffer_mem_size(buffer: Deque[tuple[tuple[int, str], RecordBatch]]) -> int:
@@ -61,9 +61,7 @@ async def fill_buffer(
         start_time_ns = time.perf_counter_ns()
 
         buffer_size = buffer_mem_size(buffer)
-        log.info(
-            f"Buffer size: {ByteSize(buffer_size).human_readable(decimal=True)} MB"
-        )
+        log.info(f"Buffer size: {ByteSize(buffer_size).human_readable(decimal=True)}")
         if buffer_size >= max_buffer_size:
             log.info(f"Buffer filled with {len(buffer)} entries")
             break
@@ -129,7 +127,10 @@ def hash_parquet_keys(keys: list[tuple[int, str]]) -> str:
 
 
 async def combine_parquet(
-    repositories: list[CodeRepository], directory: Path, max_buffer_size: ByteSize
+    repositories: list[CodeRepository],
+    directory: Path,
+    max_buffer_size: ByteSize,
+    target_size: ByteSize,
 ):
     directory.mkdir(exist_ok=True)
 
@@ -143,10 +144,10 @@ async def combine_parquet(
         max_buffer_size,  # 10 GB
     )
     log.info(
-        f"Total system memory: {ByteSize(total_memory).human_readable(decimal=True)} MB"
+        f"Total system memory: {ByteSize(total_memory).human_readable(decimal=True)}"
     )
     log.info(
-        f"Configured buffer size: {ByteSize(max_buffer_size).human_readable(decimal=True)} MB"
+        f"Configured buffer size: {ByteSize(max_buffer_size).human_readable(decimal=True)}"
     )
 
     async with httpx.AsyncClient(follow_redirects=True) as client:
@@ -172,7 +173,7 @@ async def combine_parquet(
                 write_statistics=True,
                 schema=first_buffer.schema,
             ) as writer:
-                append_buffer(fd, writer, first_buffer, roll_up_path)
+                append_buffer(fd, writer, first_buffer, roll_up_path, target_size)
 
                 while buffer or repositories:
                     if not buffer:
@@ -184,7 +185,7 @@ async def combine_parquet(
 
                     key, batch = buffer.popleft()
                     keys.append(key)
-                    if append_buffer(fd, writer, batch, roll_up_path):
+                    if append_buffer(fd, writer, batch, roll_up_path, target_size):
                         break
 
             hashed_keys = hash_parquet_keys(keys)
@@ -193,7 +194,7 @@ async def combine_parquet(
             )
 
             log.info(
-                f"Finished batch {roll_up_count}, {len(keys)} batches, {final_path.name=} {ByteSize(final_path.stat().st_size).human_readable(decimal=True)} MB"
+                f"Finished batch {roll_up_count}, {len(keys)} batches, {final_path.name=} {ByteSize(final_path.stat().st_size).human_readable(decimal=True)}"
             )
 
             roll_up_count += 1
diff --git a/src/pypi_data/datasets.py b/src/pypi_data/datasets.py
index 165eeeb3..1b854abc 100644
--- a/src/pypi_data/datasets.py
+++ b/src/pypi_data/datasets.py
@@ -124,7 +124,7 @@ async def download_dataset(self, client: httpx.AsyncClient) -> bytes | None:
         if response is None:
             return None
         log.info(
-            f"Downloaded parquet file with size {ByteSize(len(response.content)).human_readable(decimal=True)} MB"
+            f"Downloaded parquet file with size {ByteSize(len(response.content)).human_readable(decimal=True)}"
         )
         return response.content