Filter on dataset too much slowww

I have a dataset with 50M rows.
For pre-processing, I need to tokenize this and filter rows with the large sequence.

My tokenization took roughly 12mins. I used `map()` with batch size 1024 and multi-process with 96 processes.

When I applied the `filter()` function it is taking too much time. I need to filter sequences based on a boolean column.
Below are the variants I tried.
1. filter() with batch size 1024, single process (takes roughly 3 hr)
2. filter() with batch size 1024, 96 processes (takes 5-6 hrs ¯\\\_(ツ)\_/¯)
3. filter() with loading all data in memory, only a single boolean column (never ends).

Can someone please help?

Below is a sample code for small dataset.

```
from datasets import load_dataset
dataset = load_dataset('glue', 'mrpc', split='train')
dataset = dataset.map(lambda x: {'flag': random.randint(0,1)==1})

def _amplify(data):
return data

dataset = dataset.filter(_amplify, batch_size=1024, keep_in_memory=False, input_columns=['flag'])
```