Update GPT4All dataset loading for new files (#2344)

Close #2331. The multiround conversations in the previous files were
badly formatted, so the GPT4All author uploaded separate files for
singleround and multiround. Here we only load singleround examples
and can later add support for multiround

model/model_training/custom_datasets/prompt_dialogue.py

@@ -108,7 +108,7 @@ def __init__(self, mode: str, cache_dir: str = None) -> None:
         self.mode = mode
         dataset = load_dataset(
             "Nebulous/gpt4all_pruned",
-            data_files="data_pruned_3.jsonl",
+            data_files="data_singleround_pruned_3.jsonl",
             cache_dir=cache_dir,
         )
         self.rows = [(row["prompt"], row["response"]) for row in dataset["train"]]