add depr warning for preprocess --iterable

src/axolotl/cli/args.py

@@ -13,6 +13,16 @@ class PreprocessCliArgs:
     debug_num_examples: int = field(default=1)
     prompter: Optional[str] = field(default=None)
     download: Optional[bool] = field(default=True)
+    iterable: Optional[bool] = field(
+        default=False,
+        metadata={
+            "help": (
+                "[DEPRECATED] No longer supported. For streaming datasets, use "
+                "'axolotl train' and set 'streaming: true' in your YAML config, or "
+                "pass --streaming instead in the CLI."
+            )
+        },
+    )
 
 
 @dataclass

src/axolotl/cli/preprocess.py

@@ -35,10 +35,20 @@ def do_preprocess(cfg: DictDefault, cli_args: PreprocessCliArgs) -> None:
     check_accelerate_default_config()
     check_user_token()
 
+    if cli_args.iterable:
+        LOG.error(
+            "The --iterable CLI argument for 'axolotl preprocess' is no longer "
+            "supported. For training, set 'streaming: true' in your YAML config or "
+            "pass '--streaming' in your 'axolotl train' command for on-the-fly "
+            "preprocessing."
+        )
+        return
+
     for key in ["skip_prepare_dataset", "pretraining_dataset"]:
         if cfg.get(key):
             LOG.error(
-                f"You have set `{key}:`. `preprocess` is not needed. Run the `axolotl train` CLI directly instead."
+                f"You have set `{key}:`. `preprocess` is not needed. Run the 'axolotl "
+                "train' CLI directly instead."
             )
             return
 

src/axolotl/utils/schemas/validation.py

@@ -1441,7 +1441,8 @@ class StreamingValidationMixin:
             val_set_size = getattr(self, "val_set_size", 0.0)
             if val_set_size and val_set_size > 0:
                 raise ValueError(
-                    "Validation splits not supported for streaming datasets, skipping"
+                    "Validation splits not supported for streaming datasets, please "
+                    "use test_datasets: ... instead"
                 )
 
         return self