fix(save): unsloth_push_to_hub_gguf(save_method="lora") raises NameError (#7193)

* fix(save): unsloth_push_to_hub_gguf(save_method="lora") raises NameError

unsloth_push_to_hub_gguf reads is_main_process at save.py:3181 but never
declares it. Its twin unsloth_save_pretrained_gguf declares it (2783) and
uses it the same way (2839) -- the LoRA branch was copied between the twins,
the parameter it depends on was not. There is no module-level global, so the
name resolves as a global load and the branch raises NameError 100% of the
time.

save_pretrained_gguf(save_method="lora", push_to_hub=True) raises a
ValueError that tells users to "use .push_to_hub_gguf(save_method='lora')
instead" -- the documented escape hatch is the broken call.

Add is_main_process to the signature, positioned as in the twin, and forward
it to unsloth_save_pretrained_gguf on the merged path so the parameter is not
silently ignored there. Default stays True, so nothing changes for existing
callers.

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>

* fix(save): preserve GGUF push compatibility

---------

Co-authored-by: Claude Opus 4.8 <noreply@anthropic.com>
Co-authored-by: Etherll <61019402+Etherll@users.noreply.github.com>

tests/saving/test_export_dispatch.py

@@ -8,6 +8,8 @@ regressions that pure AST checks cannot (e.g. wrong scheme/suffix/outtype passed
 
 from __future__ import annotations
 
+import inspect
+
 import pytest
 
 import unsloth.save as save_mod
@@ -126,6 +128,80 @@ def test_gguf_lora_push_to_hub_is_rejected(tmp_path):
         )
 
 
+# The above rejection points users at push_to_hub_gguf(save_method='lora'), so that path
+# has to work; it is only ever exercised here.
+
+
+def test_push_to_hub_gguf_lora_dispatches(monkeypatch):
+    seen = {}
+    monkeypatch.setattr(
+        save_mod,
+        "_unsloth_save_lora_gguf",
+        lambda model, tok, sd, **kw: seen.update(kw),
+    )
+    save_mod.unsloth_push_to_hub_gguf(
+        _FakeModel(),
+        "repo/id",
+        tokenizer = object(),
+        save_method = "lora",
+        quantization_method = "q8_0",
+    )
+    assert seen.get("outtype") == "q8_0"
+    assert seen.get("push_to_hub") is True
+
+
+def test_push_to_hub_gguf_lora_skips_non_main_process(monkeypatch):
+    calls = []
+    monkeypatch.setattr(
+        save_mod,
+        "_unsloth_save_lora_gguf",
+        lambda *a, **kw: calls.append(kw),
+    )
+    result = save_mod.unsloth_push_to_hub_gguf(
+        _FakeModel(),
+        "repo/id",
+        tokenizer = object(),
+        save_method = "lora",
+        is_main_process = False,
+    )
+    assert result is None
+    assert calls == []
+
+
+def test_push_to_hub_gguf_skips_non_main_process_before_merged_conversion(monkeypatch):
+    calls = []
+    monkeypatch.setattr(
+        save_mod,
+        "unsloth_save_pretrained_gguf",
+        lambda **kw: calls.append(kw),
+    )
+    result = save_mod.unsloth_push_to_hub_gguf(
+        _FakeModel(),
+        "repo/id",
+        tokenizer = object(),
+        is_main_process = False,
+    )
+    assert result is None
+    assert calls == []
+
+
+def test_push_to_hub_gguf_preserves_positional_max_shard_size():
+    bound = inspect.signature(save_mod.unsloth_push_to_hub_gguf).bind(
+        _FakeModel(),
+        "repo/id",
+        object(),
+        "q4_k_m",
+        None,
+        None,
+        None,
+        None,
+        "token",
+        "50GB",
+    )
+    assert bound.arguments["max_shard_size"] == "50GB"
+    assert "is_main_process" not in bound.arguments
+
+
 # -- torchao PTQ / QAT dispatch ------------------------------------------------------------
 
 

unsloth/save.py

@@ -3143,6 +3143,7 @@ def unsloth_push_to_hub_gguf(
     datasets: Optional[List[str]] = None,
     save_method: str = None,
     imatrix_file = None,
+    is_main_process: bool = True,
 ):
     """
     Same as .push_to_hub(...) except 4bit weights are auto
@@ -3175,11 +3176,11 @@ def unsloth_push_to_hub_gguf(
     """
     if tokenizer is None:
         raise ValueError("Unsloth: Saving to GGUF must have a tokenizer.")
+    if not is_main_process:
+        return None
 
     # save_method="lora" exports the adapter itself as a GGUF LoRA (not a merged model).
     if save_method is not None and str(save_method).lower() == "lora":
-        if not is_main_process:
-            return None  # only the main rank converts and uploads, like the local lora branch
         _qm = quantization_method
         if isinstance(_qm, (list, tuple)) and len(_qm) == 1:
             _qm = _qm[0]  # the gguf API allows a list; unwrap a single outtype
@@ -3233,6 +3234,7 @@ def unsloth_push_to_hub_gguf(
             first_conversion = first_conversion,
             push_to_hub = False,  # Never push from here
             token = token,  # forwarded so imatrix_file=True can read a gated/private upstream
+            is_main_process = is_main_process,
             max_shard_size = max_shard_size,
             safe_serialization = safe_serialization,
             temporary_location = temporary_location,