Merge branch 'master' into convert-split

convert-hf-to-gguf.py

@@ -327,7 +327,7 @@ def write_tensors(self):
     def write(self):
         self.write_tensors()
         self.gguf_writer.init_shards()
-        self.gguf_writer.write_header_to_file()
+        self.gguf_writer.write_header_to_file(self.fname_out)
         self.gguf_writer.write_kv_data_to_file()
         self.gguf_writer.write_tensors_to_file(progress=True)
         self.gguf_writer.close()
@@ -336,7 +336,7 @@ def write_vocab(self):
         if self.gguf_writer.split_arguments.split:
             raise ValueError('Splitting the vocabulary is not supported')
         self.gguf_writer.init_shards()
-        self.gguf_writer.write_header_to_file()
+        self.gguf_writer.write_header_to_file(self.fname_out)
         self.gguf_writer.write_kv_data_to_file()
         self.gguf_writer.close()
 

gguf-py/gguf/gguf_writer.py

@@ -76,7 +76,7 @@ class GGUFWriter:
     }
 
     def __init__(
-        self, path: os.PathLike[str] | str, arch: str, use_temp_file: bool = True,
+        self, path: os.PathLike[str] | str | None, arch: str, use_temp_file: bool = False,
         endianess: GGUFEndian = GGUFEndian.LITTLE, add_architecture: bool = True
     ):
         self.fout = None
@@ -306,24 +306,17 @@ def write_tensors_to_file(self, *, progress: bool = False) -> None:
 
                 bar = tqdm(desc="Writing", total=total_bytes, unit="byte", unit_scale=True)
 
-                while True:
-                    try:
-                        tensor = self.tensors.pop()
-                    except IndexError:
-                        break
-                    tensor.tofile(self.fout)
-                    bar.update(tensor.nbytes)
-                    self.write_padding(self.fout, tensor.nbytes)
-                    del tensor
-                return
-            while True:
-                try:
-                    tensor = self.tensors.pop()
-                except IndexError:
-                    break
-                tensor.tofile(self.fout)
-                self.write_padding(self.fout, tensor.nbytes)
-            return
+            # relying on the fact that Python dicts preserve insertion order (since 3.7)
+            for ti in self.tensors.values():
+                assert ti.tensor is not None  # can only iterate once over the tensors
+                assert ti.tensor.nbytes == ti.nbytes
+                ti.tensor.tofile(self.fout)
+                if bar is not None:
+                    bar.update(ti.nbytes)
+                self.write_padding(self.fout, ti.nbytes)
+                ti.tensor = None
+        else:
+            self.temp_file.seek(0)
 
             shutil.copyfileobj(self.temp_file, self.fout)
             self.flush()