Option to split during conversion (#6942)

* support splits in convert.py

* Support split by size and dry run to write estimated shards/filesizes

* Move split functionality to new GGUFManager class

* fix improper function signature

* tentative push of convert-hf-to-gguf support

* resolve merge + SplitArguments for easier parsing

* Fix eager tensor memory leak and remove convert.py changes

Removed a memory leak caused by unexpected reference retention to eager tensors.

Also removed GGUFManager functionality in convert.py in favor of specializing for convert-hf-to-gguf.py.

* refactor SplitStrategy to be a deque

Instead of having SplitStrategy have a `data` field that is a deque, just have SplitStrategy be a subclass of deque itself.

* fix Q8 quantization

* remove unnecessary imports in gguf_manager

* fix final? merge issue

* fix gguf_writer placement and remove comments

* oops, actually fix gguf_writer placement

* reduce duplicated code from gguf_writer

* further simplify GGUFManager

* simplify even further and standardize with GGUFWriter

* reduce diffs with master

* form shards while adding tensors, SHA256 sums agree with master

* re-add type hint

Co-authored-by: compilade <[email protected]>

* GGUFWriter compatibility fix

Co-authored-by: compilade <[email protected]>

* Shard dataclass and un-negative dont_add_architecture

* type consistency in format_n_bytes_to_str

* move kv keys to constants.py

* make pathlib explicit

* base-1024 bytes to base-1000

* rename GGUFManager to GGUFWriterSplit

* Update gguf-py/gguf/constants.py

Co-authored-by: compilade <[email protected]>

* fix convert-hf-to-gguf.py permissions

* fix line endings

* Update gguf-py/gguf/gguf_writer_split.py

Co-authored-by: compilade <[email protected]>

* convert-hf : restore executable file permission

* examples/convert-legacy-llama.py: restore executable file permission

* reinstate original gguf package import and fix type annotation

* attempt to appease the linter

* attempt 2 to appease the linter

* attempt 3 to appease the linter

* comma consistency

* Update convert-hf-to-gguf.py

Co-authored-by: compilade <[email protected]>

* edit cmd line args

* use simplification from #7827

* kv/ti data are still wrong

* try to refactor kv data (still fails)

* fix ti data messiness

* tidy up

* fix linting

* actually make the linter happy

* cleanup round 1

* remove SplitStrategy, SplitArguments

* appease linter

* fix typing and clean up

* fix linting

* Update gguf-py/gguf/gguf_writer.py

Co-authored-by: compilade <[email protected]>

* progress bar, fix split logic

* Update gguf-py/gguf/gguf_writer.py

Co-authored-by: compilade <[email protected]>

* catch oversights

* Update gguf-py/gguf/gguf_writer.py

Co-authored-by: compilade <[email protected]>

* Update gguf-py/gguf/gguf_writer.py

Co-authored-by: compilade <[email protected]>

* Update gguf-py/gguf/gguf_writer.py

Co-authored-by: compilade <[email protected]>

* Update gguf-py/gguf/gguf_writer.py

Co-authored-by: compilade <[email protected]>

* Update gguf-py/gguf/gguf_writer.py

Co-authored-by: compilade <[email protected]>

* swap bar orders

* Update gguf-py/gguf/gguf_writer.py

Co-authored-by: compilade <[email protected]>

* Update gguf-py/gguf/gguf_writer.py

Co-authored-by: compilade <[email protected]>

* compatibility fix

* Update gguf-py/gguf/gguf_writer.py

Co-authored-by: compilade <[email protected]>

* Update convert-hf-to-gguf.py

Co-authored-by: compilade <[email protected]>

---------

Co-authored-by: Brian <[email protected]>
Co-authored-by: compilade <[email protected]>

Author: 3 people

convert-hf-to-gguf.py

@@ -65,7 +65,8 @@ class Model:
     # subclasses should define this!
     model_arch: gguf.MODEL_ARCH
 
-    def __init__(self, dir_model: Path, ftype: gguf.LlamaFileType, fname_out: Path, is_big_endian: bool, use_temp_file: bool, eager: bool, model_name: str | None):
+    def __init__(self, dir_model: Path, ftype: gguf.LlamaFileType, fname_out: Path, is_big_endian: bool, use_temp_file: bool, eager: bool,
+                 model_name: str | None, split_max_tensors: int = 0, split_max_size: int = 0, dry_run: bool = False, small_first_shard: bool = False):
         if type(self) is Model:
             raise TypeError(f"{type(self).__name__!r} should not be directly instantiated")
         self.dir_model = dir_model
@@ -96,7 +97,8 @@ def __init__(self, dir_model: Path, ftype: gguf.LlamaFileType, fname_out: Path,
         ftype_lw: str = ftype_up.lower()
         # allow templating the file name with the output ftype, useful with the "auto" ftype
         self.fname_out = fname_out.parent / fname_out.name.format(ftype_lw, outtype=ftype_lw, ftype=ftype_lw, OUTTYPE=ftype_up, FTYPE=ftype_up)
-        self.gguf_writer = gguf.GGUFWriter(path=None, arch=gguf.MODEL_ARCH_NAMES[self.model_arch], endianess=self.endianess, use_temp_file=self.use_temp_file)
+        self.gguf_writer = gguf.GGUFWriter(path=None, arch=gguf.MODEL_ARCH_NAMES[self.model_arch], endianess=self.endianess, use_temp_file=self.use_temp_file,
+                                           split_max_tensors=split_max_tensors, split_max_size=split_max_size, dry_run=dry_run, small_first_shard=small_first_shard)
 
     @classmethod
     def __init_subclass__(cls):
@@ -332,6 +334,8 @@ def write(self):
         self.gguf_writer.close()
 
     def write_vocab(self):
+        if len(self.gguf_writer.tensors) != 1:
+            raise ValueError('Splitting the vocabulary is not supported')
         self.gguf_writer.write_header_to_file(self.fname_out)
         self.gguf_writer.write_kv_data_to_file()
         self.gguf_writer.close()
@@ -2974,10 +2978,44 @@ def parse_args() -> argparse.Namespace:
         "--verbose", action="store_true",
         help="increase output verbosity",
     )
+    parser.add_argument(
+        "--split-max-tensors", type=int, default=0,
+        help="max tensors in each split",
+    )
+    parser.add_argument(
+        "--split-max-size", type=str, default="0",
+        help="max size per split N(M|G)",
+    )
+    parser.add_argument(
+        "--dry-run", action="store_true",
+        help="only print out a split plan and exit, without writing any new files",
+    )
+    parser.add_argument(
+        "--no-tensor-first-split", action="store_true",
+        help="do not add tensors to the first split (disabled by default)"
+    )
 
     return parser.parse_args()
 
 
+def split_str_to_n_bytes(split_str: str) -> int:
+    if split_str.endswith("K"):
+        n = int(split_str[:-1]) * 1000
+    elif split_str.endswith("M"):
+        n = int(split_str[:-1]) * 1000 * 1000
+    elif split_str.endswith("G"):
+        n = int(split_str[:-1]) * 1000 * 1000 * 1000
+    elif split_str.isnumeric():
+        n = int(split_str)
+    else:
+        raise ValueError(f"Invalid split size: {split_str}, must be a number, optionally followed by K, M, or G")
+
+    if n < 0:
+        raise ValueError(f"Invalid split size: {split_str}, must be positive")
+
+    return n
+
+
 def main() -> None:
     args = parse_args()
 
@@ -3010,6 +3048,10 @@ def main() -> None:
         "auto": gguf.LlamaFileType.GUESSED,
     }
 
+    if args.use_temp_file and (args.split_max_tensors > 0 or args.split_max_size != "0"):
+        logger.error("Error: Cannot use temp file when splitting")
+        sys.exit(1)
+
     if args.outfile is not None:
         fname_out = args.outfile
     else:
@@ -3027,7 +3069,10 @@ def main() -> None:
             logger.error(f"Model {hparams['architectures'][0]} is not supported")
             sys.exit(1)
 
-        model_instance = model_class(dir_model, ftype_map[args.outtype], fname_out, args.bigendian, args.use_temp_file, args.no_lazy, args.model_name)
+        model_instance = model_class(dir_model, ftype_map[args.outtype], fname_out, args.bigendian, args.use_temp_file,
+                                     args.no_lazy, args.model_name, split_max_tensors=args.split_max_tensors,
+                                     split_max_size=split_str_to_n_bytes(args.split_max_size), dry_run=args.dry_run,
+                                     small_first_shard=args.no_tensor_first_split)
 
         logger.info("Set model parameters")
         model_instance.set_gguf_parameters()
@@ -3038,13 +3083,13 @@ def main() -> None:
         model_instance.gguf_writer.add_quantization_version(gguf.GGML_QUANT_VERSION)
 
         if args.vocab_only:
-            logger.info(f"Exporting model vocab to '{model_instance.fname_out}'")
+            logger.info("Exporting model vocab...")
             model_instance.write_vocab()
+            logger.info("Model vocab successfully exported.")
         else:
-            logger.info(f"Exporting model to '{model_instance.fname_out}'")
+            logger.info("Exporting model...")
             model_instance.write()
-
-        logger.info(f"Model successfully exported to '{model_instance.fname_out}'")
+            logger.info("Model successfully exported.")
 
 
 if __name__ == '__main__':

gguf-py/gguf/constants.py

@@ -75,6 +75,11 @@ class Rope:
         SCALING_FINETUNED       = "{arch}.rope.scaling.finetuned"
         SCALING_YARN_LOG_MUL    = "{arch}.rope.scaling.yarn_log_multiplier"
 
+    class Split:
+        LLM_KV_SPLIT_NO            = "split.no"
+        LLM_KV_SPLIT_COUNT         = "split.count"
+        LLM_KV_SPLIT_TENSORS_COUNT = "split.tensors.count"
+
     class SSM:
         CONV_KERNEL    = "{arch}.ssm.conv_kernel"
         INNER_SIZE     = "{arch}.ssm.inner_size"