Superminaren/llama.cpp
1
0
Fork 0
mirror of https://github.com/ggml-org/llama.cpp.git synced 2026-09-04 02:37:27 +02:00
Code Issues Packages Projects Releases Wiki Activity

convert: add option to create separate dspark GGUF (#26452)

Browse Source 
* convert: add option to create separate dspark GGUF

* add --no-nextn

* fix convert bug
This commit is contained in:
Aman Gupta
2026-08-02 23:16:31 +08:00
committed by GitHub
parent c745be2a2c
commit 3581ba0cf5
4 changed files with 146 additions and 6 deletions
Download Patch File Download Diff File Expand all files Collapse all files
convert_hf_to_gguf.py
+16 -5
View File
@@ -122,8 +122,12 @@ def parse_args() -> argparse.Namespace:
help="Export only the multi-token prediction (MTP) head as a separate GGUF, suitable for use as a speculative draft. An 'mtp-' prefix will be added to the output file name.",
)
parser.add_argument(
"--no-mtp", action="store_true",
help="Exclude the multi-token prediction (MTP) head from the converted GGUF. Pair with --mtp on a second run to publish trunk and MTP as two files. Note: the split form duplicates embeddings, but even though the bundled default is more space-efficient overall, this allows differing quantization which may be more performant.",
"--no-nextn", "--no-mtp", dest="no_mtp", action="store_true",
help="Exclude NextN speculative draft tensors from the converted GGUF. Pair with --mtp or --dspark on a second run to publish target and draft as two files.",
)
parser.add_argument(
"--dspark", action="store_true",
help="Export only the DeepSeek-V4 DSpark draft tensors as a separate GGUF.",
)
parser.add_argument(
"--mistral-format", action="store_true",
@@ -254,13 +258,20 @@ def main() -> None:
from conversion.mistral import MistralModel
model_class = MistralModel
if args.mtp and args.no_mtp:
logger.error("--mtp and --no-mtp are mutually exclusive")
if sum((args.mtp, args.no_mtp, args.dspark)) > 1:
logger.error("--mtp, --no-nextn, and --dspark are mutually exclusive")
sys.exit(1)
if args.dspark:
if is_mistral_format or model_architecture != "DeepseekV4ForCausalLM":
logger.error("--dspark is only supported for DeepseekV4ForCausalLM")
sys.exit(1)
from conversion.deepseek import DeepseekV4DSparkModel
model_class = DeepseekV4DSparkModel
if args.mtp or args.no_mtp:
if not model_class.supports_mtp_export:
logger.error("--mtp / --no-mtp are not supported for %s", model_architecture)
logger.error("--mtp / --no-nextn are not supported for %s", model_architecture)
sys.exit(1)
if args.no_mtp:
model_class.no_mtp = True