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 : skip bias_vl tensor in DeepSeek-V4 DSpark conversion (#28294)

Browse Source 
* convert : skip bias_vl tensor in DeepSeek-V4 DSpark conversion

The DFLASH arch does not include FFN_EXP_PROBS_B_VL, so the DSpark
conversion failed when it tried to write the mtmd-only hash routing
tensor ffn.gate.bias_vl. Drop it like the tid2eid tensor; the DFLASH
draft only consumes ffn.gate.bias via FFN_EXP_PROBS_B.

Assisted-by: pi:llama.cpp/DeepSeek-V4-Flash-0731

* cont : fix

Co-authored-by: Sigbjørn Skjæret <[email protected]>

---------

Co-authored-by: Sigbjørn Skjæret <[email protected]>
This commit is contained in:
Georgi Gerganov
2026-09-03 10:37:23 +03:00
committed by GitHub
co-authored by Sigbjørn Skjæret
parent 5ec4eab69e
commit d646c9d155
1 changed files with 7 additions and 0 deletions
Download Patch File Download Diff File Expand all files Collapse all files
conversion/deepseek.py
+7
View File
@@ -1007,6 +1007,13 @@ class DeepseekV4DSparkModel(DeepseekV4Model):
return self._DSPARK_ROOT_MAP[name]
return super()._map_dsv4_tensor_name(name, bid)
def modify_tensors(self, data_torch: Tensor, name: str, bid: int | None) -> Iterable[tuple[str, Tensor]]:
# the DFlash draft uses the plain exp-probs bias (ffn.gate.bias -> FFN_EXP_PROBS_B);
# the mtmd-only hash routing tensors (bias_vl, tid2eid) are not part of the DFLASH arch
if name.endswith(".ffn.gate.bias_vl"):
return
yield from super().modify_tensors(data_torch, name, bid)
def set_vocab(self):
if self.target_model_dir is None:
raise ValueError("DeepSeek-V4 DSpark requires --target-model-dir with the target tokenizer")