CS348Project/llama.cpp
5
0
Fork 0
mirror of https://github.com/ggml-org/llama.cpp.git synced 2025-11-05 09:36:52 +00:00
Code Issues Packages Projects Releases Wiki Activity

override modify_tensors instead of get_tensors

Browse Source
This commit is contained in:
ibrahimkhadraoui
2025-07-07 12:00:57 +04:00
parent 53304c84db
commit 441d8d66bd
1 changed files with 12 additions and 16 deletions
Download Patch File Download Diff File Expand all files Collapse all files

16
convert_hf_to_gguf.py
View File

@@ -6600,23 +6600,19 @@ class FalconH1Model(Mamba2Model):
return mup_vector
def get_tensors(self) -> Iterator[tuple[str, Tensor]]:
for name, tensor in super().get_tensors():
if name.startswith("model.backbone") or name.startswith("model.lm_head"):
name = name.removeprefix("model.")
yield name, tensor
def modify_tensors(self, data_torch: Tensor, name: str, bid: int | None) -> Iterable[tuple[str, Tensor]]:
tensors = list(super().modify_tensors(data_torch, name, bid))
if self.ssm_multipliers is not None:
# Insert MUP vector after mamba.dt_bias
if "mamba.dt_bias" in name:
if self.ssm_multipliers is not None and "mamba.dt_bias" in name:
block_match = re.search(r"(?:model\.layers\.)?(\d+)\.mamba\.dt_bias", name)
if block_match:
block_id = int(block_match.group(1))
# Generate MUP vector with correct name format
mup_tensor = self._generate_mup_vector(block_id)
mup_name = f"blk.{block_id}.ssm_mup_vec"
logger.debug(f"Inserting MUP vector for block {block_id}: {mup_name}")
yield mup_name, mup_tensor
tensors.append((self.map_tensor_name(mup_name), mup_tensor))
return tensors
def set_gguf_parameters(self):
super().set_gguf_parameters()