upstream-archive/llama.cpp
1
0
Fork 0
mirror of https://github.com/ggml-org/llama.cpp.git synced 2026-10-03 11:27:25 -05:00
Code Issues Packages Projects Releases Wiki Activity

model : Add LFM2.5-Encoder-350M and LFM2.5-Encoder-230M (#29862)

Browse Source 
Register `Lfm2BidirectionalForMaskedLM` architecture for LFM2.5-Encoder
models.
This commit is contained in:
Tarek Dakhran
2026-10-03 08:44:45 +02:00
committed by GitHub
parent 889edf43dd
commit cb7934c52c
2 changed files with 6 additions and 3 deletions
Download Patch File Download Diff File Expand all files Collapse all files
conversion/__init__.py
+1
View File
@@ -155,6 +155,7 @@ TEXT_MODEL_MAP: dict[str, str] = {
"LevModel": "lev",
"NimbleModel": "lev",
"Lfm25AudioTokenizer": "lfm2",
"Lfm2BidirectionalForMaskedLM": "lfm2",
"Lfm2BidirectionalModel": "lfm2",
"Lfm2ForCausalLM": "lfm2",
"Lfm2Model": "lfm2",
conversion/lfm2.py
+5 -3
View File
@@ -65,19 +65,21 @@ class LFM2Model(TextModel):
yield from super().modify_tensors(data_torch, name, bid)
@ModelBase.register("Lfm2Model", "Lfm2BidirectionalModel")
@ModelBase.example("LiquidAI/LFM2.5-ColBERT-350M", "LiquidAI/LFM2.5-Embedding-350M")
@ModelBase.register("Lfm2Model", "Lfm2BidirectionalModel", "Lfm2BidirectionalForMaskedLM")
@ModelBase.example("LiquidAI/LFM2.5-ColBERT-350M", "LiquidAI/LFM2.5-Embedding-350M", "LiquidAI/LFM2.5-Encoder-350M", "LiquidAI/LFM2.5-Encoder-230M")
class LFM2ColBertModel(LFM2Model):
model_arch = gguf.MODEL_ARCH.LFM2
dense_tensor_name = "dense_2"
def set_gguf_parameters(self):
super().set_gguf_parameters()
if self.hf_arch == "Lfm2BidirectionalModel":
if self.hf_arch in ("Lfm2BidirectionalModel", "Lfm2BidirectionalForMaskedLM"):
self.gguf_writer.add_causal_attention(False)
self._try_set_pooling_type()
def modify_tensors(self, data_torch: Tensor, name: str, bid: int | None) -> Iterable[tuple[str, Tensor]]:
# masked LM checkpoints use "lfm2." prefix
name = name.removeprefix("lfm2.")
if not name.startswith(self.dense_tensor_name):
name = "model." + name