CS348Project/llama.cpp
5
0
Fork 0
mirror of https://github.com/ggml-org/llama.cpp.git synced 2025-10-27 08:21:30 +00:00
Code Issues Packages Projects Releases Wiki Activity

model : add OLMo3 support (#16015)

Browse Source 
* Add HF to gguf conversion logic for Olmo3

* Add Olmo3 implementation

* Update rope comment

* Fix indentation

Co-authored-by: Sigbjørn Skjæret <sigbjorn.skjaeret@scala.com>

* Apply suggestion from @CISC

Co-authored-by: Sigbjørn Skjæret <sigbjorn.skjaeret@scala.com>

---------

Co-authored-by: Sigbjørn Skjæret <sigbjorn.skjaeret@scala.com>
This commit is contained in:
Shane A
2025-09-17 00:01:58 -07:00
committed by GitHub
parent d5fabe3682
commit 85286f3548
2 changed files with 68 additions and 4 deletions
Download Patch File Download Diff File Expand all files Collapse all files

25
convert_hf_to_gguf.py
View File

@@ -6009,9 +6009,34 @@ class SeedOssModel(TextModel):
@ModelBase.register("Olmo2ForCausalLM")
@ModelBase.register("Olmo3ForCausalLM")
class Olmo2Model(TextModel):
model_arch = gguf.MODEL_ARCH.OLMO2
def set_gguf_parameters(self):
super().set_gguf_parameters()
rope_scaling = self.hparams.get("rope_scaling") or {}
if rope_scaling.get("rope_type", rope_scaling.get("type")) == "yarn" and "factor" in rope_scaling:
self.gguf_writer.add_rope_scaling_type(gguf.RopeScalingType.YARN)
self.gguf_writer.add_rope_scaling_factor(rope_scaling["factor"])
self.gguf_writer.add_rope_scaling_attn_factors(rope_scaling["attention_factor"])
self.gguf_writer.add_rope_scaling_orig_ctx_len(rope_scaling["original_max_position_embeddings"])
if "sliding_window" in self.hparams:
self.gguf_writer.add_sliding_window(self.hparams["sliding_window"])
sliding_window_pattern = []
if "layer_types" in self.hparams:
sliding_window_pattern = [t == "sliding_attention" for t in self.hparams["layer_types"]]
else:
# Olmo2 does not use sliding window attention.
# Olmo3 defaults to using sliding window for all layers except every 4th.
for i in range(self.hparams["num_hidden_layers"]):
sliding_window_pattern.append((i + 1) % 4 != 0)
self.gguf_writer.add_sliding_window_pattern(sliding_window_pattern)
@ModelBase.register("OlmoeForCausalLM")
class OlmoeModel(TextModel):