Skip to content

Commit 5deee92

Browse files
authored
Support Qwen3
1 parent 420eb3f commit 5deee92

File tree

1 file changed

+19
-0
lines changed

1 file changed

+19
-0
lines changed

convert_hf_to_gguf.py

Lines changed: 19 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -2390,6 +2390,25 @@ def prepare_tensors(self):
23902390
raise ValueError(f"Unprocessed experts: {experts}")
23912391

23922392

2393+
@Model.register("Qwen3ForCausalLM")
2394+
class Qwen3Model(Model):
2395+
model_arch = gguf.MODEL_ARCH.QWEN3
2396+
2397+
def set_vocab(self):
2398+
try:
2399+
self._set_vocab_sentencepiece()
2400+
except FileNotFoundError:
2401+
self._set_vocab_gpt2()
2402+
2403+
def set_gguf_parameters(self):
2404+
super().set_gguf_parameters()
2405+
if self.hparams.get("rope_scaling") is not None and "factor" in self.hparams["rope_scaling"]:
2406+
if self.hparams["rope_scaling"].get("type") == "yarn":
2407+
self.gguf_writer.add_rope_scaling_type(gguf.RopeScalingType.YARN)
2408+
self.gguf_writer.add_rope_scaling_factor(self.hparams["rope_scaling"]["factor"])
2409+
self.gguf_writer.add_rope_scaling_orig_ctx_len(self.hparams["rope_scaling"]["original_max_position_embeddings"])
2410+
2411+
23932412
@Model.register("GPT2LMHeadModel")
23942413
class GPT2Model(Model):
23952414
model_arch = gguf.MODEL_ARCH.GPT2

0 commit comments

Comments
 (0)