Apply Qwen3 PR from llama.cpp (#355)

1 files changed, 7 insertions, 0 deletions

diff --git a/convert_hf_to_gguf.py b/convert_hf_to_gguf.py
index a6ab09c0..20d27a5c 100755
--- a/convert_hf_to_gguf.py
+++ b/convert_hf_to_gguf.py
@@ -1938,6 +1938,13 @@ class Qwen2MoeModel(Model):
             if len(experts) > 0:
                 raise ValueError(f"Unprocessed experts: {experts}")
 
+@Model.register("Qwen3ForCausalLM")
+class Qwen3Model(Qwen2Model):
+    model_arch = gguf.MODEL_ARCH.QWEN3
+
+@Model.register("Qwen3MoeForCausalLM")
+class Qwen3MoeModel(Qwen2MoeModel):
+    model_arch = gguf.MODEL_ARCH.QWEN3MOE
 
 @Model.register("GPT2LMHeadModel")
 class GPT2Model(Model):