From 326daf095f832f90ada22c417a73df381e37e6c1 Mon Sep 17 00:00:00 2001
From: Dattu Sharma <venkatadattasainimmaturi@gmail.com>
Date: Mon, 28 Apr 2025 18:41:23 +0000
Subject: [PATCH] fixup qwen3

Signed-off-by: Dattu Sharma <venkatadattasainimmaturi@gmail.com>
---
 unsloth/models/_utils.py | 2 +-
 unsloth/models/llama.py  | 2 ++
 unsloth/models/qwen3.py  | 2 +-
 3 files changed, 4 insertions(+), 2 deletions(-)

diff --git a/unsloth/models/_utils.py b/unsloth/models/_utils.py
index e60881fa9c..282aa99334 100644
--- a/unsloth/models/_utils.py
+++ b/unsloth/models/_utils.py
@@ -248,7 +248,7 @@ model_architectures = ["llama", "mistral", "gemma", "gemma2", "qwen2", "granite"
 for model_name in model_architectures:
     config_filepath = f"transformers.models.{model_name}.configuration_{model_name}"
     model_filepath = f"transformers.models.{model_name}.modeling_{model_name}"
-    config_filename = f"{model_name.title()}Config"
+    config_filename = f"{model_name.title().replace('_','')}Config" # qwen3 arch folder is qwen3_moe but config is Qwen3Config. Need to remove underscore(_) for now
     exec(f"from {config_filepath} import {config_filename}", globals())
 
     try:
diff --git a/unsloth/models/llama.py b/unsloth/models/llama.py
index 722b50d27a..9c218d3e7d 100644
--- a/unsloth/models/llama.py
+++ b/unsloth/models/llama.py
@@ -2487,6 +2487,8 @@ class FastLlamaModel:
         elif model_type == "gemma2":  apply_lora_mlp = apply_lora_mlp_geglu_approx
         elif model_type == "cohere":  apply_lora_mlp = apply_lora_mlp_swiglu
         elif model_type == "granite": apply_lora_mlp = apply_lora_mlp_swiglu
+        elif model_type == "qwen3":   apply_lora_mlp = apply_lora_mlp_swiglu
+        elif model_type == "qwen3moe":   apply_lora_mlp = apply_lora_mlp_swiglu
         else:
             raise NotImplementedError(f"Unsloth: {model_type} is not yet implemented!")
         pass
diff --git a/unsloth/models/qwen3.py b/unsloth/models/qwen3.py
index aa99aa88db..9d658cf849 100644
--- a/unsloth/models/qwen3.py
+++ b/unsloth/models/qwen3.py
@@ -216,7 +216,7 @@ class FastQwen3Model(FastLlamaModel):
         # https://github.com/huggingface/transformers/pull/27931
         # https://github.com/huggingface/transformers/blob/v4.37.2/src/transformers/models/llama/modeling_llama.py
         import transformers.models.qwen3.modeling_qwen3
-        transformers.models.Qwen3.modeling_qwen3.Qwen3RotaryEmbedding = LlamaRotaryEmbedding
+        transformers.models.qwen3.modeling_qwen3.Qwen3RotaryEmbedding = LlamaRotaryEmbedding
         return
     pass