Skip QLoRA unit tests broken by latest ao nightly (#568)

tests/torchtune/models/test_lora_llama2.py

@@ -181,6 +181,7 @@ def test_lora_llama2_state_dict_parity(
         assert not unexpected
         assert all(["lora" in key for key in missing])
 
+    @pytest.mark.skip(reason="broken by ao nightly")
     def test_lora_linear_quantize_base(self):
         model = self.get_lora_llama2(
             lora_modules=["q_proj", "v_proj", "k_proj", "output_proj"],
@@ -195,6 +196,7 @@ def test_lora_linear_quantize_base(self):
             if isinstance(module, LoRALinear):
                 assert module._quantize_base
 
+    @pytest.mark.skip(reason="broken by ao nightly")
     def test_qlora_llama2_parity(self, inputs):
         with utils.set_default_dtype(torch.bfloat16):
             model_ref = self.get_lora_llama2(
@@ -222,6 +224,7 @@ def test_qlora_llama2_parity(self, inputs):
         output = qlora(inputs)
         torch.testing.assert_close(ref_output, output)
 
+    @pytest.mark.skip(reason="broken by ao nightly")
     def test_qlora_llama2_state_dict(self):
         with utils.set_default_dtype(torch.bfloat16):
             model_ref = self.get_lora_llama2(
@@ -255,6 +258,7 @@ def test_qlora_llama2_state_dict(self):
             for v in qlora_sd.values():
                 assert v.dtype == torch.bfloat16
 
+    @pytest.mark.skip(reason="broken by ao nightly")
     def test_qlora_llama2_merged_state_dict(self):
         with utils.set_default_dtype(torch.bfloat16):
             qlora = self.get_lora_llama2(

tests/torchtune/modules/peft/test_lora.py

@@ -97,6 +97,7 @@ def test_forward(self, inputs, lora_linear, out_dim) -> None:
         assert actual.shape == (BSZ, SEQ_LEN, out_dim)
         torch.testing.assert_close(actual.mean(), expected, atol=1e-4, rtol=1e-6)
 
+    @pytest.mark.skip(reason="broken by ao nightly")
     def test_lora_weight_nf4_when_quantized(self, qlora_linear):
         assert isinstance(qlora_linear.weight, NF4Tensor)