From efa1209a92a1ab19c297e2f38836dde2acb7b67e Mon Sep 17 00:00:00 2001
From: Wing Lian <wing.lian@gmail.com>
Date: Wed, 30 Oct 2024 15:40:27 -0400
Subject: [PATCH] add smoke test training

---
 src/axolotl/core/trainer_builder.py |  1 +
 tests/e2e/test_optimizers.py        | 41 +++++++++++++++++++++++++++++
 2 files changed, 42 insertions(+)

diff --git a/src/axolotl/core/trainer_builder.py b/src/axolotl/core/trainer_builder.py
index 7537f80fe2..265cfeb8c0 100755
--- a/src/axolotl/core/trainer_builder.py
+++ b/src/axolotl/core/trainer_builder.py
@@ -1638,6 +1638,7 @@ def build(self, total_num_steps):
             trainer_kwargs["max_length"] = self.cfg.sequence_len
 
         if self.cfg.optimizer in [
+            # pylint: disable=duplicate-code
             "optimi_adamw",
             "ao_adamw_4bit",
             "ao_adamw_8bit",
diff --git a/tests/e2e/test_optimizers.py b/tests/e2e/test_optimizers.py
index 119dd3d7cf..24985dde0f 100644
--- a/tests/e2e/test_optimizers.py
+++ b/tests/e2e/test_optimizers.py
@@ -65,3 +65,44 @@ def test_optimi_adamw(self, temp_dir):
 
         train(cfg=cfg, cli_args=cli_args, dataset_meta=dataset_meta)
         assert (Path(temp_dir) / "adapter_model.bin").exists()
+
+    @with_temp_dir
+    def test_soap(self, temp_dir):
+        # pylint: disable=duplicate-code
+        cfg = DictDefault(
+            {
+                "base_model": "HuggingFaceTB/SmolLM-135M",
+                "sequence_len": 1024,
+                "load_in_8bit": True,
+                "adapter": "lora",
+                "lora_r": 8,
+                "lora_alpha": 16,
+                "lora_dropout": 0.05,
+                "lora_target_linear": True,
+                "val_set_size": 0.1,
+                "special_tokens": {
+                    "pad_token": "<|endoftext|>",
+                },
+                "datasets": [
+                    {
+                        "path": "vicgalle/alpaca-gpt4",
+                        "type": "alpaca",
+                    },
+                ],
+                "num_epochs": 1,
+                "micro_batch_size": 8,
+                "gradient_accumulation_steps": 1,
+                "output_dir": temp_dir,
+                "learning_rate": 0.00001,
+                "optimizer": "soap",
+                "optim_soap_beta1": 0.95,
+                "optim_soap_beta2": 0.95,
+                "lr_scheduler": "cosine",
+            }
+        )
+        normalize_config(cfg)
+        cli_args = TrainerCliArgs()
+        dataset_meta = load_datasets(cfg=cfg, cli_args=cli_args)
+
+        train(cfg=cfg, cli_args=cli_args, dataset_meta=dataset_meta)
+        assert (Path(temp_dir) / "adapter_model.bin").exists()