From bc1c9c20e3cadb2a60163fa73370161a806124f0 Mon Sep 17 00:00:00 2001
From: Wing Lian <wing.lian@gmail.com>
Date: Mon, 13 Jan 2025 10:44:11 -0500
Subject: [PATCH] assume empty lora dropout means 0.0 and add tests (#2243)

* assume empty lora dropout means 0.0 and add tests

* remove un-necessary arg

* refactor based on pr feedback:

* chore: lint
---
 .../config/models/input/v0_4_1/__init__.py    |  7 ++
 tests/test_lora.py                            | 69 +++++++++++++++++++
 2 files changed, 76 insertions(+)
 create mode 100644 tests/test_lora.py

diff --git a/src/axolotl/utils/config/models/input/v0_4_1/__init__.py b/src/axolotl/utils/config/models/input/v0_4_1/__init__.py
index bb88a0baa..19ce7b18c 100644
--- a/src/axolotl/utils/config/models/input/v0_4_1/__init__.py
+++ b/src/axolotl/utils/config/models/input/v0_4_1/__init__.py
@@ -367,6 +367,13 @@ class LoraConfig(BaseModel):
             loraplus_lr_embedding = float(loraplus_lr_embedding)
         return loraplus_lr_embedding
 
+    @model_validator(mode="before")
+    @classmethod
+    def validate_lora_dropout(cls, data):
+        if data.get("adapter") is not None and data.get("lora_dropout") is None:
+            data["lora_dropout"] = 0.0
+        return data
+
 
 class ReLoRAConfig(BaseModel):
     """ReLoRA configuration subset"""
diff --git a/tests/test_lora.py b/tests/test_lora.py
new file mode 100644
index 000000000..b917ff3f9
--- /dev/null
+++ b/tests/test_lora.py
@@ -0,0 +1,69 @@
+"""
+tests for loading loras
+"""
+from axolotl.utils.config import normalize_config, validate_config
+from axolotl.utils.dict import DictDefault
+from axolotl.utils.models import load_model, load_tokenizer
+
+# pylint: disable=duplicate-code
+minimal_config = DictDefault(
+    {
+        "base_model": "HuggingFaceTB/SmolLM2-135M",
+        "learning_rate": 0.000001,
+        "datasets": [
+            {
+                "path": "mhenrichsen/alpaca_2k_test",
+                "type": "alpaca",
+            }
+        ],
+        "micro_batch_size": 1,
+        "gradient_accumulation_steps": 1,
+    }
+)
+
+
+class TestLoRALoad:
+    """
+    Test class for loading LoRA weights
+    """
+
+    def test_load_lora_weights(self):
+        cfg = DictDefault(
+            {
+                "base_model": "HuggingFaceTB/SmolLM2-135M",
+                "adapter": "lora",
+                "lora_r": 8,
+                "lora_alpha": 16,
+                "lora_dropout": 0.0,
+                "lora_target_linear": True,
+                "micro_batch_size": 1,
+                "gradient_accumulation_steps": 1,
+                "sequence_len": 1024,
+            }
+            | minimal_config
+        )
+        cfg = validate_config(cfg)
+        normalize_config(cfg)
+        tokenizer = load_tokenizer(cfg)
+        load_model(cfg, tokenizer)
+
+    def test_load_lora_weights_empty_dropout(self):
+        cfg = DictDefault(
+            {
+                "base_model": "HuggingFaceTB/SmolLM2-135M",
+                "adapter": "lora",
+                "lora_r": 8,
+                "lora_alpha": 16,
+                "lora_dropout": None,
+                "lora_target_linear": True,
+                "micro_batch_size": 1,
+                "gradient_accumulation_steps": 1,
+                "sequence_len": 1024,
+            }
+            | minimal_config
+        )
+        cfg = validate_config(cfg)
+        normalize_config(cfg)
+        assert cfg.lora_dropout == 0.0
+        tokenizer = load_tokenizer(cfg)
+        load_model(cfg, tokenizer)