bump flash attention 2.5.8 -> 2.6.1 (#1738)

* bump flash attention 2.5.8 -> 2.6.1 * use triton implementation of cross entropy from flash attn * add smoke test for flash attn cross entropy patch * fix args to xentropy.apply * handle tuple from triton loss fn * ensure the patch tests run independently * use the wrapper already built into flash attn for cross entropy * mark pytest as forked for patches * use pytest xdist instead of forked, since cuda doesn't like forking * limit to 1 process and use dist loadfile for pytest * change up pytest for fixture to reload transformers w monkeypathc
2024-07-14 19:11:31 -04:00
parent 219cd0d3c5
commit 98af5388ba
8 changed files with 103 additions and 14 deletions
--- a/tests/e2e/patched/test_fa_xentropy.py
+++ b/tests/e2e/patched/test_fa_xentropy.py
@@ -0,0 +1,87 @@
+"""
+E2E tests for lora llama
+"""
+
+import logging
+import os
+import unittest
+from importlib import reload
+from pathlib import Path
+
+import pytest
+from transformers.utils import is_torch_bf16_gpu_available
+
+from axolotl.cli import load_datasets
+from axolotl.common.cli import TrainerCliArgs
+from axolotl.train import train
+from axolotl.utils.config import normalize_config
+from axolotl.utils.dict import DictDefault
+
+from ..utils import with_temp_dir
+
+LOG = logging.getLogger("axolotl.tests.e2e")
+os.environ["WANDB_DISABLED"] = "true"
+
+
+@pytest.fixture(autouse=True)
+def reload_transformers():
+    import transformers.models.llama.modeling_llama
+
+    yield
+    reload(transformers.models.llama.modeling_llama)
+
+
+class TestFAXentropyLlama(unittest.TestCase):
+    """
+    Test case for Llama models using LoRA w multipack
+    """
+
+    @with_temp_dir
+    def test_lora_packing_fa_cross_entropy(self, temp_dir):
+        # pylint: disable=duplicate-code
+        cfg = DictDefault(
+            {
+                "base_model": "JackFram/llama-68m",
+                "tokenizer_type": "LlamaTokenizer",
+                "sequence_len": 1024,
+                "sample_packing": True,
+                "flash_attention": True,
+                "flash_attn_cross_entropy": True,
+                "load_in_8bit": True,
+                "adapter": "lora",
+                "lora_r": 32,
+                "lora_alpha": 64,
+                "lora_dropout": 0.05,
+                "lora_target_linear": True,
+                "val_set_size": 0.2,
+                "special_tokens": {
+                    "unk_token": "<unk>",
+                    "bos_token": "<s>",
+                    "eos_token": "</s>",
+                },
+                "datasets": [
+                    {
+                        "path": "mhenrichsen/alpaca_2k_test",
+                        "type": "alpaca",
+                    },
+                ],
+                "num_epochs": 1,
+                "micro_batch_size": 8,
+                "gradient_accumulation_steps": 1,
+                "output_dir": temp_dir,
+                "learning_rate": 0.00001,
+                "optimizer": "adamw_torch",
+                "lr_scheduler": "cosine",
+            }
+        )
+        if is_torch_bf16_gpu_available():
+            cfg.bf16 = True
+        else:
+            cfg.fp16 = True
+
+        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()