diff --git a/.github/workflows/e2e.yml b/.github/workflows/e2e.yml
index ada1fd0c48..09c26c2a67 100644
--- a/.github/workflows/e2e.yml
+++ b/.github/workflows/e2e.yml
@@ -24,6 +24,7 @@ jobs:
       - name: Install dependencies
         run: |
           pip3 install -e .
+          pip3 install flash-attn
           pip3 install -r requirements-tests.txt
 
       - name: Run e2e tests
diff --git a/src/axolotl/utils/bench.py b/src/axolotl/utils/bench.py
index 30f0985e75..b460b2ba7c 100644
--- a/src/axolotl/utils/bench.py
+++ b/src/axolotl/utils/bench.py
@@ -2,6 +2,7 @@
 
 import pynvml
 import torch
+from pynvml.nvml import NVMLError
 
 
 def gpu_memory_usage(device=0):
@@ -20,11 +21,13 @@ def gpu_memory_usage_smi(device=0):
         device = device.index
     if isinstance(device, str) and device.startswith("cuda:"):
         device = int(device[5:])
-
-    pynvml.nvmlInit()
-    handle = pynvml.nvmlDeviceGetHandleByIndex(device)
-    info = pynvml.nvmlDeviceGetMemoryInfo(handle)
-    return info.used / 1024.0**3
+    try:
+        pynvml.nvmlInit()
+        handle = pynvml.nvmlDeviceGetHandleByIndex(device)
+        info = pynvml.nvmlDeviceGetMemoryInfo(handle)
+        return info.used / 1024.0**3
+    except NVMLError:
+        return 0.0
 
 
 def log_gpu_memory_usage(log, msg, device):
diff --git a/src/axolotl/utils/config.py b/src/axolotl/utils/config.py
index 90ed409b9c..a31f34b73e 100644
--- a/src/axolotl/utils/config.py
+++ b/src/axolotl/utils/config.py
@@ -29,7 +29,7 @@ def get_device():
         cfg.device_map = "auto"
     else:
         if cfg.device.startswith("cuda"):
-            cfg.device_map = {"": cfg.local_rank}
+            cfg.device_map = {"": torch.cuda.current_device()}
         else:
             cfg.device_map = {"": cfg.device}
 
diff --git a/tests/e2e/test_lora_llama.py b/tests/e2e/test_lora_llama.py
index 7873b7ec20..905c3711fd 100644
--- a/tests/e2e/test_lora_llama.py
+++ b/tests/e2e/test_lora_llama.py
@@ -78,3 +78,45 @@ def test_lora(self):
         dataset_meta = load_datasets(cfg=cfg, cli_args=cli_args)
 
         train(cfg=cfg, cli_args=cli_args, dataset_meta=dataset_meta)
+
+    def test_lora_packing(self):
+        cfg = DictDefault(
+            {
+                "base_model": "JackFram/llama-68m",
+                "base_model_config": "JackFram/llama-68m",
+                "tokenizer_type": "LlamaTokenizer",
+                "sequence_len": 1024,
+                "sample_packing": True,
+                "flash_attention": True,
+                "load_in_8bit": True,
+                "adapter": "lora",
+                "lora_r": 32,
+                "lora_alpha": 64,
+                "lora_dropout": 0.05,
+                "lora_target_linear": True,
+                "val_set_size": 0.1,
+                "special_tokens": {
+                    "unk_token": "<unk>",
+                    "bos_token": "<s>",
+                    "eos_token": "</s>",
+                },
+                "datasets": [
+                    {
+                        "path": "mhenrichsen/alpaca_2k_test",
+                        "type": "alpaca",
+                    },
+                ],
+                "num_epochs": 2,
+                "micro_batch_size": 8,
+                "gradient_accumulation_steps": 1,
+                "output_dir": tempfile.mkdtemp(),
+                "learning_rate": 0.00001,
+                "optimizer": "adamw_torch",
+                "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)