diff --git a/intro.ipynb b/intro.ipynb
index d4f22a41bd..d7e4fbb656 100644
--- a/intro.ipynb
+++ b/intro.ipynb
@@ -1,5 +1,21 @@
 {
  "cells": [
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "outputs": [],
+   "source": [
+    "import os\n",
+    "os.environ[\"DISABLE_MLFLOW_INTEGRATION\"] = \"true\"\n",
+    "os.environ[\"WANDB_DISABLED\"] = \"true\""
+   ],
+   "metadata": {
+    "collapsed": false,
+    "pycharm": {
+     "name": "#%%\n"
+    }
+   }
+  },
   {
    "cell_type": "code",
    "execution_count": null,
@@ -18,23 +34,28 @@
   },
   {
    "cell_type": "code",
+   "execution_count": null,
+   "outputs": [],
    "source": [
+    "# Basic configuration for\n",
     "cfg = DictDefault({\n",
     "    \"base_model\": \"TinyLlama/TinyLlama-1.1B-intermediate-step-1431k-3T\",\n",
-    "    \"load_in_8bit\": True,\n",
+    "    \"model_kwargs\": {\"from_tf\": True},\n",
+    "    \"load_in_4bit\": True,\n",
     "    \"datasets\": [{\"path\": \"mhenrichsen/alpaca_2k_test\", \"type\": \"alpaca\"}],\n",
     "    \"val_set_size\": 0.1,\n",
     "    \"output_dir\": \"./lora-out\",\n",
     "    \"sequence_len\": 2048,\n",
     "    \"sample_packing\": True,\n",
     "    \"pad_to_sequence_len\": True,\n",
-    "    \"adapter\": \"lora\",\n",
+    "    \"adapter\": \"qlora\",\n",
     "    \"lora_r\": 32,\n",
     "    \"lora_alpha\": 16,\n",
     "    \"lora_dropout\": 0.1,\n",
-    "    \"lora_target_linear\": true,\n",
+    "    \"lora_target_linear\": True,\n",
     "    \"gradient_accumulation_steps\": 4,\n",
-    "    \"micro_batch_size\": 2,\n",
+    "    \"micro_batch_size\": 1,\n",
+    "    \"eval_batch_size\": 1,\n",
     "    \"num_epochs\": 1,\n",
     "    \"optimizer\": \"adamw_bnb_8bit\",\n",
     "    \"lr_scheduler\": \"cosine\",\n",
@@ -47,13 +68,12 @@
     "    \"evals_per_epoch\": 4,\n",
     "    \"saves_per_epoch\": 1,\n",
     "    \"weight_decay\": 0.0,\n",
-    "    \"wandb_disabled\": True,\n",
     "})\n"
    ],
    "metadata": {
     "collapsed": false,
     "pycharm": {
-     "name": "#%% md\n"
+     "name": "#%%\n"
     }
    }
   },
@@ -62,6 +82,7 @@
    "execution_count": null,
    "outputs": [],
    "source": [
+    "# Kick off the training\n",
     "model, tokenizer = do_train(cfg, TrainerCliArgs())\n"
    ],
    "metadata": {
diff --git a/src/axolotl/core/trainer_builder.py b/src/axolotl/core/trainer_builder.py
index 18dc353a23..6a07658df6 100644
--- a/src/axolotl/core/trainer_builder.py
+++ b/src/axolotl/core/trainer_builder.py
@@ -745,9 +745,10 @@ def build(self, total_num_steps):
         training_arguments_kwargs[
             "per_device_train_batch_size"
         ] = self.cfg.micro_batch_size
-        training_arguments_kwargs[
-            "per_device_eval_batch_size"
-        ] = self.cfg.eval_batch_size
+        if self.cfg.eval_batch_size:
+            training_arguments_kwargs[
+                "per_device_eval_batch_size"
+            ] = self.cfg.eval_batch_size
         training_arguments_kwargs[
             "gradient_accumulation_steps"
         ] = self.cfg.gradient_accumulation_steps
diff --git a/src/axolotl/utils/bench.py b/src/axolotl/utils/bench.py
index 40be0d9ac8..8f33665c69 100644
--- a/src/axolotl/utils/bench.py
+++ b/src/axolotl/utils/bench.py
@@ -20,7 +20,8 @@ def wrapper(*args, **kwargs):
             device = kwargs.get("device", args[0] if args else None)
 
             if (
-                not torch.cuda.is_available()
+                device is None
+                or not torch.cuda.is_available()
                 or device == "auto"
                 or torch.device(device).type == "cpu"
             ):