Fix 2 typos in 04_preferene-tuning-with-dpo (#356)

ch07/04_preference-tuning-with-dpo/dpo-from-scratch.ipynb

@@ -2774,7 +2774,7 @@
    },
    "source": [
     "- As we can see above, the loss continues to improve, which is a good sign\n",
-    "- Based on the downward slope, one might be tempted to train the model a bit further (and readers are encouraged to try this), but not that DPO is prone to collapse, where the model may start generating nonsensical responses\n",
+    "- Based on the downward slope, one might be tempted to train the model a bit further (and readers are encouraged to try this), but note that DPO is prone to collapse, where the model may start generating nonsensical responses\n",
     "- Next, let's take a look at the reward margins:"
    ]
   },
@@ -2823,7 +2823,7 @@
    },
    "source": [
     "- As we can see, and as it's desired, the reward margins improve; this mirrors the loss curve and is a good sign\n",
-    "- Note that DPO losses and reward margins are valuable metrics to track during training; however, they don't tell the whole store\n",
+    "- Note that DPO losses and reward margins are valuable metrics to track during training; however, they don't tell the whole story\n",
     "- Lastly, and most importantly, we have to conduct a qualitative check of the responses\n",
     "- Here, we will look at the response (in addition, you could use an LLM to score the responses similar to chapter 7)"
    ]