feat(data): Smaller extract for human eval demo

Add an additional, limited JSONL evaluation job input manifest for
running shorter demonstrative human evaluation jobs (where 20
examples take quite a while to go through)

datasets/Prepare-SQuAD.ipynb

@@ -441,6 +441,32 @@
     "            )"
    ]
   },
+  {
+   "cell_type": "markdown",
+   "id": "edd8abd8-3837-48f0-bad5-ade36e39f4bc",
+   "metadata": {},
+   "source": [
+    "For **demonstrating human evaluation** jobs, even 20 examples might be a bit tedious to wade through in a workshop: So we'll create an even shorter dataset for that as a separate file:"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "id": "ee0b1c50-b6b8-4406-9eb7-b732c55c9f3e",
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "with open(\"question-answering/eval-job-input-qa.manifest.jsonl\") as fin:\n",
+    "    with open(\"question-answering/human-demo-input-qa.manifest.jsonl\", \"w\") as fout:\n",
+    "        for ix, line in enumerate(fin):\n",
+    "            if ix >= 10:\n",
+    "                break\n",
+    "            if ix > 0:\n",
+    "                fout.write(\"\\n\")\n",
+    "            fout.write(line.strip())\n",
+    "print(f\"Wrote {ix} records to human demo file\")"
+   ]
+  },
   {
    "cell_type": "markdown",
    "id": "b0e0d33e-aac0-41c0-b2b2-8ae40e9195f9",