File size: 3,036 Bytes
477fa2f
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
{
 "cells": [
  {
   "attachments": {},
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Building an eval for LAMBADA\n",
    "\n",
    "We show how to build an eval for the LAMBADA dataset"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "# Download LAMBADA from https://zenodo.org/record/2630551 and place in examples/lambada-dataset\n",
    "!ls lambada-dataset\n",
    "import os\n",
    "import pandas as pd\n",
    "\n",
    "registry_pth = os.path.join(\"..\", \"evals\", \"registry\")\n",
    "os.makedirs(os.path.join(registry_pth, \"data\", \"lambada\"), exist_ok=True)\n",
    "\n",
    "def create_chat_prompt(text):\n",
    "    return [\n",
    "        {\"role\": \"system\", \"content\": \"Please complete the passages with the correct next word.\"}, \n",
    "        {\"role\": \"user\", \"content\": text}\n",
    "    ]\n",
    "\n",
    "df = pd.read_csv('lambada-dataset/lambada_test_plain_text.txt', sep=\"\\t\", names=[\"text\"])\n",
    "df[\"text\"] = df[\"text\"].str.split(\" \")\n",
    "df[\"input\"], df[\"ideal\"] = df[\"text\"].str[:-1].str.join(\" \").apply(create_chat_prompt), df[\"text\"].str[-1]\n",
    "df = df[[\"input\", \"ideal\"]]\n",
    "df.to_json(os.path.join(registry_pth, \"data/lambada/samples.jsonl\"), orient=\"records\", lines=True)\n",
    "display(df.head())\n",
    "\n",
    "eval_yaml = \"\"\"\n",
    "lambada:\n",
    "  id: lambada.test.v1\n",
    "  metrics: [accuracy]\n",
    "lambada.test.v1:\n",
    "  class: evals.elsuite.basic.match:Match\n",
    "  args:\n",
    "    samples_jsonl: lambada/samples.jsonl\n",
    "\"\"\".strip()\n",
    "with open(os.path.join(registry_pth, \"evals\", \"lambada.yaml\"), \"w\") as f:\n",
    "    f.write(eval_yaml)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "!oaieval gpt-3.5-turbo lambada --max_samples 20"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "# Inspect samples\n",
    "log_path = None # Set to jsonl path to logs from oaieval\n",
    "events = f\"/tmp/evallogs/{log_path}\"\n",
    "with open(events, \"r\") as f:\n",
    "    events_df = pd.read_json(f, lines=True)\n",
    "for i, r in pd.json_normalize(events_df[events_df.type == \"sampling\"].data).iterrows():\n",
    "    print(r)\n",
    "    print(f\"Prompt: {r.prompt}\")\n",
    "    print(f\"Sampled: {r.sampled}\")\n",
    "    print(\"-\" * 25)"
   ]
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "base",
   "language": "python",
   "name": "python3"
  },
  "language_info": {
   "codemirror_mode": {
    "name": "ipython",
    "version": 3
   },
   "file_extension": ".py",
   "mimetype": "text/x-python",
   "name": "python",
   "nbconvert_exporter": "python",
   "pygments_lexer": "ipython3",
   "version": "3.10.9"
  },
  "orig_nbformat": 4
 },
 "nbformat": 4,
 "nbformat_minor": 2
}