gpt-2 left context out files

run.ipynb

@@ -0,0 +1,117 @@
+{
+ "cells": [
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "import torch \n",
+    "from transformers import GPT2LMHeadModel, GPT2Tokenizer\n",
+    "\n",
+    "import numpy as np\n",
+    "\n",
+    "device = torch.device(\"cuda\" if torch.cuda.is_available() else \"cpu\")"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "torch.__version__, device"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "tokenizer = GPT2Tokenizer.from_pretrained('gpt2-medium')\n",
+    "\n",
+    "model = GPT2LMHeadModel.from_pretrained('gpt2-medium')\n",
+    "model.to(device)\n"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "import lzma \n",
+    "\n",
+    "\n",
+    "def file_iterator(file_path):\n",
+    "    print(file_path, file_path.endswith(\".xz\"))\n",
+    "    if file_path.endswith(\".xz\"):\n",
+    "        with lzma.open(file_path, mode=\"r\") as fp:\n",
+    "            for line in fp.readlines():\n",
+    "                yield line.decode(\"utf-8\")#.split(\"\\t\")[7]\n",
+    "    else:\n",
+    "        with open(file_path, \"r\", encoding=\"utf-8\") as fp:\n",
+    "            for line in fp.readlines():\n",
+    "                yield line\n",
+    "\n",
+    "def clear_line(line):\n",
+    "    return line.lower().strip(\"\\n\").replace(\"\\\\n\", \"\")\n",
+    "\n"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "K = 20\n",
+    "for file_path in ('dev-0', 'test-A'):\n",
+    "    print('Working on file from folder:', file_path)\n",
+    "    data_iterator = file_iterator(f'{file_path}/in.tsv.xz')\n",
+    "    with open(f'{file_path}/out-tr-dec.tsv', 'w', encoding='utf-8') as fp:\n",
+    "        for line in data_iterator:\n",
+    "            # print([(i, part) for i, part in enumerate(line.split('\\t'))])\n",
+    "            left_context = clear_line(line.split('\\t')[6])\n",
+    "            # print(left_context)\n",
+    "            inputs = tokenizer.encode(left_context, return_tensors='pt').to(device)\n",
+    "            preds = model(inputs)\n",
+    "            # print('\\n', preds)\n",
+    "            z_dist = preds[0][0][-1]\n",
+    "            probability_distances = torch.softmax(preds[0][0][-1], dim=0)\n",
+    "            top_k = probability_distances.topk(K)\n",
+    "            # print(top_k)\n",
+    "            results = [f'{tokenizer.decode([idx])}:{value}' for value, idx in zip(top_k.values, top_k.indices)]\n",
+    "            # print(results)\n",
+    "            line_to_write = ' '.join(results) + f' :{1 - torch.sum(top_k.values)}\\n'\n",
+    "            # print(line_to_write)\n",
+    "            fp.write(line_to_write)\n",
+    "        #     break\n",
+    "        # break"
+   ]
+  }
+ ],
+ "metadata": {
+  "kernelspec": {
+   "display_name": "mj_venv",
+   "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": "3.10.11"
+  },
+  "orig_nbformat": 4
+ },
+ "nbformat": 4,
+ "nbformat_minor": 2
+}

run.py

@@ -1,24 +0,0 @@
-#!/usr/bin/python3
-import sys
-
-
-for line in sys.stdin:
-    _, _, _, _, _, _, left_context, right_context = line.split("\t")
-    
-    previous_word = left_context.split()[-1]
-    next_word = right_context.split()[0]
-    
-    if previous_word == "United" and next_word == "of":
-        print("States:0.9 :0.1")
-    elif previous_word == "used":
-        print("to:0.4 it:0.3 as:0.2 :0.1")
-    elif previous_word.lower() == "in":
-        print("the:0.7 a:0.1 an:0.1 :0.1")
-    elif previous_word.lower() == "i":
-        print("am:0.3 was:0.3 have:0.3 :0.1")
-    elif previous_word.lower() in ["he", "she", "it"]:
-        print("is:0.3 was:0.3 has:0.3 :0.1")
-    elif previous_word.lower() in "bring":
-        print("something:0.3 it:0.3 them:0.3 :0.1")
-    else:
-        print("the:0.5 a:0.2 an:0.2 :0.1")