gpt-2 on left context

.gitignore

@@ -0,0 +1,8 @@
+
+*~
+*.swp
+*.bak
+*.pyc
+*.o
+.DS_Store
+.token

README.md

@@ -0,0 +1,21 @@
+
+wmt-2020-pl-en
+==========================================
+
+Translate from Polish to English.
+
+This is a challenge created from http://www.statmt.org/wmt20/translation-task.html . Train set is created from europarl wmt pl-en training data. Dev and test set are created from wmt pl-en development data.
+
+Directory structure
+-------------------
+
+* `README.md` — this file
+* `config.txt` — configuration file
+* `train/` — directory with training data
+* `train/train.tsv` — sample parallel corpus (Finnish text in the first column, Māori text in the second one)
+* `dev-0/` — directory with dev (test) data
+* `dev-0/in.tsv` — Finnish input text for the dev set
+* `dev-0/expected.tsv` — Māori reference translation for the dev set
+* `test-A` — directory with test data
+* `test-A/in.tsv` — Finnish input data for the test set
+* `test-A/expected.tsv` — Māori reference translation for the test set

config.txt

@@ -0,0 +1 @@
+--metric BLEU --precision 4 --tokenizer 13a -%

run-transf-dec.ipynb

@@ -0,0 +1,113 @@
+{
+ "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\")\n",
+    "\n",
+    "torch.__version__, device\n"
+   ]
+  },
+  {
+   "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",
+    "\n",
+    "def clear_line(line):\n",
+    "    return line.lower().strip(\"\\n\").replace(\"\\\\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 = [\n",
+    "                f\"{tokenizer.decode([idx])}:{value}\"\n",
+    "                for value, idx in zip(top_k.values, top_k.indices)\n",
+    "            ]\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\n"
+   ]
+  }
+ ],
+ "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
+}