{
"cells": [
{
"cell_type": "code",
"execution_count": 3,
"metadata": {},
"outputs": [
{
"ename": "ImportError",
"evalue": "cannot import name 'Benchmark' from 'ferret' (/home/cass/anaconda3/envs/eva/lib/python3.11/site-packages/ferret.py)",
"output_type": "error",
"traceback": [
"\u001b[0;31m---------------------------------------------------------------------------\u001b[0m",
"\u001b[0;31mImportError\u001b[0m Traceback (most recent call last)",
"Cell \u001b[0;32mIn[3], line 2\u001b[0m\n\u001b[1;32m 1\u001b[0m \u001b[39mfrom\u001b[39;00m \u001b[39mtransformers\u001b[39;00m \u001b[39mimport\u001b[39;00m AutoModelForSequenceClassification, AutoTokenizer\n\u001b[0;32m----> 2\u001b[0m \u001b[39mfrom\u001b[39;00m \u001b[39mferret\u001b[39;00m \u001b[39mimport\u001b[39;00m Benchmark\n\u001b[1;32m 4\u001b[0m name \u001b[39m=\u001b[39m \u001b[39m\"\u001b[39m\u001b[39mcardiffnlp/twitter-xlm-roberta-base-sentiment\u001b[39m\u001b[39m\"\u001b[39m\n\u001b[1;32m 5\u001b[0m model \u001b[39m=\u001b[39m AutoModelForSequenceClassification\u001b[39m.\u001b[39mfrom_pretrained(name)\n",
"\u001b[0;31mImportError\u001b[0m: cannot import name 'Benchmark' from 'ferret' (/home/cass/anaconda3/envs/eva/lib/python3.11/site-packages/ferret.py)"
]
}
],
"source": [
"from transformers import AutoModelForSequenceClassification, AutoTokenizer\n",
"from ferret import Benchmark\n",
"\n",
"name = \"cardiffnlp/twitter-xlm-roberta-base-sentiment\"\n",
"model = AutoModelForSequenceClassification.from_pretrained(name)\n",
"tokenizer = AutoTokenizer.from_pretrained(name)\n",
"\n",
"bench = Benchmark(model, tokenizer)\n",
"explanations = bench.explain(\"You look stunning!\", target=1)\n",
"evaluations = bench.evaluate_explanations(explanations, target=1)\n",
"\n",
"bench.show_evaluation_table(evaluations)"
]
},
{
"cell_type": "code",
"execution_count": 1,
"metadata": {},
"outputs": [
{
"name": "stderr",
"output_type": "stream",
"text": [
"/home/cass/anaconda3/envs/eva/lib/python3.11/site-packages/tqdm/auto.py:21: TqdmWarning: IProgress not found. Please update jupyter and ipywidgets. See https://ipywidgets.readthedocs.io/en/stable/user_install.html\n",
" from .autonotebook import tqdm as notebook_tqdm\n",
"2023-07-04 09:52:18.157356: I tensorflow/core/util/port.cc:110] oneDNN custom operations are on. You may see slightly different numerical results due to floating-point round-off errors from different computation orders. To turn them off, set the environment variable `TF_ENABLE_ONEDNN_OPTS=0`.\n",
"2023-07-04 09:52:18.188381: I tensorflow/core/platform/cpu_feature_guard.cc:182] This TensorFlow binary is optimized to use available CPU instructions in performance-critical operations.\n",
"To enable the following instructions: AVX2 AVX512F AVX512_VNNI FMA, in other operations, rebuild TensorFlow with the appropriate compiler flags.\n",
"2023-07-04 09:52:18.794393: W tensorflow/compiler/tf2tensorrt/utils/py_utils.cc:38] TF-TRT Warning: Could not find TensorRT\n"
]
}
],
"source": [
"import inseq"
]
},
{
"cell_type": "code",
"execution_count": 25,
"metadata": {},
"outputs": [
{
"name": "stderr",
"output_type": "stream",
"text": [
"Setting `pad_token_id` to `eos_token_id`:50256 for open-end generation.\n",
"Unused arguments during attribution: {'n_steps': 500, 'internal_batch_size': 50}\n",
"Attributing with saliency...: 100%|██████████| 45/45 [00:00<00:00, 50.01it/s]\n"
]
},
{
"data": {
"text/html": [
"<br/><b>0th instance:</b><br/>\n",
"<html>\n",
"<div id=\"usieyiccwpusnzjovqwp_viz_container\">\n",
" <div id=\"usieyiccwpusnzjovqwp_content\" style=\"padding:15px;border-style:solid;margin:5px;\">\n",
" <div id = \"usieyiccwpusnzjovqwp_saliency_plot_container\" class=\"usieyiccwpusnzjovqwp_viz_container\" style=\"display:block\">\n",
" \n",
"<div id=\"qahftrxckafnafqofpva_saliency_plot\" class=\"qahftrxckafnafqofpva_viz_content\">\n",
" <div style=\"margin:5px;font-family:sans-serif;font-weight:bold;\">\n",
" <span style=\"font-size: 20px;\">Target Saliency Heatmap</span>\n",
" <br>\n",
" x: Generated tokens, y: Attributed tokens\n",
" </div>\n",
" \n",
"<table border=\"1\" cellpadding=\"5\" cellspacing=\"5\"\n",
" style=\"overflow-x:scroll;display:block;\">\n",
" <tr><th></th>\n",
"<th>ĠTokyo</th></tr><tr><th>When</th><th style=\"background:rgba(255.0, 13.0, 87.0, 0.1959595959595959)\">0.036</th></tr><tr><th>Ġmy</th><th style=\"background:rgba(255.0, 13.0, 87.0, 0.10136660724896006)\">0.02</th></tr><tr><th>Ġflight</th><th style=\"background:rgba(255.0, 13.0, 87.0, 0.17231134878193693)\">0.031</th></tr><tr><th>Ġlanded</th><th style=\"background:rgba(255.0, 13.0, 87.0, 0.10136660724896006)\">0.02</th></tr><tr><th>Ġin</th><th style=\"background:rgba(255.0, 13.0, 87.0, 0.15654585066349747)\">0.029</th></tr><tr><th>ĠJapan</th><th style=\"background:rgba(255.0, 13.0, 87.0, 0.44032481679540497)\">0.081</th></tr><tr><th>,</th><th style=\"background:rgba(255.0, 13.0, 87.0, 0.14866310160427795)\">0.028</th></tr><tr><th>ĠI</th><th style=\"background:rgba(255.0, 13.0, 87.0, 0.03830461477520289)\">0.008</th></tr><tr><th>Ġconverted</th><th style=\"background:rgba(255.0, 13.0, 87.0, 0.06983561101208159)\">0.014</th></tr><tr><th>Ġmy</th><th style=\"background:rgba(255.0, 13.0, 87.0, 0.03830461477520289)\">0.008</th></tr><tr><th>Ġcurrency</th><th style=\"background:rgba(255.0, 13.0, 87.0, 0.23537334125569415)\">0.043</th></tr><tr><th>Ġand</th><th style=\"background:rgba(255.0, 13.0, 87.0, 0.06983561101208159)\">0.013</th></tr><tr><th>Ġslowly</th><th style=\"background:rgba(255.0, 13.0, 87.0, 0.06195286195286207)\">0.012</th></tr><tr><th>Ġfell</th><th style=\"background:rgba(255.0, 13.0, 87.0, 0.03830461477520289)\">0.008</th></tr><tr><th>Ġasleep</th><th style=\"background:rgba(255.0, 13.0, 87.0, 0.09348385818974037)\">0.017</th></tr><tr><th>.</th><th style=\"background:rgba(255.0, 13.0, 87.0, 0.08560110913052081)\">0.016</th></tr><tr><th>Ġ(</th><th style=\"background:rgba(255.0, 13.0, 87.0, 0.06983561101208159)\">0.014</th></tr><tr><th>I</th><th style=\"background:rgba(255.0, 13.0, 87.0, 0.030421865715983164)\">0.006</th></tr><tr><th>Ġhad</th><th style=\"background:rgba(255.0, 13.0, 87.0, 0.030421865715983164)\">0.006</th></tr><tr><th>Ġa</th><th style=\"background:rgba(255.0, 13.0, 87.0, 0.04618736383442265)\">0.009</th></tr><tr><th>Ġterrifying</th><th style=\"background:rgba(255.0, 13.0, 87.0, 0.08560110913052081)\">0.016</th></tr><tr><th>Ġdream</th><th style=\"background:rgba(255.0, 13.0, 87.0, 0.06195286195286207)\">0.012</th></tr><tr><th>Ġabout</th><th style=\"background:rgba(255.0, 13.0, 87.0, 0.03830461477520289)\">0.008</th></tr><tr><th>Ġmy</th><th style=\"background:rgba(255.0, 13.0, 87.0, 0.030421865715983164)\">0.007</th></tr><tr><th>Ġgrandmother</th><th style=\"background:rgba(255.0, 13.0, 87.0, 0.12501485442661908)\">0.023</th></tr><tr><th>,</th><th style=\"background:rgba(255.0, 13.0, 87.0, 0.04618736383442265)\">0.01</th></tr><tr><th>Ġbut</th><th style=\"background:rgba(255.0, 13.0, 87.0, 0.030421865715983164)\">0.006</th></tr><tr><th>Ġthat</th><th style=\"background:rgba(255.0, 13.0, 87.0, 0.03830461477520289)\">0.008</th></tr><tr><th>âĢ</th><th style=\"background:rgba(255.0, 13.0, 87.0, 0.10924935630817992)\">0.02</th></tr><tr><th>Ļ</th><th style=\"background:rgba(255.0, 13.0, 87.0, 0.04618736383442265)\">0.009</th></tr><tr><th>s</th><th style=\"background:rgba(255.0, 13.0, 87.0, 0.030421865715983164)\">0.006</th></tr><tr><th>Ġa</th><th style=\"background:rgba(255.0, 13.0, 87.0, 0.05407011289364243)\">0.01</th></tr><tr><th>Ġstory</th><th style=\"background:rgba(255.0, 13.0, 87.0, 0.10136660724896006)\">0.019</th></tr><tr><th>Ġfor</th><th style=\"background:rgba(255.0, 13.0, 87.0, 0.03830461477520289)\">0.008</th></tr><tr><th>Ġanother</th><th style=\"background:rgba(255.0, 13.0, 87.0, 0.030421865715983164)\">0.006</th></tr><tr><th>Ġtime</th><th style=\"background:rgba(255.0, 13.0, 87.0, 0.03830461477520289)\">0.008</th></tr><tr><th>).</th><th style=\"background:rgba(255.0, 13.0, 87.0, 0.10924935630817992)\">0.02</th></tr><tr><th>ĠI</th><th style=\"background:rgba(255.0, 13.0, 87.0, 0.07771836007130124)\">0.015</th></tr><tr><th>Ġwas</th><th style=\"background:rgba(255.0, 13.0, 87.0, 0.13289760348583876)\">0.025</th></tr><tr><th>Ġstaying</th><th style=\"background:rgba(255.0, 13.0, 87.0, 0.18019409784115661)\">0.033</th></tr><tr><th>Ġin</th><th style=\"background:rgba(255.0, 13.0, 87.0, 0.14866310160427795)\">0.028</th></tr><tr><th>Ġthe</th><th style=\"background:rgba(255.0, 13.0, 87.0, 0.2117250940780353)\">0.038</th></tr><tr><th>Ġcapital</th><th style=\"background:rgba(255.0, 13.0, 87.0, 1.0)\">0.182</th></tr><tr><th>Ġof</th><th style=\"background:rgba(255.0, 13.0, 87.0, 0.3536145771439889)\">0.065</th></tr><tr><th>ĠTokyo</th><th style=\"background:rgba(0.0, 0.0, 0.0, 0.0)\"></th></tr></table>\n",
"</div>\n",
"\n",
" </div>\n",
" </div>\n",
"</div>\n",
"</html>\n"
],
"text/plain": [
"<IPython.core.display.HTML object>"
]
},
"metadata": {},
"output_type": "display_data"
}
],
"source": [
"model = inseq.load_model(\"gpt2\", \"saliency\") # saliency integrated_gradients\n",
"\n",
"country = 'Japan'\n",
"text = f\"When my flight landed in {country}, I converted my currency and slowly fell asleep. (I had a terrifying dream about my grandmother, but that’s a story for another time). I was staying in the capital of\"\n",
"model.attribute(text, generation_args={\"max_new_tokens\": 1}, n_steps=500, internal_batch_size=50).show()"
]
},
{
"cell_type": "code",
"execution_count": 19,
"metadata": {},
"outputs": [
{
"name": "stderr",
"output_type": "stream",
"text": [
"Setting `pad_token_id` to `eos_token_id`:50256 for open-end generation.\n",
"Unused arguments during attribution: {'n_steps': 500, 'internal_batch_size': 50}\n",
"Attributing with saliency...: 100%|██████████| 24/24 [00:00<00:00, 61.98it/s]\n"
]
},
{
"data": {
"text/html": [
"<br/><b>0th instance:</b><br/>\n",
"<html>\n",
"<div id=\"usieyiccwpusnzjovqwp_viz_container\">\n",
" <div id=\"usieyiccwpusnzjovqwp_content\" style=\"padding:15px;border-style:solid;margin:5px;\">\n",
" <div id = \"usieyiccwpusnzjovqwp_saliency_plot_container\" class=\"usieyiccwpusnzjovqwp_viz_container\" style=\"display:block\">\n",
" \n",
"<div id=\"qahftrxckafnafqofpva_saliency_plot\" class=\"qahftrxckafnafqofpva_viz_content\">\n",
" <div style=\"margin:5px;font-family:sans-serif;font-weight:bold;\">\n",
" <span style=\"font-size: 20px;\">Target Saliency Heatmap</span>\n",
" <br>\n",
" x: Generated tokens, y: Attributed tokens\n",
" </div>\n",
" \n",
"<table border=\"1\" cellpadding=\"5\" cellspacing=\"5\"\n",
" style=\"overflow-x:scroll;display:block;\">\n",
" <tr><th></th>\n",
"<th>ĠBeijing</th></tr><tr><th>When</th><th style=\"background:rgba(255.0, 13.0, 87.0, 0.306318082788671)\">0.057</th></tr><tr><th>Ġmy</th><th style=\"background:rgba(255.0, 13.0, 87.0, 0.16442859972271742)\">0.031</th></tr><tr><th>Ġflight</th><th style=\"background:rgba(255.0, 13.0, 87.0, 0.2747870865517925)\">0.051</th></tr><tr><th>Ġlanded</th><th style=\"background:rgba(255.0, 13.0, 87.0, 0.15654585066349747)\">0.03</th></tr><tr><th>Ġin</th><th style=\"background:rgba(255.0, 13.0, 87.0, 0.2511388393741335)\">0.047</th></tr><tr><th>ĠChina</th><th style=\"background:rgba(255.0, 13.0, 87.0, 0.5979797979797981)\">0.111</th></tr><tr><th>,</th><th style=\"background:rgba(255.0, 13.0, 87.0, 0.22749059219647458)\">0.043</th></tr><tr><th>ĠI</th><th style=\"background:rgba(255.0, 13.0, 87.0, 0.06983561101208159)\">0.014</th></tr><tr><th>Ġconverted</th><th style=\"background:rgba(255.0, 13.0, 87.0, 0.10136660724896006)\">0.02</th></tr><tr><th>Ġmy</th><th style=\"background:rgba(255.0, 13.0, 87.0, 0.05407011289364243)\">0.011</th></tr><tr><th>Ġcurrency</th><th style=\"background:rgba(255.0, 13.0, 87.0, 0.2747870865517925)\">0.052</th></tr><tr><th>Ġand</th><th style=\"background:rgba(255.0, 13.0, 87.0, 0.12501485442661908)\">0.023</th></tr><tr><th>Ġslowly</th><th style=\"background:rgba(255.0, 13.0, 87.0, 0.10136660724896006)\">0.02</th></tr><tr><th>Ġfell</th><th style=\"background:rgba(255.0, 13.0, 87.0, 0.08560110913052081)\">0.016</th></tr><tr><th>Ġasleep</th><th style=\"background:rgba(255.0, 13.0, 87.0, 0.14866310160427795)\">0.029</th></tr><tr><th>.</th><th style=\"background:rgba(255.0, 13.0, 87.0, 0.18019409784115661)\">0.033</th></tr><tr><th>ĠI</th><th style=\"background:rgba(255.0, 13.0, 87.0, 0.07771836007130124)\">0.015</th></tr><tr><th>Ġwas</th><th style=\"background:rgba(255.0, 13.0, 87.0, 0.12501485442661908)\">0.024</th></tr><tr><th>Ġstaying</th><th style=\"background:rgba(255.0, 13.0, 87.0, 0.22749059219647458)\">0.043</th></tr><tr><th>Ġin</th><th style=\"background:rgba(255.0, 13.0, 87.0, 0.24325609031491383)\">0.046</th></tr><tr><th>Ġthe</th><th style=\"background:rgba(255.0, 13.0, 87.0, 0.219607843137255)\">0.041</th></tr><tr><th>Ġcapital</th><th style=\"background:rgba(255.0, 13.0, 87.0, 1.0)\">0.185</th></tr><tr><th>Ġof</th><th style=\"background:rgba(255.0, 13.0, 87.0, 0.32208358090711037)\">0.06</th></tr><tr><th>ĠBeijing</th><th style=\"background:rgba(0.0, 0.0, 0.0, 0.0)\"></th></tr></table>\n",
"</div>\n",
"\n",
" </div>\n",
" </div>\n",
"</div>\n",
"</html>\n"
],
"text/plain": [
"<IPython.core.display.HTML object>"
]
},
"metadata": {},
"output_type": "display_data"
}
],
"source": [
"country = 'Japan'\n",
"model.attribute(text, generation_args={\"max_new_tokens\": 1}, n_steps=500, internal_batch_size=50).show()"
]
}
],
"metadata": {
"kernelspec": {
"display_name": "eva",
"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": "3.11.0"
},
"orig_nbformat": 4
},
"nbformat": 4,
"nbformat_minor": 2
}