NumeroIssue,IdIssue,TituloIssue,DescricaoIssue,CriacaoIssue,RepositorioIssue,LinkIssue 354,2540380013,Fix GPT-2 paper link,Fixes the link to the GPT-2 paper.,2024-09-21T16:08:05Z,gpt-2,https://github.com/openai/gpt-2/pull/354 343,2404895719,Does the evaluation code get released?,"Hello, thank you for your contribution. I am trying to reproduce the results of the evaluation experiment reported in Table 3 of the paper, but found it a bit difficult for a novice. I would like to know if there is any available tutorial and code for gpt2 evaluation on datasets like LAMBADA and CBT? Thank you very much for your help!",2024-07-12T06:59:06Z,gpt-2,https://github.com/openai/gpt-2/issues/343 342,2349809452,سلام,"import subprocess from telegram import Update from telegram.ext import Updater, CommandHandler, CallbackContext import logging # تنظیمات لاگینگ logging.basicConfig(format='%(asctime)s - %(name)s - %(levelname)s - %(message)s', level=logging.INFO) logger = logging.getLogger(__name__) # توکن تلگرام خود را وارد کنید TELEGRAM_TOKEN = '7124126923:AAHVVtgGQlJA32M_shpw-9WhehfMLndNcHY' # توابع برای دستورات مختلف def start(update: Update, context: CallbackContext) -> None: update.message.reply_text(""خوش آمدید!"") def help_command(update: Update, context: CallbackContext) -> None: update.message.reply_text( - - - - ساخت ربات - ساخت ربات جدید به صورت - بازیابی - راه‌اندازی - پاک کردن - نصب پیش ) def update_bot(): subprocess.run([""git"", ""fetch"", ""--all""]) subprocess.run([""git"", ""reset"", ""--hard"", subprocess.run([""git"", ""pull"", ""origin"", ""persian""]) subprocess.run([""chmod"", ""+x"", ""bot""]) logger.info(""بروزرسانی اطلاعات با موفقیت انجام شد."") def update_command(update: Update, context: CallbackContext) -> None: update_bot() update.message.reply_text(""بروزرسانی اطلاعات با موفقیت انجام شد."") def create_command(update: Update, context: CallbackContext) -> None: # کد ساخت ربات جدید update.message.reply_text(""ربات جدید با موفقیت ساخته شد."") def createmanual_command(update: Update, context: CallbackContext) -> None: # کد ساخت ربات به صورت دستی update.message.reply_text(""لطفاً شماره دلخواه خود را وارد کنید."") def fix_command(update: Update, context: CallbackContext) -> None: subprocess.run([""git"", ""reset"", ""--hard"", ""FETCH_HEAD""]) update.message.reply_text(""بازیابی اطلاعات به آخرین بروزرسانی انجام شد."") def autolaunch_command(update: Update, context: CallbackContext) -> None: update.message.reply_text(""راه‌اندازی خودکار ربات‌ها هر 20 دقیقه فعال شد."") def clear_command(update: Update, context: CallbackContext) -> None: update.message.reply_text(""لطفاً شماره شناسه رباتی که می‌خواهید پاک کنید را وارد کنید."") def install_command(update: Update, context: CallbackContext) -> None: update.message.reply_text(""نصب پیش نیازهای ربات در حال انجام است."") # هندلر خطا def error_handler(update: Update, context: CallbackContext) -> None: logger.error(msg=""Exception while handling an update:"", exc_info=context.error) if update and update.message: update.message.reply_text('خطایی رخ داده است. لطفاً دوباره تلاش کنید.') def main() -> None: updater = Updater(TELEGRAM_TOKEN) dispatcher = updater.dispatcher dispatcher.add_handler(CommandHandler(""start"", start)) dispatcher.add_handler(CommandHandler(""help"", help_command)) dispatcher.add_handler(CommandHandler(""update"", update_command)) dispatcher.add_handler(CommandHandler(""create"", create_command)) dispatcher.add_handler(CommandHandler(""createmanual"", createmanual_command)) dispatcher.add_handler(CommandHandler(""fix"", fix_command)) dispatcher.add_handler(CommandHandler(""autolaunch"", autolaunch_command)) dispatcher.add_handler(CommandHandler(""clear"", clear_command)) dispatcher.add_handler(CommandHandler(""install"", install_command)) dispatcher.add_error_handler(error_handler) updater.start_polling() updater.idle() if __name__ == '__main__': main()",2024-06-12T22:58:25Z,gpt-2,https://github.com/openai/gpt-2/issues/342 334,2217223861,GPT-2 implementation problem,"""Hi, I am reading the GPT-2 paper and encountering a problem with the following phrase related to implementation: 'A modified initialization method is used to account for the accumulation on the residual path with model depth. We scale the weights of residual layers at initialization by a factor of where N is the number of residual layers.' My problem is that we normalize after accumulation (addition then normalization). So, why do we need to scale weights? Aren't we doing this to reduce the impact of accumulation?""",2024-03-31T22:18:39Z,gpt-2,https://github.com/openai/gpt-2/issues/334 331,2102993273,Update original GPT-2 blog post to working link, ,2024-01-26T22:28:35Z,gpt-2,https://github.com/openai/gpt-2/pull/331 329,2010053150,Problems trying to build gpt-2,"Hi, I am trying to build gpt-2 on Ubuntu 22.04.2. I git cloned this repository. In DEVELOPERS.md, it says that you have to pip install tensorflow==1.12.0. When I try to install tensorflow==1.12.0, it makes an error: ERROR: Could not find a version that satisfies the requirement tensorflow==1.12.0 (from versions: 2.8.0rc0, 2.8.0rc1, 2.8.0, 2.8.1, 2.8.2, 2.8.3, 2.8.4, 2.9.0rc0, 2.9.0rc1, 2.9.0rc2, 2.9.0, 2.9.1, 2.9.2, 2.9.3, 2.10.0rc0, 2.10.0rc1, 2.10.0rc2, 2.10.0rc3, 2.10.0, 2.10.1, 2.11.0rc0, 2.11.0rc1, 2.11.0rc2, 2.11.0, 2.11.1, 2.12.0rc0, 2.12.0rc1, 2.12.0, 2.12.1, 2.13.0rc0, 2.13.0rc1, 2.13.0rc2, 2.13.0, 2.13.1, 2.14.0rc0, 2.14.0rc1, 2.14.0, 2.14.1, 2.15.0rc0, 2.15.0rc1, 2.15.0) ERROR: No matching distribution found for tensorflow==1.12.0 I tried omitting the ==1.12.0 and it successfully installed a newer version of tensorflow, but if I use the new tensorflow, when I try to run the model, it makes an error: --top_k 40 2023-11-24 17 15.771054: I Could not find cuda drivers on your machine, GPU will not be used. 2023-11-24 17 16.687450: E Unable to register cuDNN factory: Attempting to register factory for plugin cuDNN when one has already been registered 2023-11-24 17 16.687553: E Unable to register cuFFT factory: Attempting to register factory for plugin cuFFT when one has already been registered 2023-11-24 17 16.885832: E Unable to register cuBLAS factory: Attempting to register factory for plugin cuBLAS when one has already been registered 2023-11-24 17 17.264367: I Could not find cuda drivers on your machine, GPU will not be used. 2023-11-24 17 17.267425: I This TensorFlow binary is optimized to use available CPU instructions in performance-critical operations. To enable the following instructions: AVX2 AVX512F FMA, in other operations, rebuild TensorFlow with the appropriate compiler flags. 2023-11-24 17 19.771254: W TF-TRT Warning: Could not find TensorRT Traceback (most recent call last): File line 9, in import model, sample, encoder File line 3, in from tensorflow.contrib.training import HParams ModuleNotFoundError: No module named 'tensorflow.contrib' In this stackoverflow article: someone says that tensorflow.contrib no longer exists in tensorflow 2.0.x and its modules were moved. Perhaps there is a way to get the modules from wherever they were moved? Any help on building gpt-2 would be appreciated!",2023-11-24T17:54:49Z,gpt-2,https://github.com/openai/gpt-2/issues/329 317,1698930108,Ppok, ,2023-05-07T07:06:56Z,gpt-2,https://github.com/openai/gpt-2/issues/317 316,1697124950,Looking to purchase GPT-4 API Keys,"Is anyone still on the waiting list for GPT-4? Don't worry, [https this platform offers a reliable GPT-4 API key rental service, which many people are already using. This is really great!",2023-05-05T07:10:16Z,gpt-2,https://github.com/openai/gpt-2/issues/316 314,1601146627,"Installation, revisited","Summary: Run e.g. this one) for a properly running version. Details: Almost nothing has worked out of the box by now, even [the dockers way](https ` load metadata for ------ failed to solve with frontend dockerfile.v0: failed to create LLB definition: rpc error: code = Unknown desc = failed to parse unexpected end of JSON input ` or and such, partially resolvable via a docker builders purge plus: rm rm but then: The tensorflow ""dll hell"" got worse since then, obviously. Only **this below has worked** for me so far, so pro memoriam: etc. to intialize it and then: and then get out: Interestingly, it is OpenAI ChatGPT plus Bing Chat , the daughers of this one, who has helped me troubleshoot these. The Github Spaces box where mine runs, another fyi: > (master) $ neofetch > > OS: Ubuntu 20.04.5 LTS x86_64 > Kernel: 5.4.0-1103-azure > CPU: Intel Xeon Platinum 8272CL (4) 2.5GHz > Memory: 2019MiB 7957MiB > Shell: 5.0.17 > Python: 3.10.4 > ------ > Uptime: 1 day, 1 hour, 41 mins > Packages: 844 (dpkg) > Terminal: > ------ > CPU Usage: 2% > Disk 17G 31G (59%) > ------ > Host: Virtual Machine 7.0 > Public IP: 20.105.169.104 Ver. 1.0.1",2023-02-27T13:03:33Z,gpt-2,https://github.com/openai/gpt-2/issues/314 313,1575356413,GPT-2 pretrain loss.,"Hello, Thanks for your great work. I want to know how to calculate the loss given the raw text. For example: I have a sample in training data: "" I want to go to school"". When I input the string into the GPT-2 model, every output logits has a loss value. So the total loss is the sum of all output logits loss? ",2023-02-08T02:32:30Z,gpt-2,https://github.com/openai/gpt-2/issues/313 312,1575326475,How does ChatGPT support languages except English?,"How does OpenAI achieve multi-language support? How did OpenAI do so many human-labeling works? I guess OpenAI is using all the world users' feedback data.",2023-02-08T01:56:07Z,gpt-2,https://github.com/openai/gpt-2/issues/312 310,1563347280,Chats won't appear on desktop ChatGPT website,"On ChatGPT January 30 version, chats won't appear on desktop website. I'm using desktop Google Chrome version 109.0.5414.199 and Fedora Linux 37 system ![Captura desde 2023-01-30 20-06-43](https 1. Open ChatGPT website ( 2. Type any sentence on the box 3. Press Enter Expected result: Chats should appear on website as usual. Actual result: Chats won't appear on website. ",2023-01-30T23:08:11Z,gpt-2,https://github.com/openai/gpt-2/issues/310 309,1465854854,How to use gpt-2 for question answers task?, ,2022-11-28T05:52:45Z,gpt-2,https://github.com/openai/gpt-2/issues/309 308,1449790873,docker image ,"the command : return this : ",2022-11-15T13:31:54Z,gpt-2,https://github.com/openai/gpt-2/issues/308 306,1318810837,interactive_conditional_samples.py crashes if there is more than one context token,"I can run the generate_unconditional_samples.py script on my GPU without issue, however, when I run the interactive_conditional_samples.py script, it crashes if there is more than one context token. The interactive_conditional_samples.py script works fine as long as the model prompt only produces one context token, for instance using the prompt ""please"" produces the list of tokens [29688] and correctly generates text. However, it crashes if the model prompt produces two or more context tokens, for instance using the prompt ""pig"" produces the list of tokens [79, 328] and crashes immediately. When it crashes I'm getting the error: And a little further down I see: If anyone has any insight on what might be going wrong, and how I can fix it, I'd really appreciate the help.",2022-07-26T22:25:38Z,gpt-2,https://github.com/openai/gpt-2/issues/306 305,1301618734,Dose the pre-training data also use this prompt structure related to downstream tasks？,"I read the gpt2 paper, but not sure whether the pre-training data from WebText will add format information. For example, we konw data format will be in the translation task. So during pre-training time, will we add similar promt to the training data? Thanks!",2022-07-12T06:50:48Z,gpt-2,https://github.com/openai/gpt-2/issues/305 304,1296846995,Local path resolution,"There was a regression in the code base where absolute paths are stored in the training metadata and tokenizer index files. This meant that these files couldn't be sent to others and used on their system. This PR introduced the wrapper class which allows access to file paths relative to the project root and as an absolute path. The code was updated to use this class, and to never save an absolute path to one of the aforementioned JSON files. Additionally some logic was added to the class to detect the absolute paths from other systems and to a best effort to resolve them on the new person's machine.",2022-07-07T05:38:13Z,gpt-2,https://github.com/openai/gpt-2/pull/304 302,1203270274,Can't download the 1558M model,"Using the by running I was able to download all 3 models, but running returns me just and not the model itself.",2022-04-13T13:01:29Z,gpt-2,https://github.com/openai/gpt-2/issues/302 300,1078870813,"Fix error: 'tensorflow' has no attribute 'sort'""","add tensorflow 1.13.1 to force update into docker build and fix de ex…ecution issue ""AttributeError: module 'tensorflow' has no attribute 'sort'""",2021-12-13T18:55:50Z,gpt-2,https://github.com/openai/gpt-2/pull/300 299,1048716203,TypeError: add_code_sample_docstrings() got an unexpected keyword argument 'tokenizer_class', ,2021-11-09T15:06:55Z,gpt-2,https://github.com/openai/gpt-2/issues/299 298,973320056,to EXTRA LARGE I only could download with this parameter: **1558M**,"to EXTRA LARGE I only could download with this parameter: **1558M** > _$ python download_model.py 1558M_ EXTRA LARGE 1542M according to 1.5BM according to __Originally posted by in ",2021-08-18T06:25:24Z,gpt-2,https://github.com/openai/gpt-2/issues/298 295,914725610,ModuleNotFoundError: No module named 'fire',"Hi there! Please, help to find out why I get this issue after module's import and installation. Behind I've added some proofs (stderr from my terminal): (venv) sasha pip3 install fire Processing Requirement already satisfied: six in (from fire) (1.14.0) Requirement already satisfied: termcolor in (from fire) (1.1.0) Installing collected packages: fire Successfully installed fire-0.4.0 (venv) sasha python Traceback (most recent call last): File line 4, in import fire ModuleNotFoundError: No module named 'fire' Part of my script with import: import logging import fnmatch import subprocess import fire import os def setup_log(name): logger = logging.getLogger(name) # > set up a new name for a new logger logger.setLevel(logging.INFO) Also I've tried use pip for installation, but there wasn't any impact on result... ",2021-06-08T09:54:54Z,gpt-2,https://github.com/openai/gpt-2/issues/295 291,870539088,ModuleNotFoundError with TensorFlow,"I have tried with multiple versions of tensorflow and python (1.12.0, 1.14.0, 1.13.1, 2.4 for tf, 3.9, 3.6.0, and 3.6.5 for python) and every time i try to run any sample the same error message pops up. (py365) --top_k 40 Traceback (most recent call last): File line 7, in import tensorflow as tf File line 41, in from tensorflow.python.tools import module_util as _module_util ModuleNotFoundError: No module named 'tensorflow.python' is there something i've done wrong? is there any way to fix this? any help would be greatly appreciated! thank you loads! and if any additional information is needed please let me know! p.s. i apologize, i dont know how to format on github, so that's why the text isnt in a codeblock or anything. sorry again.",2021-04-29T01:49:49Z,gpt-2,https://github.com/openai/gpt-2/issues/291 290,857068165,in domains.txt,"in domains.txt there is a domain named ""ashemaletube""",2021-04-13T15:19:42Z,gpt-2,https://github.com/openai/gpt-2/issues/290 288,824981046,Installation,I am wondering how to install it on mac. So far it only shows how to install on windows.,2021-03-08T21:42:32Z,gpt-2,https://github.com/openai/gpt-2/issues/288 287,819956670,text export," ",2021-03-02T12:07:54Z,gpt-2,https://github.com/openai/gpt-2/issues/287 286,818844119,Beginner questions ,"ModuleNotFoundError: No module named 'tensorflow.contrib' May I know How can I solve this issue. Thank you ",2021-03-01T13:01:17Z,gpt-2,https://github.com/openai/gpt-2/issues/286 285,818042942,Checkpoint not generating,"I ran the program of GPT-2 on a txt training data which has 3 stories. I used the 117M parameters model, and it runs, it trains the model, but once it stops it creates checkpoint folder inside it is run1 folder, but none of these files are generated: - checkpoint - model-xxx.data-00000-of-00001 - model-xxx.index - model-xxx.meta > Use standard file APIs to check for files with this prefix. > Loading dataset... > 100%|████████████████████████████████████████████████████████████████████████████████████| [00 00, > > dataset has 12863 tokens > > Training... > > [1 | 22.35] loss=3.69 avg=3.69 > [2 | 40.40] loss=3.48 avg=3.58 > [3 | 72.00] loss=3.34 avg=3.50 > [4 | 91.34] loss=3.45 avg=3.49 > [5 | 111.14] loss=3.32 avg=3.45 > [6 | 130.68] loss=3.63 avg=3.48 > [7 | 146.00] loss=3.35 avg=3.46 > [8 | 164.12] loss=3.33 avg=3.45 > [9 | 187.81] loss=3.44 avg=3.45 > [10 | 212.46] loss=3.41 avg=3.44 > [11 | 238.91] loss=3.35 avg=3.43 > [12 | 265.70] loss=3.07 avg=3.40 > [13 | 286.85] loss=3.36 avg=3.40 > [14 | 309.50] loss=3.32 avg=3.39 > [15 | 327.70] loss=3.26 avg=3.38 > [16 | 344.01] loss=3.22 avg=3.37 > [17 | 358.19] loss=3.41 avg=3.37 > [18 | 371.93] loss=2.95 avg=3.35 > [19 | 386.32] loss=3.19 avg=3.34 > [20 | 400.90] loss=3.51 avg=3.35 > [21 | 415.34] loss=3.06 avg=3.33 > [22 | 430.17] loss=3.47 avg=3.34 > [23 | 444.54] loss=3.06 avg=3.33 > > forrtl: error (200): program aborting due to control-C event > > Image PC Routine Line Source > libifcoremd.dll 00007FFD7D033B58 Unknown Unknown Unknown > KERNELBASE.dll 00007FFDC9D6B443 Unknown Unknown Unknown > KERNEL32.DLL 00007FFDCC487034 Unknown Unknown Unknown > ntdll.dll 00007FFDCC5BD241 Unknown Unknown Unknown > What should I do?",2021-02-27T22:18:21Z,gpt-2,https://github.com/openai/gpt-2/issues/285 284,815271828,GPT-2 on Wordpress,As i've wondering is it possible to make a GPT-2 work inside the wordpress?,2021-02-24T09:05:10Z,gpt-2,https://github.com/openai/gpt-2/issues/284 283,810050725,Access Denied error when downloading model 345M,"I get the following error when downloading model 345M using download_model.py: Manually opening in the browser produces the same error. All links in this comment also produce the same error. I am using download_model.py in Google Colab. I tried to download the link in a different browser with empty cash and I get the same mistake. ",2021-02-17T10:22:40Z,gpt-2,https://github.com/openai/gpt-2/issues/283 282,807664698,Error with python download_model 117M," ... ... Access
Anonymous caller does not have storage.objects.get access to the Google Cloud Storage Getting this error, even when I visit the link normally from the browser.",2021-02-13T03:26:35Z,gpt-2,https://github.com/openai/gpt-2/issues/282 281,807628900,Error with batch_size > 1,"Hello. Everything works fine until I increase nsamples and batch_size to >1. Tried in docker and in conda environment, with different versions of TF(from 1.12 to 1.15), on Tesla V100 and GF1060 - nothing helps. Will be appreciate for any help. Thank you.",2021-02-13T00:20:07Z,gpt-2,https://github.com/openai/gpt-2/issues/281 280,792723211,Reformat file and added documentation,Added documentation for different model sizes,2021-01-24T05:14:38Z,gpt-2,https://github.com/openai/gpt-2/pull/280 279,792281471,Update interactive_conditional_samples.py,code has some depreciated functions used,2021-01-22T20:24:00Z,gpt-2,https://github.com/openai/gpt-2/pull/279 278,773058157,Custom pre-trained model...,"Hello! I have custom txt dataset, how can I create my pre-training model? Do I understand correctly that the concept ""fine-tuning"" what is the pretranded model(example 355M) + my training data?",2020-12-22T16:04:31Z,gpt-2,https://github.com/openai/gpt-2/issues/278 277,771622754,Is there a way to use gpt-2 with the newest tensorflow (2.4)?,"I tried and it didn't work. I am having trouble installing tensorflow 1.12.0. I am trying Which returns: `ERROR: Could not find a version that satisfies the requirement tensorflow==1.12.0 (from versions: 2.2.0rc1, 2.2.0rc2, 2.2.0rc3, 2.2.0rc4, 2.2.0, 2.2.1, 2.3.0rc0, 2.3.0rc1, 2.3.0rc2, 2.3.0, 2.3.1, 2.4.0rc0, 2.4.0rc1, 2.4.0rc2, 2.4.0rc3, 2.4.0rc4) ERROR: No matching distribution found for tensorflow==1.12.0` ",2020-12-20T16:14:47Z,gpt-2,https://github.com/openai/gpt-2/issues/277 273,712049303,How can i use GPT to correct the alignment and spellings in a sentence?,"I have a dataset with 20lakh food item names. And another clean dataset with all the correct vocabularies. I want to use GPT to build a item name corrector. For eg: IN: ""Cheeesseee Pijjja"" OUT: ""Cheese Pizza"" I am aware that GPT language model is capable to do this but I'm not sure how it can be implemented. Any help would really be appreciated",2020-09-30T15:28:10Z,gpt-2,https://github.com/openai/gpt-2/issues/273 268,690735277,got an error while trying to download,"did ""python3 download_model.py 1558M Traceback (most recent call last): File ""download_model.py"", line 4, in from tqdm import tqdm ModuleNotFoundError: No module named 'tqdm' please help ",2020-09-02T05:45:47Z,gpt-2,https://github.com/openai/gpt-2/issues/268 267,689035981,tf_upgrade_v2 fails.,"Hi, I've tried to upgrade the code to tf2 using tf_upgrade_v2 tool and got the following error: `Traceback (most recent call last): File line 8, in sys.exit(main()) File line 152, in main args.input_tree, output_tree, args.copy_other_files) File line 1050, in process_tree _, l_report, l_errors = self.process_file(input_path, output_path) File line 900, in process_file temp_file) File line 958, in process_opened_file lines = in_file.readlines() File line 26, in decode return codecs.ascii_decode(input, self.errors)[0] UnicodeDecodeError: 'ascii' codec can't decode byte 0xc2 in position 786: ordinal not in range(128)` The reason for this error is the following line in encoder.py: I think that I can bypass this issue by deleting the line, converting the code, and then inserting the line back. Do you think it's a valid solution, or maybe it can break something I didn't think about? Thanks. ",2020-08-31T08:47:57Z,gpt-2,https://github.com/openai/gpt-2/issues/267 266,687245654,KeyError: 'para_index' coming when using Question-Generation(interact.py),"I was curious to try using the question generation component. Based on the overall README, I might expect to be able to run interact.py without any arguments, but this doesn't work: That is ok, I downloaded a pretrained model myself from Google drive and tried interact.py again (python3 --model_checkpoint When I did so, I first hit the below issue: Reading the overall README, I inferred that I needed to download instances_dev.pkl, and I found something closely named instances_dev.pickle and instances_coref_dev.pickle here. I assumed I'd need the coref labeled one, as the pretrained folder is named ""gpt2_corefs_question_generation"". But when I tried to use this, I hit the next issue: Would you mind clarifying the steps to test out just this component? It would greatly speed up my efforts.",2020-08-27T13:31:36Z,gpt-2,https://github.com/openai/gpt-2/issues/266 265,686078408,Running GPT-2 with Tensorflow2.2+ issues,"Hi! I have been trying to install GPT-2 locally through several methods, but due to Tensorflow 2.2 (and greater) I have encounter many issues. First with in the first lines of . I solved this issue thanks to this: After that several errors appeared related to Tensorflow not having attributes like , , etc, which I solved using at the beginning of the file, but that developed in other complex errors when the console encounters these lines of code in models.py: The console outputs And at this point I have no idea how to fix this issue. So I tried this: But it tries to install tensorflow 2.0.1, which is not availabe. I have also tried to install earlier versions of tensorflow, but apart to the fact that the module is deprecated in TF 2.0, my console output an error saying that it was not compatible with it. Is there a different way to run GPT-2 now that only Tensorflow 2.2(and greater) is available? I would really love to try GPT-2 in many projects, and it would also be a really good practice until GPT-3 API is released. I will really appreciate any help. Thank you",2020-08-26T07:21:53Z,gpt-2,https://github.com/openai/gpt-2/issues/265 264,684915705,BPE tokenizer has problem in training the low source language like persian,"hi everybody, I'm trying to start train gpt2 in a large amount of Persian data for the special tasks.. but now I got a problem with this tokenizer... after training one data, the .json and .txt frequency information files include some unknown characters:((((( for example: ""ĠبادÙĩا"" something like this... it's good to mention that BPE tokenizer has no problem in English texts... and it makes me confused because this had trained with the Persian dataset but couldn't encode simple Persian sentence... it is a related piece of code... make me happy by your guidance:((( ",2020-08-24T19:37:23Z,gpt-2,https://github.com/openai/gpt-2/issues/264 262,684220249,upgrade to TensorFlow 2,"This PR upgrades the script to work with TensorFlow 2. Main goal was to be able to run it on Google colab, which has the new version. - Used TensorFlow utility to generate compatible code: - Replaced hyperparameters model to simply use a dictionary",2020-08-23T17:09:25Z,gpt-2,https://github.com/openai/gpt-2/pull/262 261,680787211,The specified module was not found,"### This exception occurred after I change TensorFlow version into 1.13.1 Traceback (most recent call last): File line 58, in from tensorflow.python.pywrap_tensorflow_internal import * File line 28, in _pywrap_tensorflow_internal = swig_import_helper() File line 24, in swig_import_helper _mod = imp.load_module('_pywrap_tensorflow_internal', fp, pathname, description) File line 242, in load_module return load_dynamic(name, filename, file) File line 342, in load_dynamic return _load(spec) ImportError: DLL load failed while importing _pywrap_tensorflow_internal: 找不到指定的模块。 During handling of the above exception, another exception occurred: Traceback (most recent call last): File line 7, in import tensorflow as tf File line 24, in from tensorflow.python import pywrap_tensorflow # pylint: disable=unused-import File line 49, in from tensorflow.python import pywrap_tensorflow File line 74, in raise ImportError(msg) ImportError: Traceback (most recent call last): File line 58, in from tensorflow.python.pywrap_tensorflow_internal import * File line 28, in _pywrap_tensorflow_internal = swig_import_helper() File line 24, in swig_import_helper _mod = imp.load_module('_pywrap_tensorflow_internal', fp, pathname, description) File line 242, in load_module return load_dynamic(name, filename, file) File line 342, in load_dynamic return _load(spec) ImportError: DLL load failed while importing _pywrap_tensorflow_internal: 找不到指定的模块。 Failed to load the native TensorFlow runtime. I can only find ""_pywrap_tensorflow_internal.pyd"" this path.",2020-08-18T08:04:55Z,gpt-2,https://github.com/openai/gpt-2/issues/261 259,678767073,download_model.py never gets requests,"Hi, I'm having an issue trying to use download_model.py to get any of the models. I've tried some debugging and it looks like the requests.get commands are never returning. When I manually navigate to the address in browser, it looks like the page contains no Have the locations of the models changed? This occurs for all 4 sets of model data.",2020-08-13T22:19:28Z,gpt-2,https://github.com/openai/gpt-2/issues/259 252,654498100,test,,2020-07-10T04:19:27Z,gpt-2,https://github.com/openai/gpt-2/pull/252 251,654495022,file added,,2020-07-10T04:09:26Z,gpt-2,https://github.com/openai/gpt-2/pull/251 247,633568601,Build error with Dockerfile.gpu,"I get the below error when I tried this on my Jetson Xavier NX. Anyone had a similar issue? My device has ARM v8 processor, OS Type 64-bit sudo docker build --tag gpt-2 -f Dockerfile.gpu .** Sending build context to Docker daemon 4.712MB Step : FROM ---> 413b9533f92a Step : LABEL com.nvidia.volumes.needed=""nvidia_driver"" ---> Using cache ---> 05ba5810d177 Step : LABEL com.nvidia.cuda.version=""${CUDA_VERSION}"" ---> Using cache ---> ebbf24a709ba Step : ENV NVIDIA_VISIBLE_DEVICES=all NVIDIA_DRIVER_CAPABILITIES=compute,utility NVIDIA_REQUIRE_CUDA=""cuda>=8.0"" LANG=C.UTF-8 ---> Using cache ---> ca9f7af4b739 Step : RUN pip3 install -r requirements.txt ---> Running in 7839cb2e5a88 **standard_init_linux.go exec user process caused ""exec format error"" The command -c pip3 install -r requirements.txt' returned a non-zero code: 1** ",2020-06-07T16:47:36Z,gpt-2,https://github.com/openai/gpt-2/issues/247 246,627880304,got this error ," python3 Traceback (most recent call last): File line 9, in import model, sample, encoder File line 3, in from tensorflow.contrib.training import HParams ModuleNotFoundError: No module named 'tensorflow.contrib' ",2020-05-31T03:01:35Z,gpt-2,https://github.com/openai/gpt-2/issues/246 245,623719420,checksum does not match error for 1558M model,"Downloaded the master branch and 1558M model. Got checksum error. Is there a way to bypass the checksum and keep working ? Command given: python3 --model_name=1558M Error in linux: 2020-05-23 23 34.717724: W OP_REQUIRES failed at save_restore_v2_ops.cc:184 : Data loss: Checksum does not match: stored 1096252745 vs. calculated on the restored bytes 1479428755 Traceback (most recent call last): File line 1334, in _do_call return fn(*args) File line 1319, in _run_fn options, feed_dict, fetch_list, target_list, run_metadata) File line 1407, in _call_tf_sessionrun run_metadata) tensorflow.python.framework.errors_impl.DataLossError: Checksum does not match: stored 1096252745 vs. calculated on the restored bytes 1479428755 [[{{node During handling of the above exception, another exception occurred: Traceback (most recent call last): File line 79, in fire.Fire(sample_model) File line 138, in Fire component_trace = _Fire(component, args, parsed_flag_args, context, name) File line 468, in _Fire target=component.__name__) File line 672, in _CallAndUpdateTrace component = fn(*varargs, **kwargs) File line 67, in sample_model saver.restore(sess, ckpt) File line 1276, in restore {self.saver_def.filename_tensor_name: save_path}) File line 929, in run run_metadata_ptr) File line 1152, in _run feed_dict_tensor, options, run_metadata) File line 1328, in _do_run run_metadata) File line 1348, in _do_call raise type(e)(node_def, op, message) tensorflow.python.framework.errors_impl.DataLossError: Checksum does not match: stored 1096252745 vs. calculated on the restored bytes 1479428755 [[node (defined at ]] Caused by op defined at: File line 79, in fire.Fire(sample_model) File line 138, in Fire component_trace = _Fire(component, args, parsed_flag_args, context, name) File line 468, in _Fire target=component.__name__) File line 672, in _CallAndUpdateTrace component = fn(*varargs, **kwargs) File line 65, in sample_model saver = tf.train.Saver() File line 832, in __init__ self.build() File line 844, in build self._build(self._filename, build_save=True, build_restore=True) File line 881, in _build build_save=build_save, build_restore=build_restore) File line 513, in _build_internal restore_sequentially, reshape) File line 332, in _AddRestoreOps restore_sequentially) File line 580, in bulk_restore return io_ops.restore_v2(filename_tensor, names, slices, dtypes) File line 1572, in restore_v2 name=name) File line 788, in _apply_op_helper op_def=op_def) File line 507, in new_func return func(*args, **kwargs) File line 3300, in create_op op_def=op_def) File line 1801, in __init__ self._traceback = tf_stack.extract_stack() DataLossError (see above for traceback): Checksum does not match: stored 1096252745 vs. calculated on the restored bytes 1479428755 [[node (defined at ]] ",2020-05-23T18:47:56Z,gpt-2,https://github.com/openai/gpt-2/issues/245 243,616757022,ModuleNotFoundError: No module named 'tqdm',"After installing the requirements.txt with pip3 I get: when trying to even if I have tqdm 4.31.1 installed. I also tried uninstalling and installing the latest version of tqdm 4.46.0",2020-05-12T15:37:48Z,gpt-2,https://github.com/openai/gpt-2/issues/243 242,609336026,"Native Installation Leads to a ""no attribute sort"" error?","After following the instructions carefully in bash, for the native installation they inevitably lead to this error: I wrote a SO post about it here which has the full stack trace: What should I do to have a successful install and resolve this issue? ",2020-04-29T20:26:19Z,gpt-2,https://github.com/openai/gpt-2/issues/242 239,586046279,Can GPT-2 be used to generate sentences from set of keywords?,"I have a use case where I have set of keywords and target sentences. I want to build a model which will take keywords as inputs and generate sentences as outputs. Can this be achieved using GPT-2 in some way? Example - set of keywords expected output ",2020-03-23T09:08:05Z,gpt-2,https://github.com/openai/gpt-2/issues/239 238,583809266,ModuleNotFoundError when training the model ,"Hi guys, I am working on finetuning a GPT-2 in Google Collab to use in (behavioral) research, but I am very new to this and I don't understand what's going on. I have been using it without errors for a few weeks now, but now the algorithm does not cooperate with me anymore lol :) **Traceback (most recent call last): File line 14, in import model, sample, encoder File line 3, in from tensorflow.contrib.training import HParams ModuleNotFoundError: No module named 'tensorflow.contrib'** I've tried to import tensorflow.contrib manually (import tensorflow.contrib) but it does not solve the problem. Also, I have tried to see which tensorflow was installed, uninstalled and reinstalled newer versions but it still does not work. Can someone please help me? Thanks so much. If needed, here is the google collab link: Cheers, Marloes",2020-03-18T15:25:50Z,gpt-2,https://github.com/openai/gpt-2/issues/238 233,575082734,gpt-2 slow inference,"Using gpt-2 345M model to run inferences in batches between 10 and 100 documents with approximately ~60 tokens is taking ~15ms in a Tesla T4 GPU machine. Why? That looks really bad if someone would like to hook this up in a realtime pipeline... It means ~66 which isn't sufficient for many realtime systems. Am I losing something? Is there any up-to-date benchmark I can use to compare my numbers? pytorch 1.4.0 transformers 2.3.0 CUDA 10.1 apex 0.1 ",2020-03-04T02:00:23Z,gpt-2,https://github.com/openai/gpt-2/issues/233 231,568510944,Modifying to work for tensorflow 2.0,"Since I am running Arch, it is an absolute pain to deal with older versions of software. Therefore, I want to update this to tensorflow 2.0. The biggest problem is in the model.py script, since it relies on tensorflow.contrib which no longer exists (even in tensorflow.compat.v1). How can I easily modify this script to make this work with the latest and greatest tensorflow?",2020-02-20T19:13:09Z,gpt-2,https://github.com/openai/gpt-2/issues/231 229,558750137,ZeroDivisionError: integer division or modulo by zero,"Hey, I've been trying to get this to run, but keep running into an issue when I get to the training phase. I keep getting this back: Loading checkpoint Loading dataset... 100%|████████████████████████████████████████████████████████████████████████████████| [00 00, dataset has 0 tokens Training... Traceback (most recent call last): File line 297, in main() File line 275, in main feed_dict={context: sample_batch()}) File line 251, in sample_batch return [data_sampler.sample(1024) for _ in range(args.batch_size)] File line 251, in return [data_sampler.sample(1024) for _ in range(args.batch_size)] File line 74, in sample self.chunks ZeroDivisionError: integer division or modulo by zero I saw someone else was able to fix this problem by increasing the the lines of text, but I've tried doing that with no luck. (I've tried running it with 1000, 4000, and 10,000 lines of text) Has anyone else had this issue or found a way to fix it? Thank you. ",2020-02-02T20:12:17Z,gpt-2,https://github.com/openai/gpt-2/issues/229 228,556184830,src/model.py gelu uses numpy functions,"The gelu function in the script uses numpy.sqrt and numpy.pi, how does this affect GPU performance, and does it even work with GPU? If not, it should be changed to similar functions in tf.",2020-01-28T12:43:48Z,gpt-2,https://github.com/openai/gpt-2/issues/228 227,555154293,BPE using sequence of bytes. HOW ?,"Hello, I read the paper about GPT 2 : it says that they used BPE on a sequence of bytes and that they only needed a vocab size of 256. I researched the internet but didn't find any explanation on how BPE on sequence of bytes work and why the 256 vocab size. I am confusing since I don't know how this works compared to applying BPE on normal characters and what are the clear motivations since they also say that level LMs don't work great. How this is different. THANKS.",2020-01-25T23:41:49Z,gpt-2,https://github.com/openai/gpt-2/issues/227 226,548348604,recommended VRAM for the small 117M model,"Hello, is it possible to train the small 117M GPT-2 model with 6 GB VRAM using FP16? Recommended Vram is 12 GB so with fp16 I could halve the memory consumption? ",2020-01-11T00:12:37Z,gpt-2,https://github.com/openai/gpt-2/issues/226 225,547350355,Serving gpt-2 model with Flask for conditional samples,"Hi there , There are so many developers want to try gpt-2. I found a lot of people confuse and suffering for deploy gpt-2 as a web service, may be this commit will help them avoid this trouble and save more time focus on the model itself. ",2020-01-09T09:21:05Z,gpt-2,https://github.com/openai/gpt-2/pull/225 222,541255513,"enc.encoder[""<|endoftext|>""] is wrong and nobody realizes it.","Relevant tweet chain: Basically, you're prompting the model with <|endoftext|> (a single token with BPE value 50256 or whatever), but the BPE encoder encodes <|endoftext|> as <| end of text|>, five separate tokens. It's completely different. ",2019-12-21T00:27:21Z,gpt-2,https://github.com/openai/gpt-2/issues/222 221,540816440,how can i get the input and output node names to generate pb file?,"I am working to make file of the checkpoint but i don't know the input and output nodes of the model so i am unable to make frozen graph of the model for prediction. i tried so many ways to make file. In one of the way i am getting error that How to resolve this issue? Thank you for your time.",2019-12-20T07:14:20Z,gpt-2,https://github.com/openai/gpt-2/issues/221 217,537002693,Generate text samples by giving some features or text as a parameter,"I am using GPT2 small model (124M). I have trained the model on real Estate home descriptions samples. So it start generating homes description. I am use the GPT2 following generate function to generate home description samples. **gpt2.generate(sess,length=500,temperature=0.9,nsamples=10,run_name='run1')** Well the generated description well look like this, **2700 NW Dogwood St Unit H104 is a condo in Seattle. WA 98146. This 830 square foot condo features 2 bedrooms and 2 bathrooms having 3 stories and 3 floors.This property was built in 1992 and last sold for $40000. .Based on Redfin's Seattle data. we estimate the home's value is $472830. Comparable nearby homes include 2700 NW Dogwood St Unit H101. 2700 NW Dogwood St Unit H202.** So my question is that how can i give the parameter like **""4 stories 4 floors""** to GPT2 generate function and it will generating the home description samples with this feature (**""4 stories 4 floors""**) and if i give parameter like **2 bedrooms and 2 bathrooms** it will start generating samples with this feature ( **3 bedrooms and 3 bathrooms** )",2019-12-12T14:12:57Z,gpt-2,https://github.com/openai/gpt-2/issues/217 216,535470332,Are there some research papers about text-to-set generation?,"I know this question is a little out of topic. But it is helpful to me. Thank you. Text-to-(word)set generation or sequence-to-(token)set generation. For example, input a text and then output the tags for this text: --> Thank you!",2019-12-10T03:33:03Z,gpt-2,https://github.com/openai/gpt-2/issues/216 215,534201625,No such file or directory: 'models/774M/encoder.json,"I am using mac 10.15.1 and I am using: However, I have an error: FileNotFoundError: [Errno 2] No such file or directory: Checking in the model 774M folder I can see the file: Here the entire message (tf1) XXX:~ XXX$ python3 Traceback (most recent call last): File line 79, in fire.Fire(sample_model) File line 138, in Fire component_trace = _Fire(component, args, parsed_flag_args, context, name) File line 471, in _Fire target=component.__name__) File line 675, in _CallAndUpdateTrace component = fn(*varargs, **kwargs) File line 44, in sample_model enc = encoder.get_encoder(model_name, models_dir) File line 109, in get_encoder with open(os.path.join(models_dir, model_name, 'encoder.json'), 'r') as f: FileNotFoundError: [Errno 2] No such file or directory: Any advise? Thanks",2019-12-06T18:48:18Z,gpt-2,https://github.com/openai/gpt-2/issues/215 213,525303588,Out of Memory Error,"I get the below errors when running gpt-2. The model runs in the end and seems to work, but is there any way to fix this? Thanks! ",2019-11-19T22:24:00Z,gpt-2,https://github.com/openai/gpt-2/issues/213 210,523076958,No module named 'tensorflow.contrib',"I am trying to use python3 --top_k 40 **But it gives me this error** Traceback (most recent call last): File line 9, in import model, sample, encoder File line 3, in from tensorflow.contrib.training import HParams ModuleNotFoundError: No module named 'tensorflow.contrib' **Details of Tensorflow** tensorflow 2.0.0 tensorflow-addons 0.6.0 tensorflow-estimator 2.0.1 **How can I solve this issue?** Can I replace ""from tensorflow.contrib.training import HParams"" in ? if yes, how? Thanks",2019-11-14T20:06:51Z,gpt-2,https://github.com/openai/gpt-2/issues/210 209,522833474,The true names/sizes of the 4 GPT-2 models,"Can you please clarify the true names of the 4 now available models? : - according to and and - according to and : - according to and and - according to : - according to and - according to - according to - according to This makes downloading them through incredibly hard. It'd be really useful if you could put the true names either on the readme or in the download script. Thanks",2019-11-14T12:43:37Z,gpt-2,https://github.com/openai/gpt-2/issues/209 207,520736310,AllenNLP and a Question Generator,"I would love to see gpt-2 used to generate questions. I have a working example here that uses BIDAF and an automatic question generator. Do you think the full model will ever be released, or can you submit the full model to AllenNLP with some type of com-pliancy license? ",2019-11-11T04:40:37Z,gpt-2,https://github.com/openai/gpt-2/issues/207 206,520497084,Exposing a programmatic REST API,"I was wondering if there was a way to expose an API of some sort for the conditional generator. I am not good with Python, and I don't think that wrapping with Expect library is the way to go. Also, ideally one would be allowed to put characters in input to the model, since I think right now if you use carriage return it will treat it as a separate input. And getting the formatting right is critical to some things like poetry, lists, etc. Anyone know of some code to expose an API? I didn't see any in the code itself, nor do I see any issues opened regarding this. Thanks.",2019-11-09T18:48:17Z,gpt-2,https://github.com/openai/gpt-2/issues/206 205,520482647,Documentation out of date,"I noticed this PR which updates the DEVELOPERS file: It is not yet merged. I was reading through I tried pip3 install tensorflow==1.12.0. I found that version doesn't exist. I used: pip3 install tensorflow==1.13.2 It seems to work with this version - however it stopped mid sentence and had some warnings. I am using a Linode, the default size version with the latest Ubuntu. ",2019-11-09T17:13:20Z,gpt-2,https://github.com/openai/gpt-2/issues/205 204,520242654,Running using Docker instructions results in libcuda.so.1 missing error," ` $ docker build --tag gpt-2 -f Dockerfile.gpu . ... $ docker run -it gpt-2 bash ... root export PYTHONIOENCODING=UTF-8 root python3 Traceback (most recent call last): File line 58, in from tensorflow.python.pywrap_tensorflow_internal import * File line 28, in _pywrap_tensorflow_internal = swig_import_helper() File line 24, in swig_import_helper _mod = imp.load_module('_pywrap_tensorflow_internal', fp, pathname, description) File line 242, in load_module return load_dynamic(name, filename, file) File line 342, in load_dynamic return _load(spec) ImportError: libcuda.so.1: cannot open shared object file: No such file or directory During handling of the above exception, another exception occurred: Traceback (most recent call last): File line 7, in import tensorflow as tf File line 24, in from tensorflow.python import pywrap_tensorflow # pylint: disable=unused-import File line 49, in from tensorflow.python import pywrap_tensorflow File line 74, in raise ImportError(msg) ImportError: Traceback (most recent call last): File line 58, in from tensorflow.python.pywrap_tensorflow_internal import * File line 28, in _pywrap_tensorflow_internal = swig_import_helper() File line 24, in swig_import_helper _mod = imp.load_module('_pywrap_tensorflow_internal', fp, pathname, description) File line 242, in load_module return load_dynamic(name, filename, file) File line 342, in load_dynamic return _load(spec) ImportError: libcuda.so.1: cannot open shared object file: No such file or directory Failed to load the native TensorFlow runtime. See for some common reasons and solutions. Include the entire stack trace above this error message when asking for help. root `",2019-11-08T21:30:02Z,gpt-2,https://github.com/openai/gpt-2/issues/204 203,519961501,Error when testing with Docker build / run,"I got an error when testing with Docker build run and I can't see how to fix it. Can someone assist? Build: Run: ",2019-11-08T11:22:45Z,gpt-2,https://github.com/openai/gpt-2/issues/203 201,518308205,training from scratch,Will the code for training from scratch be released after the 1.5B model? ,2019-11-06T08:11:43Z,gpt-2,https://github.com/openai/gpt-2/issues/201 200,515912992,Question,"I am very sorry, but I just want to know what is ""fire"" in generate_unconditional_samples.py?",2019-11-01T05:00:07Z,gpt-2,https://github.com/openai/gpt-2/issues/200 195,511362154,Retraining with Beatles’ lyrics only,"Suppose I want a model of all Beetles’ lyrics, how do I train the system from scratch with them. Assume I have them in a txt file. What would be the steps? My goal is to input Beatles lyrics and then output new lyrics in the same style. Detailed answers or pointers are highly appreciated. I noticed this project is aiming to achieve ""general"" capabilities as opposed to ""specialized"" ones. If I am going about my problem the wrong way, or misunderstanding the premise of this project, please kindly correct me and provide pointers on how I'd achieve my goal elsewhere if possible. Thank you. -Andy",2019-10-23T14:27:52Z,gpt-2,https://github.com/openai/gpt-2/issues/195 192,507927122,What is the maximum model prompt size for interactive samples?," Is there a maximum size that's recommended? I notice that if I feed it with very long prompts it'll write a few sentences, then devolve into a schizophrenic rant. > I was, was, was, was, was, like, was, like, was, like, like, was, was like, was, was like, like, was, like, like, I was,, went away, was, gone, was like, was, was, a, was, was, was, > What we > ’ > ’ ’ ’ ’ ’ ’ > > > > ‵ ’ 
 ’ > ’ ’ > ’   ― > > > ’ > That ’ ’ ’ > ’ ’ > ’ ’ ’ ’",2019-10-16T15:27:48Z,gpt-2,https://github.com/openai/gpt-2/issues/192 191,506274808,Does the model prompt support multiple lines of input?,Currently I have found no way to enter multiple paragraphs or a list format. Enter and all other newline methods I've tried do not work.,2019-10-13T05:00:23Z,gpt-2,https://github.com/openai/gpt-2/issues/191 189,501348841,Code not working for tensorflow==1.12.0,"I created a virtual environment with tensorflow==1.12.0 to run this code, as mentioned in starter guide. But, program is throwing errors on tf 1.12. But, it is working fine on tf 1.14. Please update the starter guide.",2019-10-02T08:28:22Z,gpt-2,https://github.com/openai/gpt-2/issues/189 188,500868673,Feature/Question: With GPT-2 is it possible to get previous word prediction?," With GPT-2 is it possible to get previous word prediction? Hi, I say this after seeing this And wondering how I could maybe write a method that would allow me to predict the previous word? (ideally for GPT2) Many thanks, Vince.",2019-10-01T12:40:00Z,gpt-2,https://github.com/openai/gpt-2/issues/188 187,500440080,Questions about nucleus sampling/Potential Bugs,Why is nucleus sampling implemented on top of the _top-k_ logits? ( in ) My understanding from the paper is that the threshold should be implemented on all of the logits.,2019-09-30T18:20:44Z,gpt-2,https://github.com/openai/gpt-2/issues/187 186,499103228,Quick installation instructions.,These are some quick install instructions for users that aren't very familiar with docker. I had a large docker install script but I accidentally wiped it for Nvidia. ,2019-09-26T20:35:59Z,gpt-2,https://github.com/openai/gpt-2/pull/186 185,497283730,confused about vocab.bpe and encoder.json,"I'm reading the source code. And I have two questions about vocab and encoder. Please help me with that. Thank you in advance. 1. For vocab, I take the second row (Ġ t) for example. But I found ""Ġ"" appears in many rows(for example the third row). So why isn't it one-to-one correspondence? 2. Are the items in encoder.json the subtokens from BPE? I take for example. Why does appear here? ",2019-09-23T19:28:21Z,gpt-2,https://github.com/openai/gpt-2/issues/185 184,496677094,Docker installation does not work.,"I installed the Dockerfile.gpu version, at which I found that gpu passthrough is unavailable in Windows 10 Home, so I just changed the tensorflow version to the cpu version, and had to change to 1.13.1 in order to use the sort function as shown in this pull request: Now I get the following output: As you can see, the program seems to get killed by something in the virtual environment. I am unsure what it is as I have not installed anything else in this environment.",2019-09-21T16:56:55Z,gpt-2,https://github.com/openai/gpt-2/issues/184 183,496628575,Generate encoder.json for other languages,"Hi all, Please tell me the way to generate encoder.json for other languages French, Spanish... I tried to create a encoder.json but it's not exactly the same example of GPT and failed in run model",2019-09-21T08:07:15Z,gpt-2,https://github.com/openai/gpt-2/issues/183 182,490543355,Large model?,"The code in never calls the large model, it uses 124M. Is this correct?",2019-09-06T22:15:22Z,gpt-2,https://github.com/openai/gpt-2/issues/182 180,489779325,"error : ""AttributeError: 'HParams' object has no attribute 'override_from_dict'""","error : ""AttributeError: 'HParams' object has no attribute 'override_from_dict'""",2019-09-05T14:09:17Z,gpt-2,https://github.com/openai/gpt-2/issues/180 179,487805013,"Installation guide, please","This code looks great and all, but without a proper installation guide, most people can't even start playing with it. I have checked a couple of pages with supposed installation guides, but none seem to work for me. Please, just add one, or create an .exe file.",2019-08-31T19:42:56Z,gpt-2,https://github.com/openai/gpt-2/issues/179 178,487576713,Developers.md," says that we should install: But when I run: it throws: Looks like sort method was launched in tensorflow 1.14 onwards. Hence this readme can be updated. ",2019-08-30T16:43:38Z,gpt-2,https://github.com/openai/gpt-2/issues/178 176,486959756,while text generation first past is repeating again and again.,"After first generating first word, the present is not getting updated. Please change it to ",2019-08-29T13:30:00Z,gpt-2,https://github.com/openai/gpt-2/issues/176 175,485643271,Slow inference speed,"Hello, I am currently fine-tuning GPT-2 with my own dataset. There are 90000 documents in my training set and it takes around 10 seconds to finish one epoch. However, when I try the inference with the interactive_conditional_samples.py, it takes me more around 30 seconds to generate one document. I cannot explain why it becomes so slow. Is it because of every document is sent to the BPE encoding separately? All suggestions are welcomed. Thanks. Helena ",2019-08-27T07:46:58Z,gpt-2,https://github.com/openai/gpt-2/issues/175 174,484851648,Conditional generation from input file of contexts,"Fix for #145 Instead of passing contexts one by one interactively, takes an input text file containing a list of contexts (one context per line) and outputs the generated text to an output file (one sample per line)",2019-08-24T18:51:05Z,gpt-2,https://github.com/openai/gpt-2/pull/174 173,484832348,"ResourceExhaustedError: OOM when allocating tensor with shape[3,16,1024,1024]","I'm getting this error when training the model 345M on a dual GTX 1080i - 8GB RAM. ",2019-08-24T15:08:40Z,gpt-2,https://github.com/openai/gpt-2/issues/173 172,484055140,How to get reproducible interactive conditional samples?,"The 'seed' argument of interactive_conditional_samples.py helps getting reproducible sample if I restart the script, i.e., if I start from a new session. However, I cannot find a way to get the exact same sample for the same prompt within a single session. How can this be achieved without creating a new session?",2019-08-22T15:14:25Z,gpt-2,https://github.com/openai/gpt-2/issues/172 169,483388144,Why not the 1.5B model as well? Even 8.3B (instructions) have been posted by NVIDIA,"Now as I understand, and have audited the code myself - NVIDIA has released an improved 8.3B solution, albeit not a MODEL - I was able to follow it through to the point that I fully understood what actually RUNNING this monstrosity would take in computing power and dollars considering I'd probably need to settle for cloud TPU through google or amazon. The cost was unappealing to say the least. Now we're already to the 755M model - THANK YOU! Don't think I'm not grateful. But it just makes sense to release the full 1.5B as most of your partners have said there is little difference between the two anyways, and smaller researchers who don't want to rack up a $3000 compute bill to train a model would love to have it. I understand this is actually probably the biggest ""danger"" you guys were referring to early on - certain people not having access to re-train the software. But at least having a point of reference helps us connect the dots. Thanks again. ",2019-08-21T12:23:15Z,gpt-2,https://github.com/openai/gpt-2/issues/169 168,483107945,Does gpt-2 use a POS tagger? ,"Hi, sorry it is not an issue just a question, does GPT-2 use a POS tagger, a parser? If not, loosing such data isn't a loss in potential accuracy? ",2019-08-20T22:13:50Z,gpt-2,https://github.com/openai/gpt-2/issues/168 165,481068209,conv1d naming?,"I am wondering why is this function named conv1d? It seems to be a linear transformation to me, not a conv1d operation. ",2019-08-15T09:23:48Z,gpt-2,https://github.com/openai/gpt-2/issues/165 164,476053840,ZeroDivisionError: integer division or modulo by zero,"When I run my train.py, this is the output Loading checkpoint W0802 09 48.051176 6828 deprecation.py:323] From checkpoint_exists (from tensorflow.python.training.checkpoint_management) is deprecated and will be removed in a future version. Instructions for updating: Use standard file APIs to check for files with this prefix. Loading dataset... 100%|███████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████| [00 00, dataset has 0 tokens Training... Traceback (most recent call last): File ""train.py"", line 293, in main() File ""train.py"", line 271, in main feed_dict={context: sample_batch()}) File ""train.py"", line 247, in sample_batch return [data_sampler.sample(1024) for _ in range(args.batch_size)] File ""train.py"", line 247, in return [data_sampler.sample(1024) for _ in range(args.batch_size)] File line 74, in sample self.chunks ZeroDivisionError: integer division or modulo by zero My idea is that I may have too small of a dataset, but upon consistently increasing size, it is not to be. I have moved from 300 lines to 3100. Help?",2019-08-02T08:11:50Z,gpt-2,https://github.com/openai/gpt-2/issues/164 163,475866945,Failed to interpret file %s as a pickle,"Every time I run train.py with my dataset, I get this error. Loading dataset... 0%| | [00:00 main() File ""train.py"", line 173, in main chunks = load_dataset(enc, args.dataset, args.combine) File line 27, in load_dataset with np.load(path) as npz: File line 450, in load ""Failed to interpret file %s as a pickle"" % repr(file)) OSError: Failed to interpret file 'lyric.npz' as a pickle Any solutions? What am I doing wrong?",2019-08-01T20:10:24Z,gpt-2,https://github.com/openai/gpt-2/issues/163 162,473633252,Training for Style vs. Topic,"When using GPT-2 via the interactive prompt or finetuning with a secondary corpus, the model seems to learn both writing style and what topic to write about simultaneously. Let's say I wanted to teach GPT-2 to write about modern architecture (topic), in a Twitter vernacular (style). Is there a way to indicate to GPT-2 which parts of the input are topic related and which are stylistically important? ",2019-07-27T12:23:57Z,gpt-2,https://github.com/openai/gpt-2/issues/162 161,473574476,Update README.md,,2019-07-27T00:02:55Z,gpt-2,https://github.com/openai/gpt-2/pull/161 160,472328406,Batch of samples with different length,"Hi, I'd like to feed short samples of various length to the model, and I would like to put them in a single batch. Hence my question is, does the model support padding? If yes what is the token? ",2019-07-24T14:50:04Z,gpt-2,https://github.com/openai/gpt-2/issues/160 158,469198605,Training question,"Hi, I have taken a look into your model and looks really nice! I do have a question due to I am pretty new to Tensorflow. How can I do it in order to train my model using for example a Spanish Dataset? Thank you!",2019-07-17T13:37:15Z,gpt-2,https://github.com/openai/gpt-2/issues/158 157,467814571,Arguments for multi-layer decoder-only Transformer,"Hello, I recently started studying language modeling and GPT(-2) in particular. While I start to understand the way it is I do have some questions about its architecture. In OpenAI's paper it is stated that GPT (and GPT-2) is a multi-layer decoder-only Transformer. From a higher perspective I can understand that an architecture is useful for sequence 2 sequence applications, but that it becomes less attractive for language modeling tasks. Therefore, it seems logical OpenAI decided to stick with the multi-layer decoder only. However, during the stage of GPT, in these decoding-layers, tokens are still encoded and eventually decoded, right? I'm not sure whether my question is clear, but it basically comes down to this: in GPT's paper it is stated that they use a decoder-only transformer, but I cannot find any arguments this decision is based on. What would be the difference if they stuck to the regular Transformer architecture for example? I hope someone is able to give me more insight into this. Many thanks in advance.",2019-07-14T09:22:48Z,gpt-2,https://github.com/openai/gpt-2/issues/157 156,467672940,no such file 'models\\\\345M\\\\encoder.json' but I have encoder.json,"Running generate_unconditional_samples.py outputs this error: outputs the same error. I even tried replacing in all the files with because I downloaded the 345M model. I also tried putting a direct path to replace line 109 in encoder.py: I tried and it gives me the error I don't know what's going on with the at the last line there. Thank you! ",2019-07-13T05:13:55Z,gpt-2,https://github.com/openai/gpt-2/issues/156 155,465188489,Legit real-world use cases?,"Hello, I've been reading about this project, and many of the articles referring to it discuss the potential abuse or whether it does or not represent an actual risk. I also read the mega thread in #16 and draw some conclusions for myself: 1. Most people who wants the full model release argue it's ""for the sake of knowledge"" 2. I feel like an ample percent of those are actually internet trolls that want a fun-and-easy to use tool for generating scam emails and such 3. Some people is actually concerned about the potential abuse and understand the caution on not releasing the full model Now, what I didn't saw, neither on that thread or in the articles speaking about this project, are actual legitimate use cases for this technology - far from the obvious ""research purposes"". So let's forget about fake news and internet trolling, I honestly don't see a situation where this would be of any help - or let me rephrase it... where this **should** be of any help. I've seen commercial offerings that pretty much sum up to ""are you too lazy as for interpreting your own data? let our bot write reports for your stakeholders so they feel your project is going somewhere even if you don't know what you are doing at all"". The other real world use cases I can think of would be (non)writers who instead of paying others to write their books as they do currently, would use some sort of AI to bake standardized best-sellers for their own profit (with a huge marketing effort ofc - this could also help on that regard), or crappy news agencies who deprecated all the reporters in favor of interns who can type one paragraph so the text auto-generation tool can fill the rest of the made up article. To sum up, I'd really love to hear some legitimate real world use cases which don't completely suck for this technology, from people who are actually working on it. Cheers!",2019-07-08T10:52:39Z,gpt-2,https://github.com/openai/gpt-2/issues/155 154,465037419,Why gpt-2 could apply to other tasks without fine-tune?,Thank you!,2019-07-08T03:28:14Z,gpt-2,https://github.com/openai/gpt-2/issues/154 151,460893652,Generating text on custom encoder.json of much smaller size. ,,2019-06-26T10:38:25Z,gpt-2,https://github.com/openai/gpt-2/issues/151 150,459587272,GPU configuration,What GPU configuration is required to train the 345M model,2019-06-23T15:19:46Z,gpt-2,https://github.com/openai/gpt-2/issues/150 149,458793821,Samples lenght,"I want to adjust it so that it gives out shorter samples. If you limit the length, then gpt2 breaks phrases",2019-06-20T18:01:11Z,gpt-2,https://github.com/openai/gpt-2/issues/149 148,457907619,Intermediate layer output,"Is it possible to use the intermediate layer outputs and generate text ignoring the layers on top? Basically, I want to check quality of generations as we keep on adding more layers. What modifications in the script would I have to make for the same? Thanks.",2019-06-19T09:35:30Z,gpt-2,https://github.com/openai/gpt-2/issues/148 147,456654359,How to freeze checkpoint graph to .pb format?,"Trying to freeze GPT2 fine-tuned model but unable to figure out what the output node name will be. Using this code as a reference, I put this together:- but I get > AssertionError: is not in graph So what should I put as argument 3 output node name in freeze_graph?",2019-06-16T15:40:47Z,gpt-2,https://github.com/openai/gpt-2/issues/147 146,454638219,"Why does the generated text contain ""??"" (⁇) characters?","The generated samples during learning contain sequences like ""⁇ ames"", with ""??"" being a single unicode character. It seems that this character is used when the model cannot complete a word sentence piece, but I am pretty sure that it should be able to complete ""ames"" to ""James"" with a vocabulary of 50000 and an input text that contains ""James"" quite a few times. On the other hand, it seems to be quite consistant in not learning uppercase ""J"". I see ""⁇ us"" as well or ""⁇ udging"". At some other places, ""⁇"" stands between two words without any hint on what it should represent. I am 100% sure, that the vocabulary was generated from the same text as I used as the input. If it is some issue with the vocabulary: Is there an option to fix it for a trained model?",2019-06-11T11:37:41Z,gpt-2,https://github.com/openai/gpt-2/issues/146 145,453908317,How to call input.txt and write to output.txt for interactive_conditional_samples,"I did setup the system on windows 10, x64, 32 gb ram, core i7 2600k and GTX 1050Ti When I run this It asks Model prompt >> Then I type my word such as Pikachu and output is as below. This output obviously seeems incorrect and unrelated So my first questions are as below **1 )** How can i give input.txt as input which will have few sentences and write output to a text file? **2 )** Also are there any other parameters than these? interactive_conditional_samples.py hparams.json **3)** How can i make it work on GPU instead of CPU? **4)** Can I make it return more logical output for given sentences? 5) How can I make it run on GPU instead of CPU? I have GTX 1050 Ti, 4 GB ram Here is the full input and output of my CMD ",2019-06-09T15:21:07Z,gpt-2,https://github.com/openai/gpt-2/issues/145 144,453904632,pip install -r requirements.txt - Errors,"I run this command and i am getting the below output My OS is windows 10, X64 ",2019-06-09T14:43:33Z,gpt-2,https://github.com/openai/gpt-2/issues/144 143,453843851,Subreddit, that is composed entirely of different gpt-2 models that are posting and commenting on each others posts and comments.,2019-06-08T23:01:06Z,gpt-2,https://github.com/openai/gpt-2/issues/143 142,451195304,Add verification for number of tokens in encoded input,"Encoded tokens array should not be greater than model's n_ctx 2 or the one supplied by --length parameter. Fixes a crash on too long inputs. Issue: #121 ",2019-06-02T14:03:16Z,gpt-2,https://github.com/openai/gpt-2/pull/142 141,451191555,"Using VScode in Conda , no models found ""import model, sample, encoder""",,2019-06-02T13:22:28Z,gpt-2,https://github.com/openai/gpt-2/issues/141 140,450885364,Announce dates of staged releases,"It's great to see that OpenAI plans to release better models for public use in the future. However, I disagree with the current staged released strategy in place. Rather than announce when these releases will occur, OpenAI is just releasing them sporadically whenever the time is right for them. This is even worse than not releasing the model at all - isn't the whole point of staged releases to give time for others to prepare? Partnerships do not necessarily fix this issue; the general public should at least know approximately _when_ these new models will release.",2019-05-31T16:39:36Z,gpt-2,https://github.com/openai/gpt-2/issues/140 139,449142071,ERROR: Could not find a version that satisfies the requirement tensorflow-gpu==1.12.0 (from versions: none),"After running I see I'm running on a macOS Mojave. I see a closed issue for this repo that says only Python 3.6 is supported. And I see a closed issue for this repo that says only Python 3.6 is supported. So that's a bit confusing. If I have to use Python 3.6 on my mac, is it best to use a closed issue for this repo that says only Python 3.6 is supported? Also which version of Conda? Looks like there are a few versions; Anaconda®, Miniconda, Anaconda Repository, Anaconda Enterprise. And doing something a closed issue for this repo that says only Python 3.6 is supported seems right?",2019-05-28T08:49:02Z,gpt-2,https://github.com/openai/gpt-2/issues/139 138,448422352,Estimate of sequence probability ?,"Is there a simple way to get the probability of the language model generating a sequence? Given different possible sequences is there an easy way to compare their relative likelihood ? Maybe as an optional output given for encoding and for in batch generation? The use case is for comparing the systems preference for multiple choice options, and for ranking its preferences given a batch.",2019-05-25T05:35:07Z,gpt-2,https://github.com/openai/gpt-2/issues/138 137,447354410,How to increase the maximum length?,I'm trying to figure out how to bypass the 1024 limit on Length but it crashes. ,2019-05-22T21:36:58Z,gpt-2,https://github.com/openai/gpt-2/issues/137 134,445150679,Fix typo in docstring,indefinately -> indefinitely,2019-05-16T20:42:32Z,gpt-2,https://github.com/openai/gpt-2/pull/134 133,444719614,Implement nucleus (top-p) sampling as an alternative to top-k.,Implements the sampling from ,2019-05-16T02:03:47Z,gpt-2,https://github.com/openai/gpt-2/pull/133 132,443698358,Why does GPT-2 use a 500MB model.ckpt txt file that contains Asian characters LOL?,"model.ckpt is almost 500MB in size for my download of GPT-2. Is this training data? Why the asian characters? How does GPT-2 USE this? 별ꠧ㰝噛빲࿆붣�㶣굑뵇ᦁ㸯䱝㶹甲밉ꬶ붨᷏붸手㹓㪚붒쑧㺏飩㷭뙲벧虧븽쁥㲚쉦㵒ᄃ봁హ㹞츙㸼㷑렊볕 㸀㶎뷻뵱㐄㶙訪브偱붤㴃둿㳆解㶔ͽ㱞餑부椨뵠㱏㶏⦱뷕伝뵋䁅㸉捺㴘鴆㸁벻蔺볯빱鐺범꿗봍☦븴಄뷆킔㸆윂㷕ᵜ봐횵㱊ᲁ㷘㗛㹶蘤뵅ꔲ뷄頱㸬끢㴂擄㸜㦞㸻㖨㸍ɳ㸄킹뷁봣ᰇ㸨겲뷆꯰㹺∼㷦暮븝쓏㹂⢻빯㸘븕뇈㷎史빇瓌벂쾼브붔也뷫ἶ믨�뺉ⶠ븩䫿뷝먮붛ꡌ븜䖏㶌꩘㸢륛뺅ఝ㮫홷㴃Ꝫ㹝뵆鼔㹢濟㷆쎪㶶鋈㸒쫜밲溝㯐﬒봈㳠븘邙㷞䉍빑캐㴏펿㱚ꂢ㴜℅붡㶓큠볥㶳郟㴿珌㴷⯄뷫⌁벂쫱봃ᦦ㶖붳잁뷰忿㷙쎧뷞༳㶧穀믘䭠븎齵븎軳볞�㶙ਜ믧ᕞ㲿㶿밲邝㴽툛븏茉뵑犚㷝新㲗♮봺㗾뷘儔㷶捴㱾㳺㷢ὕ븚狕㷮䦱㳞卦뷬㳦ᶀ봞Ꝼ㳲㵔将㶉夸㶻㷅븚㶒窲븘圄㷩檳㸉捯봉竂㸃维빂⃛㸥뺼뷪㲺취봕뢘븹盔븋ˮ붲⯠㭑é붳䥜㶟憱㸗㵋㭨㺇䧜뺗�뵼ꦀ븰ኪ㸛췪빋慌㹈௨뷳♂붝ื㶝玭㹛ꔯ㸾蹚㷌꽳뵈緁붙㭽벶܈㶕寛빯醢봼蘱㶸ꁊ뺇뷇驃㴊㷋鋹㶢૬붴ꚟ븃ᴶ빲减밋֗㷉敦봿唺㷷㣄㰃엓봉伟빓斫븅냿㴙❋벏뱐ছ뭫뵚櫅㶆丌㷸鼌뷦㶈ෞ뭙而㶏ⵓ븨뷛洄빏�㪡Ȭ븟햩㷸�㷟ꮓ봼᭴뵯콏뷗붭玖㰳귡㶣೜뷤ᖟ붉䋓븤璶벛犀뷣낥㷲볫嗷비ﭓ붵껺봥胒㷼斛봥Ḅ뷈봊꾢븞ꄯ㶍皀㶹捴벯ﮎ㷚腑㺁䄂㸙項㶍到㞳픧㶁鳠㸿ᇯ븗봪脚㵨㶎铙㷑༂㷥싐㲶皠붤ೢ븪ው㷻㸮Ⓛ㳵痂붔후㰓껬븜궱볃糄㷦㞔㹒뷵ᔁ벟稯믑八㱢ἡ봰렝㴅ਨ㴁췏㸷℉빨荟붼혴㵻❋빜鸾봾빯韇㹯ᣔ㹪劮뷪䔼배뵳グ붌砼뵭ⴀ㸚�㸃㷄ᑉ㷂捙㸼튆㶗�박㸊㲲⨷㸲無볷䁅㴛թ㺐퀱㯲鎌㸐ꇕ봀餉볤볙붓ၶ빙ꇩ븑뺃ᤩ㩎І뵏䓅㺆䛟㴑꘸㺁끹㸝哃붉遲㯴◫붢蕽㰑鉣붑괻븃㴶墧븽⋾붖厬㷲䘦㴓ﹿ㸇䈹붇㸆㵸᧨㹀䧤㷫e㮓䒩뱋㖩뷬ᬝ㵓鹴붯灩뺝줠㸱淵부뵜㡻㶛㣎㴝ꥨ㷱艳㶈㔇븲駖븷㶅ઌ붔䷘뷜唯빉灾㺁뎋뵸殨븧쓠㺨뷧⮳뵔툒붯鍭㴪핁㸤㵙�벶㷮镪봿烉㹂㺬붽�븢₰믪궔붕핚봔퍢뵎짲㲂⼹㶎薪㵾᮸벺䄗뵮㷏㸐髞㸲흌㵣ಥ붍箧밉鄖㵝ᡧ법喗㴡豮㳕ⷤ㴀䯹뱟☬㴾䤚븀픱㴔쎮붬砡븨དྷ밬彚븴뜃븉ⶎ빑葝붢紉븻ᕁ㶪淡㸊솝㷀뱒㲀䞹㸑橣㳹幍몀⦧㴣䴆㵸絲븬뮯㹂㌯븿皫㷰䒦븖檉㸴퍉봗묤붬ꜷ봔苚붱쳂볝ຓ㸺瘄벚㏲㶒త㴦軨부㗇㶓≱벗붢㸗봌㕙봎䟷뺦劺㴅䊾붢♀붽㲴뭈❽㳦䧆㳫ጾ뷠뺥㷍ꦤ봸嗒㷏䍩뵻毁㷉ﬡ물窹㺼뾟㬓붹썖붿뷀녃㸓鴓븛澹붣ჼ빠㌬㶱簤므ꟙ㶠ᰬ㹺襊붒�㵾ᤎ㵫邑㶛짚뵛┚㴬炚뱅ᶫ볅逃㹱뺋ၴ㵃ẇ붐쮫뷢К㱉൛㲕旕㱩뷃᪐㶰ꬳ㱗ᆓ㶑▙㷽邶븴蟆㺚႒㴍謂㳸볧瑂뺓潢㴮㴗㎛빟煉㯃ꀤ뷖朴㸬螆빔絷빪↧뱃㴿ᶊ㸂징빺翝븶꥙븴Ⱙ㴫䏽벿琘븎렭㴁飬㺧ˮ믫آ㮫纹뺂㷄밗㠺㴘빈匩붎ᚍ㸓攣㶂ꤶ볞㞻붉䴵붹붸ʒ㳘귂뷝잧붕ၷ㶗틈㵓臞㳔獶㷮著뵌㷝䅻㶢忙뮱鳠붨〿봤븍㶛㺜붖㸪Ꮿ㳇膩붍䯟㱃謺㱚笲벓꽱㲸⋒붴沆뷗⨭㴜禯밥␲뱻욁뷫좤뵧칕븛㬡❲㹅땳붐䦪㴐߅㯚⼏봕붼轝㶼≽벪㴝㶉뷅봰ꩯ빈ꄕ뵄ꄱ볨춆㶗⏋㵟冶㫜諧㪚⢤뱓龳밻頖뷴�븊髵븧�㳸䚻봣ࣗ븧ꧢ㱣嶪㴋㿆㺣䊵븘�㳽慩㺁줞붥蔟볛邰㹔ピ뷌쟏븿虝뷊ՠ븭ᓕ밮�붐笉㵣俜밃ꕜ㲮⃋㰂瘗븱褤㵪ꦪ봕�뵗쨾㳴⧬㴅ࣝ㷘鋼㵶븫㺕㵫暱㸥䫼뷛땨빂췺븚燛㸟ﶎ㸶뤚㶍㣕㶫摋㶐篃붦㖑㰮의㸚㶪븉뺁ﴃ븴硎븕螸㳍ഝ붲ὺ빅侖봕眶븧媺㷘ᐇ븨敨㸢嫇㵚ﰥ㴏虱㷶㶤洏뷣薻뷰麋㶈篰븾�뷝댣㨸ᛎ㺙脋봜䵠㶵횘뵫럘㸏ㆾ뺜弖봼ﳨ뷝询붧婴㸤뤰븉᳒뷱봢椗븽&븟첃㸃匁㶤췈㵫乆㸀뵐뷜訡붮�㹨㷨쇁빜ﳃ붠밡㵖䙾뱔ₛ믈㰐뷢蠽㺱㴼齊㯵霙㸈荭븊㸛꽲붒⿜㶦圗뷓⼓뵚묧㱃致㶜ᆏ븋戄봦孫㸆ឍ㸚�㶈籔㷕뱎�뮑脬븤➮㺑蹀㲫᣿㲙쪲뱰�㷾艎㱺ᙈ㴉捝㷳ᰇ봐덜뷠놇붣뢻㶆鰖븪詛붗ጶ㶆뵚띃㷣㶹撠븅Ჳ붧〞뺼娊㷠攖㺁纙봊⣯뷱욹뷮�뵼ᥒ㺝懒㸬盋㲟뵤釐븑븂鬦㷢쪏㸳ᤧ㶪Ѭ㲄咘㷖㔟㸡ꨵ㶨豯㵼㺎⧣㸐ﶚ븷�㶸�뺬ࣹ빆鞲㶲㴜坸븘鍿㲘ᅾ㲰ඤ㺇쑂뱌黡뵂賾㭑暑붏呋㺮趨㸣跔뵸쀘붉빦㹩聢㸙빨�㶯匡불꼿㯑㶸樝빧儵㸫낺㵖癋㹀튗뷕牂븯殞빗㬘㺫癪㴂腚㺬啠붡ߦ벙미ҍ븞蝲변笼붌헁㹴浑븀�㷤꘎㸜멀븟ⶕ㸺邁㬈绅㹋숊뱳㷞뵰⮾봜뷂뤎뷌䘶봳ᬧ㺑㷵⪥㻈欄븻怚봾䕯㴫歬㶙㦁㻄꩒뷟๹뷽猪믜핚㴤빂塑㸳⬂㲺誨㶲뷄훳㳈仆㳓辥봌镆㶋輨뷷◛빇웉㶊窳㸖출㸡㵫ጷ밞榟븝졤㷠恣㵸핊뵺迿㹁ꓱ븕髚㱔檌분䅔㹠ᮌ㶝Ϛ㸰뵟褿㴲릳붮鰄㲪趬븂�봨ꌇ벰롥붠陷뫽坬㷷䆑븒毭븝欃㵵㢕볪璘봃돟㺅剸㴤⼂붷륱븾댓㶧翠㸸ܶ㹄㉫뮼腏㴘辮붎㚶㲈ﺹ㴽ୱ븈ꩭ붺꼁봟㸙턨㴮挳㷧瑌㸚埜벐㒚뷡뱢봓덢㪯脕붩玔㷗옺붣ᦥ븫ௌ붼鱩븬藍뵽⇯㸒붉㴮銎뵭亢븄ꮾ뵍꺻부ᮏ봦辁믨쏼㸳뀗밃索㶠ⴞ㶛⿱㺗蒚㸜뽍㸢᠎뭔�믮젵㸀⸐㶻䲝붶糞㶎㸂싽㲏嗹㹔ꏬ㷚⢌㵀裹뮯ĩ붂䫜㹗鼜㸄Ŕ㵮㮨㸍읹뷣亏㶉ㄉ붐鐱벛箕뵓ᰳ㸁諼㻖뻖︇붚˅㸸῵㵃뷂堔㶰蔤㹼㳄붍뎡봰㓰뷷틛븲봚쥂붗弞㸾눲붟挓벃笏㶢쇲㳟㾑㸂�㲨ധ㸭甓봮绸㺊欦㮁꣈㳢ᕚ㶰㜈㸃磫㶓墋㸚Ჸ붶尷㹧�㵢䭯㷻禥봘䈚㸾黭봗䷆볪ꘖ뺅�뷂ᗒ㺆㮔븓⭢북꣑븨熧㰞﹩뷴㮔퍁뷀鰋㱤窪븗桙븄뷧敆㷵�㸠㙄㷃筶㴌㶀컝㷚귶㷪婬불謺㶵짽붽摏붛굹붒駦㷑쑗뺓ᚘ㹎廝㸅홡뱥ȿ빌幽뱧✱㴟絅빤֊㶚웳빖㣹빁鐂뺬陵뵇烱밞躟㴫ᐍ㳾⅘㶽紻㺚㸮᧦분ሡ㺕ꏸ㶔ۨ㫸죸㶄Ṫ븳ﲙ㴼쑷㵇掔뷤퓟붤캱뷍揜㸣嚙㹗汓뷩隈빥㪟㺳㊎㵝汜㸀뀮㺸♁㸌밂渫벇畛㬅㹀餬㶬燩븾뵃ဨ뷳⢧㹚姗㸰셾㶺斢븛㸄骖븝Ხ㲱�㶰炿빈˟㳭�㺎䥬㺨鍪㹹馑㸃髢릾빰騱㶈㶈줲㹶条㷀䩢㹻譶㲼䣨㹫覆㺇ᇔ뵎⍨븀뭷㰋틙붎䑔㸩㙴븝솶㺣箿뵜㪀㶬ᇌ봰眮㶽봥溴㸶쵵㻀魡봶븚䰜㶍鍾㶯튻뷰ẩ㹒跰㰺㿼뻐㊬벍㠝봳뒩㱭镰㸤笠㴚蠫봧분㳙唾㵑蓡㷎鞺㸫㙷㴈퀓㶫暓믒㷏㺁⟧㺑᱿㱻셅뭉밵붟䘺븘栴㶺軄㺈㵳늴붃炻㹬붛ᯎ븟퇵벂뺃弥붼໚㸅㸋ꍍ뵍㏾㸛訳㷉胯㯉᠓㷖ⅷ㸒貋㶥놎뷊送빰墷벏鞐㴐웮벓冉㷥쉥㶄반㯸곸㲼❐㸻걢㸘뉤㹫ꈬ뭢쭐㷶쀋뤓琨㵑ꯧ뷄ꋔ봖좇㱬ᤘ㶾볭㹙솫㶂䨼㶹㭒벐틡㵅빱㴀嘢붻⯟볓것뵱歮㶛氺㴔㓓뷎빡涨㮉隟㺓㶵䨸㶷␴㹤牘뵜䛄빽ꪮ㶽䪐병ӧ㸝뜜뮯붕᪖뵌쵥㹞莮㷦릹붲ꭰ뷨誯봗怵뱩辍㶄ᕂ㸁붒韷㷫䢿㰮㍵벸ߞ먿䩅㸯䫀븸ป㷵牲븫㸍ቄ㶈᥷븴臌㲿覬봡혝붧뷰糮붕ᶻ벓ꛤ뵻瑕㶉첟㹗�븰櫴㹈�뷙⽁뷦뼪㷯ﶞ㳇甖㹉瞚㵛ᩇ봫잪붑ঈ㷵↪빺Ƙ빉輇뵓寅㰫鍥㺁㷖䁸㶌깐㵁뾫붗䘁뷹鎵㹆첷㷏�뷝➪㴩䵓뷲郿븏ꯧ봡�뷷鸕㴎﹍㶃ᬿ븣몼빒凅븿ヱ㸿䀶㵋ꧡ붺␵㸚㶆騷㹀펆㶕ꐺ㻟떢㸦舉㴭鮱븿狽㷌ច㮿㴋㹴說빖䦻븎뀊븝⢫㴤㣼붟㷥䐻㲓耪㶐㧦븳㋨뮃䄺㳾ᙨ㵀軷㸔쒛㹼謱밀ႌ㲀浃뵕䗑㸛薳㸼偫㹨ᄎ뵓繞붝尅벻㨰뺁柸㹆阽봒䭳㶴儡㸗㷔ῐ붋䩽㺐㤏㴍퉭㹨ဨ㶲輞㸲봜⧜㴩꓇붥ﹾ뺃钅붘⑜㲎䀈빸㵛뵐૵㱕㲍먷븃빫㷯≔먓퐇뷉揓㸌抹㲘蠍㵊圇㶞㹿௻㸶阾부ᥠ뺈㷜봘놁븢攟㶫罺붚왈븁ᤲ븐Ἶ분㋁뮒客븅֯㺂㙂㵁玖㷁맢브ㄻ㶐뙲붻뽄뵶癋봭ꢁ㶦旸㸞錩㸖붙⋳㸆豈㸪妾븍㭖㸅㹚᪇㷚淏㸥⟍븤㸔ᢓ뷻㵿䜟㸨뙘볖﫶㵡㵡붬衤㸐㴸䯔붒�뷰먪㵺泃봺鎔붝篶뵐劗븶⎄㱶㸂欴㷉僄㵾챜뵰�뺪킌㶹浂㶦嫷㴕㝆붉늻㮰쨋㴬簁붯쇖㸬乮뵫b붆髦봱䢩볣ꤱ뻌誌볻禑뺙眇뱃揗㸧쿠뷒蔆㷷븒ꙷ뵴啬㷘͆㶷囉뷦뺁挵볾☝㲏㶁찀㵔⃢븛腾븹隶빚뇁봳蠛뺔ک㷞㶣갅㶮꧰밣誜㷁ﳽ븀䭴뷽ු빝噽븩আ블붵㊁봆혟붴꽜㶿ㄿ㱌칵뵘顝뷋⑞㵔�㹭㇐봉欀뷇㸤ⓐ뺂傖㶃㜫붺녽벧ᒑ뷕勧붨맸뷇�뷥Ṷ㲂䐀㹸벮賢밖᫩뱃㩏㷼㝇밫眮㵿囥㳤븮뮗㸇ઁ㹰짭㳍﫵봠⠧㯶롡㳲㹊찢묏뢮㱇老㴶橈㶩뭣㲦�㵷鮎뫞哲붩ꗶ㹋弦붳资㶒蒜볶㹛㦮볅뷮ꯂ㹒짒㴵䂦㸝᪏㷃﫪㹕ړ비揺뷊㋍뺴㳓㷜ჱ㸊ꇵ븊Ⓩ㹼搤㸺북퇳뷻滘뵯兀붻㶊ￔ㷥൷빛ꬍ㹑㾔븚踑뷤竂뱒ٙ㸤붻㑵붊厬뷞䄌먛렵봻ᴧ㷬蜔빀㲮됳㹊ḉ밿峂뷂橃뱥皠빊죞빓ꥬ빪붃⎮뱰蛫븤仔㬪枤봆끠㶏㳇흢㴤존㰮軐븩뻧빨㲼㳥Ṃ붍〨붤膝붭ᛍ븛㴻Ѣ벗娐븖㒘봎ﲡ벡鯠㸂왻㷳甿뫄➘㲡♂㴦艓벙欺봄莤붃汍벒駿뷀뵿ﱄ㸤版㵑䁪㯓濗붙�뵑刳㵞붋멠㶽血㷙棢㷡琭빻Ѝ㶐瘼븸㠾붙ᓫ㵒좴㷃봙뷌붨�북曑㪡颾㸣묧㺤쟓㸪᳙㲉⣟㰼⾢븯䁄밲콇㲓ণ㶡ꂩ빿ፉ㲑�㴒榁붭狶붦﨩㷈埿붙븎㟭뺏詡벓棛㴟妷뵙籍뵱蚶㸠Ꮙ㶧㸠抗봫됖뷅峝붎捘㸍嵅뷬鍄법嶯㶺䴤빩뱶硥봻숅㵊癃븟�봛ઌ붍⧛보鬇㴐介뷟㒎봯坓뷤ⱷ㸼䱒뵮縜㵩뵙Ø뷟퉾봓ᚈ봿ጃ뺣뫼븏樓㶑⋁븓䰝㷓삛㰈㵪㷘贼뷵쁙뺁�㱶勃㹁輁㷂菪㲂픷봉ꁣ뷌៍㱕㶈ᚽ븸朰빞賞벾굣븗럒붑黔㳵븚鐺븛謶㷥鰨븳瑻㷅뒮뱀㓧㳰븺诿㷱팔봇䉬븳뷺繰㺑㵙㗔봠វ㵔՗㵵ৌ붮ࡉ붅兀뵞僇㱎໕븞묣봣䰃븼텇㷁쩈㷠뵭툱봅怿㷡螸봒㼥븓웓뺄捧㯸湰㵣釟㹵붰剙㱶ꩪ㷫閕㳷⿾뵨㪊붸륷㸒촙뷀뺍㶉뵅ួ봴殝뵎ở뵀㷥胴㶭㸑ᣑ㶛赹㶛蛳㳨ᮧ봫䲲빀଍붯룲볐�빌ᅏ붱ऩ㳱禬㱦ꑺ볇よ㹁졽㶄잠빜ꮯ㷹᧻㬃�㶗␱㲁퍒빍䍂밓案㷝ᛃ㺂禲붧畤뵕쨆빹羜빅봝㜝㸹갥븖譿븘㺙㸶頞뺌䧧빋㓗밼䰶㵀單분㲎慿㶗노㵯ẝ㯆㴵뀩붶岣붓㹰䩻뻈᧍붶Ľ뺍㿳붒鄯붑뼜뷕杒붆㩰벨竛㴿肽빇慥붙罹㱱祍붇䍵㳰�븧禲빇㏇㶚貓㸏虋㸊鎾봞熤봲唙㶬䰟뵴蹹㶜痉븎鶣㺒ᕔ㵹腱㶔丕붮즽㴂늘병㴄ᄩ뷖洚븒ﴈ벓큲㶵컷㷋⪘빉郪볨䗏붯ﺣ㸔檥㷦�㷈봫䝐㲏ﰈ㸅ઞ뺏檭㶷䝲뷿헼㵓햍㷮嫟봁�뵐뼞볛眷부ﾚ㷘३㵉驒뷸⪥봛丗붚㑐㵿᪵뮜ਗ㷢㷝䋕㶞℟㹷쮺㳃ᅒ볜噒㱚춀㰯Ḍ㯸縠붷馴붎쑺볆ꒌ뷚⁹뷻羚㱀츱븈벅嵃뷄縸㷚썛비愭㴤؊빒뎣붪ꊤ뷮镟붥휪㹇睊븉歾븬빭퉈벘㬷붦ꉅ㴭㽫븙㤞㳫㶾㸁⪲뵨Ǐ㶧섦벺䛣㺝㶴�봳遝븛觴붫蔍㱉ꑇ붯甚㸀볾븾儞㶅菤붂㵣홃㷑趋㵅殽볓Ӡ㶖냡벰ﲉ불蛌볠ᦞ뺁ྴ㴣ꄡ붌渏㸕ਹ㷱䡿뷱⻳㷢﯏뷿᰺㩌蕤뵚븮뺟栧붲᱿㸎଺㷊ᙋ뷰唩뱩匁빔셝븛가㵺㷚顳㲸毇㳢ᨹ㵸ퟒ㯅蛸붍沈븵⇑뺪ﹹ㳇ȱ㷇쬛뭋婩벵揉㴡ꔱ뺅葋빻諬볯⓻㸃൝뺋汾㸔걑㶆᪕뷆법갊㳑䡨븝䧓㵈ᄙ뷀㰹븟뺅鰽㲳㏀㴲�㱘㴳칩붚彗뷅㬇봽뎭봶㷝㸬홰㰣뵒㴥⬟㴂㻵㷅ᚆ㹃˩부앟봒ꤖ㶫⇇붚貽㶫쏗밠롭벴뮂㰹ྈ벱ᳰ붃椝㹶綍봙髑볕䫲뷫Ц븈㹟澘별ॼ㳦姶븑㾤뱀빰縣봧Ⅻ붸涕㸒㿜붩颪㶖䰱뵏㟩㯆皋뺁老뺂迯㵅낥볋鶷뷼찕봼ᇩ뵳踲북炙뵣᱇㱠ⳓ㵌襥㲌�봫즈㶐ﺝ㸅᱾㲔뎗㷐틿㴖ꢠ뷇ፅ㯝뱎ᙂ붮允봷藫㸗炊㵘쨢㶄涰붫⠴뵾ه붥≤㷫㞡빜瘟붕१븒�봦饺븲⎌봕餫㸢㜏㰷杯뷤꽔붅ፖ볃쟃㶁돢㹉垚븯鋶뮐떈㲨鯯㺄ᆉ㸱靈㳵⸅붔㊸믵﹆븒場빈賠㷃㖢㸜巁붡軠븄꾸㶻�븲菾빸ﴮ빘蛝봎㇠㷯밵㴕︯먷궶먀劬뵏풑㴫뱳﯏㶸㯛㯽皿㸗豜븎嚴㶝天㴶㣇봙ꄛ뷬닑붕쯤㴧汹붚㳲븘⶞㲆꼉㷥銇봏໎백핫봂㺉뵀붂怍뷍ᩄ㴖룅㸞홯밃죭㸎暆밄閊븳ﹺ㶈䇭㺎ꖤ븬ᵡ빇휆븟�㳈㶝잻㺢뛉뷦時먃輻봇㫚뺠뷼붑埐㶨ୗ빡ᅏ㵏㬑㶌╓㰚餏㸗焯발⿱빘퇢㹃侌㳢빐㳟鷎뱚轂뱵式벐梅㴕鄛붙㶈匚㶱ᕋ㸪쳗붛㾇㵹᫱뱪࢈븏㭧兑뷲聳붗襨봾院봪穞봢퓉믨蕲㶱᩵뷫๻㩩篍㸉ꕅ㸈鑡㲜্밐땗뷠檔뷅㵣붏ﱂ법⺀빜Ⲱ빨ࡘ뺯砬븯䊬㷐ᖎ뷇나㹊墝㷄送봭븯믖㶸े㴶哣붐鯖봏�블ꚢ묢�㸮䛟㸑῱뷋ꎆ㵐ډ㲼有㶣꺑㸜쯔㶼ٱ㬒嬨빠娩볊콺㴁ｖ㯳ܘ㹀㹴얍㶃童백帢븅ꁬ븠ᣯ㶻朰벸ڡ빰䟮븐침㹍킠빓㷏冃볼诈뺵歂㶍痎붨놺㴚䜑㸜䒅믟뤻벯ॏ㴚糗㸐䷿㶻鴭뷺�㵐큕㸵鰴븃빆妅㶨샓㺑੃뵤㏹벾瀒㴮ꁢ㷈辰뺢柵붊ᾘ뷍酫㳚Ｋ붯빓븵붘ᗄ뷠ᯰ븞㿻㴢빕앿㸕穦븮갔빮￳봜໹㷂꘎븨ӗ㷜顺㰦睹붇⪚㸲ꖺ봝ી㳈╒㳷뀿붵䈍㵌 ",2019-05-14T03:55:34Z,gpt-2,https://github.com/openai/gpt-2/issues/132 131,443031373,Release raw lambada dataset,"Is it possible to release the Lambada dataset used to generate accuracy numbers in Table 3 of the paper? This would make it easier to do comparisons with other models :) ",2019-05-11T20:03:04Z,gpt-2,https://github.com/openai/gpt-2/issues/131 130,442909074,Added Console Commands To GPT-2 Interactive Module,"I cleaned up the loops a bit since the second one was not needed and instead could be fixed with a continue statement. I also added console commands to better user control. The end goal is for the user to eventually be able to change the launch parameters inside the application. I also noticed it doesn't always exit on Control-C or Control-Z so i created a kill command (named #kill) to cleanly close the process.",2019-05-10T21:58:39Z,gpt-2,https://github.com/openai/gpt-2/pull/130 129,441893748, allow models to be in a separate folder via models_dir argument," allow models to be in a separate folder via models_dir argument default values unchanged",2019-05-08T19:28:14Z,gpt-2,https://github.com/openai/gpt-2/pull/129 127,438865868,Can't open download_model.sh in colab,"When I try to download the model I see the error: > sh: 0: Can't open download_model.sh What am I doing wrong? I used this code in colab: ",2019-04-30T16:26:12Z,gpt-2,https://github.com/openai/gpt-2/issues/127 125,438036639,Interested in programming length recognition.,"Hello! I was wondering if you guys would be interested in collaborating on programming in a stop-word or length recognition to adjust the text output. (like dicing it instead of 3-4 paragraphs every time.) Was this considered, and I was wondering if there was any harm in a PR to adjust the values for each text input. Like saying: Input: The cat is blue output: and brown and grey. Instead of: Input: The cat is blue output: and brown and grey... 5 paragraphs.",2019-04-28T09:38:16Z,gpt-2,https://github.com/openai/gpt-2/issues/125 124,435955578,Create download_model.sh,add curl shell version for download model files.,2019-04-23T01:18:40Z,gpt-2,https://github.com/openai/gpt-2/pull/124 123,435428218,Issue when i try to train using my own dataset of bigger size,"I am trying to train the GPT-2 using my own dataset which is substantially larger size. However, the process gets stopped at a point after displaying below details - 2019-04-20 18 42.851036: I CPU Frequency: 2300000000 Hz 2019-04-20 18 42.851356: I XLA service 0x1b06260 executing computations on platform Host. Devices: 2019-04-20 18 42.851408: I StreamExecutor device (0): , 2019-04-20 18 43.106241: I successful NUMA node read from SysFS had negative value (-1), but there must be at least one NUMA node, so returning NUMA node zero 2019-04-20 18 43.106797: I XLA service 0x1b05e40 executing computations on platform CUDA. Devices: 2019-04-20 18 43.106854: I StreamExecutor device (0): Tesla T4, Compute Capability 7.5 2019-04-20 18 43.107280: I Found device 0 with properties: name: Tesla T4 major: 7 minor: 5 memoryClockRate(GHz): 1.59 pciBusID: 0000 04.0 totalMemory: 14.73GiB freeMemory: 14.60GiB 2019-04-20 18 43.107310: I Adding visible gpu devices: 0 2019-04-20 18 43.920575: I Device interconnect StreamExecutor with strength 1 edge matrix: 2019-04-20 18 43.920645: I 0 2019-04-20 18 43.920654: I 0: N 2019-04-20 18 43.920962: I Created TensorFlow device GPU:0 with 14115 MB memory) -> physical GPU (device: 0, name: Tesla T4, pci bus id: 0000 04.0, compute capability: 7.5) WARNING From colocate_with (from tensorflow.python.framework.ops) is deprecated and will be removed in a future version. Instructions for updating: Colocations handled automatically by placer. WARNING From to_float (from tensorflow.python.ops.math_ops) is deprecated and will be removed in a future version. Instructions for updating: Use tf.cast instead. WARNING From multinomial (from tensorflow.python.ops.random_ops) is deprecated and will be removed in a future version. Instructions for updating: Use tf.random.categorical instead. WARNING From to_int32 (from tensorflow.python.ops.math_ops) is deprecated and will be removed in a future version. Instructions for updating: Use tf.cast instead. WARNING From div (from tensorflow.python.ops.math_ops) is deprecated and will be removed in a future version. Instructions for updating: Deprecated in favor of operator or tf.math.divide. Loading checkpoint WARNING From checkpoint_exists (from tensorflow.python.training.checkpoint_management) is deprecated and will be removed in a future version. Instructions for updating: Use standard file APIs to check for files with this prefix. Loading dataset... 0% [00:00 fire.Fire(interact_model) File line 127, in Fire component_trace = _Fire(component, args, context, name) File line 366, in _Fire component, remaining_args) File line 542, in _CallCallable result = fn(*varargs, **kwargs) File line 42, in interact_model enc = encoder.get_encoder(model_name) File line 110, in get_encoder encoder = json.load(f) File line 299, in load parse_constant=parse_constant, object_pairs_hook=object_pairs_hook, **kw) File line 354, in loads return _default_decoder.decode(s) File line 339, in decode obj, end = self.raw_decode(s, idx=_w(s, 0).end()) File line 357, in raw_decode raise JSONDecodeError(""Expecting value"", s, err.value) from None json.decoder.JSONDecodeError: Expecting value: line 1 column 1 (char 0) (gptvenv) User MINGW64 help much appreciated",2019-03-07T14:28:52Z,gpt-2,https://github.com/openai/gpt-2/issues/94 93,417903541,Cool and fast model,Guys thanks for releasing this model. It is very fast compared with LSTM models. Cudos!,2019-03-06T16:44:24Z,gpt-2,https://github.com/openai/gpt-2/issues/93 92,417660539,Train model on small data set,"Great work releasing this! Is it possible to get the data used to train the 117M model?",2019-03-06T07:29:06Z,gpt-2,https://github.com/openai/gpt-2/issues/92 91,417581409,Extracting representations from the LM,"Hello, Is it possible to extract features from the language model using the current version of the code? For example, given a string such as ""the cow goes moo"", can we extract a vector for use in downstream NLP tasks? Thanks",2019-03-06T01:36:37Z,gpt-2,https://github.com/openai/gpt-2/issues/91 90,417541520,"error downloading model using ""python download_model.py 117M""","got this permission error: Access
Anonymous caller does not have storage.objects.get access to ",2019-03-05T22:59:39Z,gpt-2,https://github.com/openai/gpt-2/issues/90 89,416731557,Python download script,Python download script with tqdm progress bars.,2019-03-04T10:25:23Z,gpt-2,https://github.com/openai/gpt-2/pull/89 87,415594980,clarification,"Hi, can you just try to clarify in human readable text what can be done with this project? I find the usage explanation a bit shallow. Basically can I or cant I use this model to start from sample text to generate text accordinglyas a POC ? if so, can you describe more accurately how to actually do this? many tnx",2019-02-28T12:15:35Z,gpt-2,https://github.com/openai/gpt-2/issues/87 86,415512197,Is it possible to edit the code based on this project to train from scratch?,"Is it possible? To achieve maybe 90% of the experiment result in the paper. Thank you very much! ",2019-02-28T08:55:08Z,gpt-2,https://github.com/openai/gpt-2/issues/86 85,415364482,"Problem with re.compile in encoder.py, Python 3.6+ x64","This is from 9d1e704fec1726d5e051e98336b655ae8ce49242: When I try it in Python 3.6 or 3.7 x64 on either Windows or Ubuntu, I get Works fine in Python 2.7",2019-02-27T22:52:11Z,gpt-2,https://github.com/openai/gpt-2/issues/85 84,415243818,How can I train your model on my own text corpus?,"I know that you trained your model on a corpus of news texts. I would like to train it on my own set of documents. I'm not looking for precision, rather for poetry. Could you please tell me how I could create my own training set and reactivate this model?",2019-02-27T17:35:25Z,gpt-2,https://github.com/openai/gpt-2/issues/84 83,415217989,Windows installation,"Hi. Found this software through a game. This could help a lot with the roleplay posts. A game changer! Unfortunately, total noob here. So, I was wondering if someone could guide me through a detailed windows 8 installation?",2019-02-27T16:39:13Z,gpt-2,https://github.com/openai/gpt-2/issues/83 82,415003766,OutOfRangeError," OutOfRangeError (see above for traceback): Read fewer bytes than requested [[node (defined at F 45) = RestoreV2[dtypes=[DT_FLOAT, DT_FLOAT, DT_FLOAT, DT_FLOAT, DT_FLOAT, ..., DT_FLOAT, DT_FLOAT, DT_FLOAT, DT_FLOAT, DT_FLOAT], ",2019-02-27T08:36:15Z,gpt-2,https://github.com/openai/gpt-2/issues/82 81,414955434,Add documentation for help flags,"Add documentation for help flags in unconditional and conditional sample generators and modify README with how to use help. Also in interactive_conditional_samples.py, change default for batch_size to 1 instead of None to make it more readable, this does not impact execution in any way as it was already being performed in the code earlier: ` if batch_size is None: batch_size = 1 `",2019-02-27T05:46:42Z,gpt-2,https://github.com/openai/gpt-2/pull/81 80,414901459, Why \\u0120 (Ġ) is in so many pairs?,"It seems obvious that it's used as a ""blank byte"" but I don't understand how this ""blank byte"" ends up being represented by this ""Ġ"". Can someone explain this? ",2019-02-27T01:30:49Z,gpt-2,https://github.com/openai/gpt-2/issues/80 79,414510206,How to cite the paper,How should the paper be cited? I could not find bibtex entries on the internet. Thank you,2019-02-26T09:25:31Z,gpt-2,https://github.com/openai/gpt-2/issues/79 78,414302881,Question about reported perplexities,"I've been reading your paper, interesting work. I have a question about how you compute perplexities, especially over datasets that are already tokenized (e.g., wikitext-103). I understand that your encoding can assign probabilities to any string, but I'd expect the LM to do poorly when fed pre-tokenized input. For example, the tokenized wikitext-103 input looks like How do you report perplexity in this case?",2019-02-25T20:55:45Z,gpt-2,https://github.com/openai/gpt-2/issues/78 77,413869759,WIP: Add chat example,"I'm experimenting with the 117M model, trying to build a simple interactive chat example. Results are _acceptable_: However, I have a feeling my code has an error because when I try generating text with the other scripts ( and ), the output quality seems subjectively better. The approach I've taken is to keep the model in memory, and pass the whole conversation at each point, to generate the bot's reply. To try out the code on this branch, run: Some open questions in case anyone (OpenAI team or community) wants to chime in: - Is it correct to use the same model instance for the whole ""chat session""? Or should I restore from a checkpoint before generating each new line? - Is it correct to prompt the model with the whole conversation at each stage, or should I only send the new dialogue lines? - Any other ideas on how to improve output quality? (tweaking , ?) ",2019-02-24T21:21:20Z,gpt-2,https://github.com/openai/gpt-2/pull/77 76,413771052,Minor: tee stdout to /tmp/samples in readme examples,"Example in readme generates a file. To avoid checking this output into source control, add it to .",2019-02-24T02:04:44Z,gpt-2,https://github.com/openai/gpt-2/pull/76 75,413727635,Rephrasing a sentence,"Can this model rephrase a given sentence or even paraphrase? At the prompt, I tried to give a sample such as: Symptoms of influenza include fever and nasal congestion = A stuffy nose and elevated temperature are signs you may have the flu. Giraffes like Acacia leaves and hay, and they can consume 75 pounds of food a day. = But the sample returned something completely different.",2019-02-23T19:09:48Z,gpt-2,https://github.com/openai/gpt-2/issues/75 74,413644486,Port model download script to python,,2019-02-23T02:24:34Z,gpt-2,https://github.com/openai/gpt-2/issues/74 73,413602039,Getting all exclamation marks on Windows,"Once model is downloaded, I tried the interactive python and does not matter what I write, the output is SAMPLE 1 === !!!!!!!!!!!!!!!!!!!!!!!!!!!! And 3 or 4 lines or exclamation marks. This happens also with samples.",2019-02-22T22:20:43Z,gpt-2,https://github.com/openai/gpt-2/issues/73 72,413563721,No module named 'fire' issue ,"Hi, I wanted to try to generate some text but I have an issue while trying to use commands for text generation, the console tells me that no module named fire has been found. I have installed fire 0.1.3 but there is still an error tristan python3 --top_k=1 Traceback (most recent call last): File line 3, in import fire ModuleNotFoundError: No module named 'fire' any ideas what i could have done wrong ? I'm using windows 10 and the ubuntu console Thank you ",2019-02-22T20:20:40Z,gpt-2,https://github.com/openai/gpt-2/issues/72 71,412921864,Clarify who has access to the model or code,"In your Introducing OpenAI post you mentioned collaboration with institutions and companies. Could you clarify whether any of those or your sponsors, has access to the model the code of GPT-2? Thank you!",2019-02-21T13:22:50Z,gpt-2,https://github.com/openai/gpt-2/issues/71 70,412906208,add .gitattributes file to ensure files copied to docker container ha…,"On windows machines git defaults to cloning with windows style line breaks. This breaks the *.sh file, so we should ensure that all files copied to the docker container keep their unix style line endings.",2019-02-21T12:43:18Z,gpt-2,https://github.com/openai/gpt-2/pull/70 69,412794551,Minor: update readme,Add note about setting env var for running examples,2019-02-21T08:02:52Z,gpt-2,https://github.com/openai/gpt-2/pull/69 68,412694928,Model code is not same as paper describled ,"After read the model source code, I notice that the code uses a “past” context for multi head attention. Could you tell us more about how this “past” context is organized into the training pipeline? I guess this “past” context maybe very critical for the Gpt-2 model!",2019-02-21T00:37:50Z,gpt-2,https://github.com/openai/gpt-2/issues/68 67,412648096,UnicodeEncodeError: 'ascii' codec can't encode character,"I've reproduced this Python error 3 times already. First I tried out the unconditional samples: That didn't work, but I got a few simple interactive conditional samples working until I entered this model prompt: Thought it didn't like the or characters so I removed those but still observed the same error: I'm running GPT-2 on 99af6d7 in a container built from the image running on a single core Nvidia Tesla T4 node provisioned with Google Kubernetes Engine.",2019-02-20T21:58:04Z,gpt-2,https://github.com/openai/gpt-2/issues/67 66,412620417,fixed seed arg to ensure reproducibility in interactive_conditional_samples.py,"Before, setting up flag with a specific value did not make any changes in the results, as noted in #58 The reason was that the random seed needed to be within the scope (line 34 in updated version). This change allows for reproducibility of results every time is ran, is passed with the same specific int value assigned, and the same prompt is input. See example below how I run with and prompt ""Well, I just made my first pull request to an open source project."" Then I interrupt my session, and re-run with . I prompt again ""Well, I just made my first pull request to an open source project."" and obtain the same results. Note: It DOES NOT reproduce the same output if passing the same prompt two or more times within the same session. You must interrupt it and run the file again. ",2019-02-20T20:46:17Z,gpt-2,https://github.com/openai/gpt-2/pull/66 65,412592291,fixed seed arg to ensure reproducibility in generate_unconditional_samples.py ,"Before, setting up flag with a value did not make any changes in the results, as noted in #58 The reason was that the random seed needed to be within the scope (line 30 in updated version). I updated the file to reflect the change. Now setting flag to a specific value ensures you get the same results everytime you run the code: I suspect same issue occurs for interactive_conditional_samples.py ",2019-02-20T19:33:25Z,gpt-2,https://github.com/openai/gpt-2/pull/65 64,412414375,After These Messages We'll Be Right Back,"Very minor and semi-amusing, but noticed gpt-2 with the 117M model will generate ""Advertisements"" as a string in isolation on a line, so I'm guessing the training data has that somewhere that could be scrubbed whenever is convenient. The implications of AIs accidentally learning to value commercial interruption is left as an exercise for the reader :)",2019-02-20T13:03:37Z,gpt-2,https://github.com/openai/gpt-2/issues/64 63,412357725,Request: Release Bloom Filters for WebText (or provide other method to check a given text is in WebText),"Hi, Thank you for releasing your pretrained model and saving us training time. I am currently exploring possible applications, but I ran into a problem that might also annoy many researchers trying to use your model. AFAIK, you have not released the WebText corpus (although I know this is currently discussed in issue #24). This is fine by me, except for one aspect: it makes it impossible for me to know if my test data is somehow included in WebText. Which, in turns, makes it impossible for me to tell if any improvement I am getting is due to the quality of GPT or the fact that the pretrained model has already seen my test data. If you do not plan to release WebText in the very near future, I was thinking you could release the bloom filters you describe in your technical paper (code + filled filters). This would allow us to evaluate the proportion of 8-grams in our test data that is also in WebText. Would this be possible? Thank you. ",2019-02-20T10:39:49Z,gpt-2,https://github.com/openai/gpt-2/issues/63 62,412268487,Generate conditional output based on input keywords,"I think it would be nice if it would be possible to generate an output sentence based on input keywords, where the lenght of the sentence, or the words ratio could be fine tuned. In this way we could create a higher system (motive) which could use this model to generate a guided conversation.",2019-02-20T06:39:18Z,gpt-2,https://github.com/openai/gpt-2/issues/62 61,412234797,encoder.json download error,The download_model.sh script can't download encoder.json from online repository. help me! thanks in advance.,2019-02-20T04:02:57Z,gpt-2,https://github.com/openai/gpt-2/issues/61 60,412232856,Installation question,Is the Docker installation an alternative to Native installation or are both needed?,2019-02-20T03:53:15Z,gpt-2,https://github.com/openai/gpt-2/issues/60 59,412224061,Requirements on Arch Linux ,"These who use archlinux should not follow the instructions but try to After uninstall everything that will be install by pacman. I did piping install the sample did nothing and exit",2019-02-20T03:12:19Z,gpt-2,https://github.com/openai/gpt-2/issues/59 58,412211097,Fixing seed and/or setting top_k to 1 don't make sampling deterministic,,2019-02-20T02:12:32Z,gpt-2,https://github.com/openai/gpt-2/issues/58 57,412103096,sh doesn't do anything on Windows 10,"Hello, what operating system do the instructions apply to? sh doesn't do anything on Windows 10. How would I install this on Win10? Also, is the first step to clone the repo? The instructions don't seem to make sense otherwise. Thanks.",2019-02-19T20:10:26Z,gpt-2,https://github.com/openai/gpt-2/issues/57 56,412073443,fixing requirements error,"Was getting this when trying to install the requirements... ` Could not find a version that satisfies the requirement tensorflow>=1.2 (from -r requirements.txt (line 2)) (from versions: 1.12.0rc1, 1.12.0rc2) No matching distribution found for tensorflow>=1.2 (from -r requirements.txt (line 2)) ` Adding an 'r' in in front of the tensorflow version worked for me",2019-02-19T18:54:07Z,gpt-2,https://github.com/openai/gpt-2/pull/56 55,411994886,Which python should be used? 3.7.2 catches exception,"Using 3.7.2 on any tutorial cmd > python3 --top_k 40 > RuntimeWarning: compiletime version 3.6 of module 'tensorflow.python.framework.fast_tensor_util' does not match runtime version 3.7 > return f(*args, **kwds) > 2019-02-19 17 24.116332: I Your CPU supports instructions that this TensorFlow binary was not compiled to use: AVX2 FMA > Traceback (most recent call last): > File line 68, in > fire.Fire(interact_model) > File line 127, in Fire > component_trace = _Fire(component, args, context, name) > File line 366, in _Fire > component, remaining_args) > File line 542, in _CallCallable > result = fn(*varargs, **kwargs) > File line 42, in interact_model > temperature=temperature, top_k=top_k > File line 76, in sample_sequence > back_prop=False, > File line 3291, in while_loop > return_same_structure) > File line 3004, in BuildLoop > pred, body, original_loop_vars, loop_vars, shape_invariants) > File line 2924, in _BuildLoop > c = ops.convert_to_tensor(pred(*packed_vars)) > File line 3259, in > math_ops.logical_and(i < maximum_iterations, orig_cond(*lv))) > File line 4365, in logical_and > ""LogicalAnd"", x=x, y=y, name=name) > File line 510, in _apply_op_helper > preferred_dtype=default_dtype) > File line 1146, in internal_convert_to_tensor > ret = conversion_func(value, dtype=dtype, name=name, as_ref=as_ref) > File line 229, in _constant_tensor_conversion_function > return constant(v, dtype=dtype, name=name) > File line 208, in constant > value, dtype=dtype, shape=shape, verify_shape=verify_shape)) > File line 542, in make_tensor_proto > append_fn(tensor_proto, proto_values) > File line 134, in tensorflow.python.framework.fast_tensor_util.AppendBoolArrayToTensorProto > File line 547, in asscalar > return a.item() > UnboundLocalError: local variable 'a' referenced before assignmen",2019-02-19T15:54:49Z,gpt-2,https://github.com/openai/gpt-2/issues/55 54,411979958,Integrate a training feature to pass JSON markup convo data ,"Other alternatives to this software have allowed such a system to exist problem free, cakechat by Replika.ai for example; allows and encourages people to check out the 40GB reddit corpus you guys build GPT-2 from. I was getting highly similar results with altering their token model and marking up the JSON with emotion compared to what I am seeing with your currently released solution. **Examples of things more dangerous than your software that turned out safer than imagined...:** Example #1: Fusor.net provides detailed instructions on how to build a nuclear fusion reactor, 12 year olds in south america have even made them. Yet Bogota hasn't become Chernobyl. Example #2: 3D printed weapons have been around for years and actually are far more safe less accessible than normal weapons. Nobody has ever been harmed by one. Example #3: The early days of Bitcoin, although they were accompanied by some awful things - didn't do anything as bad as what the Sinaloa cartel managed under El Chapo daily for decades unchecked. I'll protest this until I get my own copy of this software. It's a human right for you guys to release your full, open and honest production software instead of choose profit and proprietary. ",2019-02-19T15:25:33Z,gpt-2,https://github.com/openai/gpt-2/issues/54 53,411902339,[PROPOSAL] Allow remote testing of larger models ,"Your paper and this implementation could mark a significant evolution in UMLs. Especially (for what I'm concerned) in CoQa & translation. It would be nice to kick the tires of the full model even remotely, without having direct access to it. Would you consider setting up -say- a MQTT server with some channels to be able to interact with a couple of functionalities? Abuse of this service could be moderated by simply pruning out suspicious requests by sending authorization to validated mail. --R ",2019-02-19T12:35:27Z,gpt-2,https://github.com/openai/gpt-2/issues/53 52,411871922,Charmap error," File line 55, in fire.Fire(sample_model) File line 127, in Fire component_trace = _Fire(component, args, context, name) File line 366, in _Fire component, remaining_args) File line 542, in _CallCallable result = fn(*varargs, **kwargs) File line 52, in sample_model print(text) File line 19, in encode return codecs.charmap_encode(input,self.errors,encoding_table)[0] UnicodeEncodeError: 'charmap' codec can't encode character in position 1410: character maps to ",2019-02-19T11:15:40Z,gpt-2,https://github.com/openai/gpt-2/issues/52 51,411816654,Is it fine not use Softmax after forwarding?," I found something different from the paper. It may be mabe some error when function because there will be value less than zero I try with softmax but it does not shows good result for text-generator.. This is result when use softmax function Not using softmax is your original meant? I added softmax before multinomial sampling when implement to Pytorch to avoid encountering probability entry < 0 ",2019-02-19T09:07:24Z,gpt-2,https://github.com/openai/gpt-2/issues/51 50,411728596,Can you make site/blog that publish content from GPT ?,It will intersting and entertainment everyone (maybe ??),2019-02-19T03:29:25Z,gpt-2,https://github.com/openai/gpt-2/issues/50 49,411667109,Wrong Scaling in multiattention?,"Hello, Not quite sure if this is intended or not, You are scaling by rsqrt(dim value) where as it is usual to scale by rsqrt( dim key ). Is this a typo or does it work better? Thanks",2019-02-18T22:20:22Z,gpt-2,https://github.com/openai/gpt-2/issues/49 48,411473816,SentencePiece,"Why we not used SentencePiece for BPE ( Can you provide SentencePiece unigram model from your dataset?",2019-02-18T13:23:36Z,gpt-2,https://github.com/openai/gpt-2/issues/48 47,411363541,Implementation about gpt-2-pytorch here," This is my code about Implementation of gpt-2-pytorch But in first Commit about BackBone Code, This is not running because model can not load *.ckpt file in Pytorch. Also, I wait license on original gpt-2 repository !! I hope it's good for pythorch developers. PS) Thanks for for allowing",2019-02-18T08:51:44Z,gpt-2,https://github.com/openai/gpt-2/issues/47 46,411228612,\\\\ issue in Windows,"When running Getting this issue ? any -idea is it due to ? FileNotFoundError: [Errno 2] No such file or directory: ",2019-02-17T20:03:37Z,gpt-2,https://github.com/openai/gpt-2/issues/46 45,411195211,How to download on windows?,"I'm using Windows 10, 64x. Could someone please explain to a novice how to download this? 1) Following the instructions, I downloaded gsutil and started a new configuration. Run: sh download_model.sh 117M But receive error: download_model.sh: download_model.sh: No such file or directory 2) Tried both: gsutil cp -r dir gsutil cp -r dir But receive error: AccessDeniedException: 403 [my gmail] does not have storage.objects.list access to gpt-2. 3) Tried the solution from loretoparisi at But I think I'm doing something wrong here. I downloaded Curl and Grep, then created a .bat file in Notepad++ with his script. Executed the file, but it only opens and closes. Any help would be greatly appreciated. Thanks, Pete",2019-02-17T14:48:18Z,gpt-2,https://github.com/openai/gpt-2/issues/45 44,411171500,I dont know why 'while_loop' only loop one time?,"hello. I dont know why 'tf.while_loop' only loop one time Nevertheless return value of is always invarient on False. ",2019-02-17T10:17:33Z,gpt-2,https://github.com/openai/gpt-2/issues/44 43,411147153,模型依然没法下载," 0 0 0 0 0 0 0 0 -- -- 0 15 -- -- 0curl: (7) Failed to connect to drive.google.com port 443: Operation timed out 你好，我们（中国）这里还是没法下载。真心请把模型文件放到一个网盘。或者是否可以直接把模型发送到我的邮箱：1224070840 感谢感谢啦。",2019-02-17T04:23:47Z,gpt-2,https://github.com/openai/gpt-2/issues/43 42,411144236,Syntax error in sh_download_model.sh,"When running ""sh download_model.sh 117M"", I am told that there is a syntax error on line 14: 'ownload_model.sh: line 14: syntax error near unexpected token `do download_model.sh: line 14: `for filename in checkpoint encoder.json hparams.jso' model.ckpt.data-00000-of-00001 model.ckpt.index model.ckpt.meta vocab.bpe; do Unfortunately, I don't know how to write shell scripts and can't troubleshoot this myself, but I don't see this error reported anywhere else.",2019-02-17T03:42:25Z,gpt-2,https://github.com/openai/gpt-2/issues/42 41,411099905,Better example parameters for conditional sample command,"This PR adds better initial parameters to the example conditional sample generation command in the docs. The results are pretty poor in the interactive script with the default settings. Now, you'll get better results if you run the interactive samples using the example command from the README.",2019-02-16T18:39:17Z,gpt-2,https://github.com/openai/gpt-2/pull/41 40,411099491,Template to use in Google Colab and avoid settings issues,There can be problems preparing the dependencies and python 2 and 3 requirements. This Google Colab template helps to visualize the examples in an easy manner.,2019-02-16T18:34:23Z,gpt-2,https://github.com/openai/gpt-2/pull/40 39,411083098,Fetch model files using curl directly from GCS,"Since each publicly accessible GCS bucket is reachable via , there is no need for gsutil to be present for downloading files.",2019-02-16T15:34:57Z,gpt-2,https://github.com/openai/gpt-2/pull/39 37,411078552,Is it fine to implemente gpt2 on Pytorch?,"Thanks for the awesome work! 👍 Can I try to implement gpt2 on Pytorch? Of course, I'll write down the sources of the papers and codes.",2019-02-16T14:38:53Z,gpt-2,https://github.com/openai/gpt-2/issues/37 36,411078492,模型下载失败,"你好。 我是python3的环境，可是gsutil requires python 2.7。除了通过gsutil，请问有没有其他的下载链接方式呢？谢谢。如果可以，是否可以发送到我邮箱：1224070840@qq.com。谢谢",2019-02-16T14:38:14Z,gpt-2,https://github.com/openai/gpt-2/issues/36 35,411078091,"Markers of ""end of text"" appear when generating samples","When generating samples interactively sometimes an ""end of text"" marker is included on the text. Is this related to the sanitization of the training data?",2019-02-16T14:33:48Z,gpt-2,https://github.com/openai/gpt-2/issues/35 34,411072092,My CPU doesn't support Tensorflow AVX instructions ,"I was able to install all the requirements. However while generating samples, getting the following error. I have an Intel i3 First gen Processor and running Ubuntu 18. _2019-02-16 03 49.453982: F The TensorFlow library was compiled to use AVX instructions, but these aren't available on your machine. Aborted (core dumped)_ I then installed Tensorflow 1.5 (pip3 install tensorflow==1.5). The sample was generated, however another warning popped up as shown below. Will this affect the quality? Do I need to compile TensorFlow on my system? _2019-02-16 03 19.785441: I Your CPU supports instructions that this TensorFlow binary was not compiled to use: SSE4.1 SSE4.2_ ",2019-02-16T13:17:22Z,gpt-2,https://github.com/openai/gpt-2/issues/34 33,411071420,Abstractive summarization,How can I create an abstractive summary of a text document?,2019-02-16T13:08:52Z,gpt-2,https://github.com/openai/gpt-2/issues/33 32,411066093,Please close it.,,2019-02-16T11:57:49Z,gpt-2,https://github.com/openai/gpt-2/issues/32 31,411044864,Is there a plan to open a Chinese model?,,2019-02-16T08:07:00Z,gpt-2,https://github.com/openai/gpt-2/issues/31 30,410933477,vocab.bpe contains an invalid unicode char and can't be read in Windows,"The reader causes an exception on Windows 10 when reading the vocab.bpe file. I have tried with Unix endlines and Windows endlines. The byte that fails has a value of 0x81 and causes an exception when running the interactive mode.",2019-02-15T20:35:24Z,gpt-2,https://github.com/openai/gpt-2/issues/30 29,410928481,Powershell script to do the same than the download_model.sh does,"I have had to manually download the model in Windows10, because even when using bash, the path is not exported. I can contribute a powershell script to download the model without using bash, using powershell. I don't see any reason of this being bad. Would that be desirable? ",2019-02-15T20:19:45Z,gpt-2,https://github.com/openai/gpt-2/issues/29 28,410906660,Update encoder.py to work on windows,"This fixes ` File line 112, in get_encoder bpe_data = f.read() File line 23, in decode return codecs.charmap_decode(input,self.errors,decoding_table)[0] UnicodeDecodeError: 'charmap' codec can't decode byte 0x81 in position 66951: character maps to `",2019-02-15T19:14:36Z,gpt-2,https://github.com/openai/gpt-2/pull/28 27,410884119,Sampling code flags descriptions (support for --help?),"Is there a list of the flags for both conditional and unconditional models with their definitions? (I looked in the blog and paper and couldn't find any mention.) In particular, for reproducibility purposes, it'd be great to know the definition of and and how choosing different values for these affect the results. Thanks!",2019-02-15T18:08:05Z,gpt-2,https://github.com/openai/gpt-2/issues/27 25,410768970,Try in Colaboratory,"For a quick run: ",2019-02-15T13:26:35Z,gpt-2,https://github.com/openai/gpt-2/issues/25 24,410765001,Any plans to release WebText corpus?,"I've seen #16 and appreciate the valid concerns raised about releasing the model, but the WebText corpus could be a tremendous help to general research if you were able to release it. Are there plans to do so? I did wonder if this might simply enable people to recreate the unreleased GPT-2 but presumably this is no trivial matter, needing expertise and thus deterring the causal mischief maker! Anyway, whatever you end up doing, I wanted to thank you for what you have released already which is really interesting 🙂",2019-02-15T13:16:23Z,gpt-2,https://github.com/openai/gpt-2/issues/24 23,410732941,AccessDeniedException: 403 xxx@xxx.com does not have storage.objects.list access to gpt-2.,"Hello! I am unable to download the model. When doing: I get this error: . Any advice? Greetings!",2019-02-15T11:46:45Z,gpt-2,https://github.com/openai/gpt-2/issues/23 22,410720562,Can't install sh_download_model.sh,"Noob here (linguist, with rudimentary knowledge of computers) I've installed the gcloud sdk but I can't get the command: _sh download_model.sh 117M_ to run. I get: 'sh' is not recognized as an internal or external command. Any help would be greatly appreciated.",2019-02-15T11:13:41Z,gpt-2,https://github.com/openai/gpt-2/issues/22 21,410714732,removed gsutil dependency,I have removed the dependency using and google drive. This approach is well-known and used in several frameworks that need to download large models files (like in FastText),2019-02-15T10:59:19Z,gpt-2,https://github.com/openai/gpt-2/pull/21 20,410687964,Translation task,"What was the format for translation task? Do you provide sequence of pairs delimited by new lines, e.g. ""sentence1 = translation_of_sentence1 sentence2 = translation_of_sentence2 ... testing_sentence = ""? Does the training dataset consist of similar format translations?",2019-02-15T09:56:06Z,gpt-2,https://github.com/openai/gpt-2/issues/20 19,410618504,how to train it ,how to train it ,2019-02-15T06:02:47Z,gpt-2,https://github.com/openai/gpt-2/issues/19 18,410617961,"It is painful to install gsutil, why not wget/curl",As shown in title,2019-02-15T06:00:31Z,gpt-2,https://github.com/openai/gpt-2/issues/18 17,410614714,Is there a plan to support PyTorch?,"Thanks for the awesome work, hope to support PyTorch.",2019-02-15T05:45:15Z,gpt-2,https://github.com/openai/gpt-2/issues/17 16,410572332,Release The Full Model!,I understand your concerns but I still think it's better to release the full model now and let people poke at it's abilities and discover potential issues quicker.,2019-02-15T02:02:26Z,gpt-2,https://github.com/openai/gpt-2/issues/16 15,410538437,Add a Dockerfile and ignore example artifacts,"If you'd like, here's a Dockerfile to toss up as an alternate installation method. Also quickly gitignored the file and file generated by running the example in the README. ",2019-02-14T23:27:04Z,gpt-2,https://github.com/openai/gpt-2/pull/15 14,410531605,Errors during model downloading,"When I try to download the model on my Ubuntu Linux 14.04 LTS box I get the following errors from : Also, do you need a CUDA card just to run the model?",2019-02-14T23:01:51Z,gpt-2,https://github.com/openai/gpt-2/issues/14 13,410524809,-,"Download.sh fails: + gsutil cp AccessDeniedException: 403 [user] does not have storage.objects.list access to gpt-2. ",2019-02-14T22:38:28Z,gpt-2,https://github.com/openai/gpt-2/issues/13 12,410496731,Improve environment setup & fix path errors & exec,"Added instructions on creating a virtual environment with . Obviates the need to use . Also ensures that the requirements installation will work: there won't be any conflicts in the environment. Fixed the path import errors by adding , explicitly importing: , and by moving the executable Python scripts to the project's root. Updated README documentation too. Made the script exectuable and changed it to use the first executable found on the system's . Additionally, made the Python scripts _not_ executable (they should be run with at any rate).",2019-02-14T21:19:00Z,gpt-2,https://github.com/openai/gpt-2/pull/12 11,410488746,generate_unconditional_samples.py Only return {text},"I used tensorflow_gpu (didn't have the patience to use the cpu) No errors, but when I run generate_unconditional_samples.py with and without flags only return {text} Can you please advise?",2019-02-14T20:56:59Z,gpt-2,https://github.com/openai/gpt-2/issues/11 10,410479772,License,"Hi, is there a license associated with plans towards MIT licensing this model?",2019-02-14T20:32:49Z,gpt-2,https://github.com/openai/gpt-2/issues/10 9,410474539,Help doing transfer learning to generate spanish-language text?,"Hi! Amazing results I know this is an open-ended and lazy question, but I'd appreciate if you could give me some pointers into how to re-train the model with additional text in another language (e.g spanish). I already have a small (6 MB) dataset in spanish, and I'm not very well versed in ML but I'm curious about playing with your model. Thanks! I'll be sure to report results back if I somehow figure it out :)",2019-02-14T20:18:42Z,gpt-2,https://github.com/openai/gpt-2/issues/9 8,410461687,Fix ModuleNotFoundError: No module named 'src',fix #7 ,2019-02-14T19:44:56Z,gpt-2,https://github.com/openai/gpt-2/pull/8 7,410458016,ModuleNotFoundError: No module named 'src'," Python can't import from a subfolder unless it's on the Python path.",2019-02-14T19:35:39Z,gpt-2,https://github.com/openai/gpt-2/issues/7 6,410456478,Python 3.6 requirement must be explicitly stated,You need to use _exactly_ 3.6.x since 3.6 has f-strings (which this repo uses) and 3.7 doesn't support TensorFlow.,2019-02-14T19:31:56Z,gpt-2,https://github.com/openai/gpt-2/issues/6 5,410452672,Window size error when trying to sample,"I'm trying to sample from 117M using the following commands: And I get the following error: > File line 34 > raise ValueError(f""can't get samples longer than window size: {hparams.n_ctx}"") > ^ > SyntaxError: invalid syntax",2019-02-14T19:22:27Z,gpt-2,https://github.com/openai/gpt-2/issues/5 4,410446073,fix: create models directory,Fixes #3 .,2019-02-14T19:05:24Z,gpt-2,https://github.com/openai/gpt-2/pull/4 2,410444981,Fix download_model.sh by removing trailing forward-flash,"Running as-is results in: Removing the trailing forward-slash fixes the issue.",2019-02-14T19:02:27Z,gpt-2,https://github.com/openai/gpt-2/pull/2 1,410440239,Issue with gsutil download_model.sh,"Hi, I'm not familiar with gsutil. Installed it freshly using the 6 steps of : Upon running the script : When I'm not logged in on cloud. When I'm logged in on cloud : Thanks ",2019-02-14T18:49:43Z,gpt-2,https://github.com/openai/gpt-2/issues/1