diff --git a/solutions/web_crawl_Q&A/requirements.txt b/solutions/web_crawl_Q&A/requirements.txt new file mode 100644 index 00000000..787fa27f --- /dev/null +++ b/solutions/web_crawl_Q&A/requirements.txt @@ -0,0 +1,80 @@ +aiohttp==3.8.3 +aiosignal==1.3.1 +appnope==0.1.3 +asttokens==2.2.1 +async-timeout==4.0.2 +attrs==22.2.0 +backcall==0.2.0 +beautifulsoup4==4.11.1 +blobfile==2.0.1 +bs4==0.0.1 +certifi==2022.12.7 +charset-normalizer==2.1.1 +comm==0.1.2 +contourpy==1.0.7 +cycler==0.11.0 +debugpy==1.6.5 +decorator==5.1.1 +docopt==0.6.2 +entrypoints==0.4 +executing==1.2.0 +filelock==3.9.0 +fonttools==4.38.0 +frozenlist==1.3.3 +html==1.13 +huggingface-hub==0.11.1 +idna==3.4 +ipykernel==6.20.1 +ipython==8.8.0 +jedi==0.18.2 +joblib==1.2.0 +jupyter_client==7.4.8 +jupyter_core==5.1.3 +kiwisolver==1.4.4 +lxml==4.9.2 +matplotlib==3.6.3 +matplotlib-inline==0.1.6 +multidict==6.0.4 +nest-asyncio==1.5.6 +numpy==1.24.1 +openai==0.26.1 +packaging==23.0 +pandas==1.5.2 +parso==0.8.3 +pexpect==4.8.0 +pickleshare==0.7.5 +Pillow==9.4.0 +pipreqs==0.4.11 +platformdirs==2.6.2 +plotly==5.12.0 +prompt-toolkit==3.0.36 +psutil==5.9.4 +ptyprocess==0.7.0 +pure-eval==0.2.2 +pycryptodomex==3.17 +Pygments==2.14.0 +pyparsing==3.0.9 +python-dateutil==2.8.2 +pytz==2022.7.1 +PyYAML==6.0 +pyzmq==24.0.1 +regex==2022.10.31 +requests==2.28.1 +scikit-learn==1.2.0 +scipy==1.10.0 +six==1.16.0 +soupsieve==2.3.2.post1 +stack-data==0.6.2 +tenacity==8.1.0 +threadpoolctl==3.1.0 +tiktoken==0.1.2 +tokenizers==0.13.2 +tornado==6.2 +tqdm==4.64.1 +traitlets==5.8.1 +transformers==4.25.1 +typing_extensions==4.4.0 +urllib3==1.26.13 +wcwidth==0.2.5 +yarg==0.1.9 +yarl==1.8.2 \ No newline at end of file diff --git a/solutions/web_crawl_Q&A/web-qa.ipynb b/solutions/web_crawl_Q&A/web-qa.ipynb new file mode 100644 index 00000000..94724161 --- /dev/null +++ b/solutions/web_crawl_Q&A/web-qa.ipynb @@ -0,0 +1,1285 @@ +{ + "cells": [ + { + "cell_type": "code", + "execution_count": 20, + "metadata": {}, + "outputs": [ + { + "name": "stdout", + "output_type": "stream", + "text": [ + "https://openai.com/\n", + "https://openai.com/blog/tags/announcements\n", + "https://openai.com/blog/introducing-openai\n", + "https://openai.com/blog/authors/ilya\n", + "https://openai.com/blog/requests-for-research-2\n", + "https://openai.com/blog/authors/diederik\n", + "https://openai.com/blog/block-sparse-gpu-kernels\n", + "https://openai.com/blog/authors/alec\n", + "https://openai.com/blog/fine-tuning-gpt-2\n", + "https://openai.com/blog/authors/paul\n", + "https://openai.com/blog/concrete-ai-safety-problems\n", + "https://openai.com/blog/learning-to-summarize-with-human-feedback\n", + "https://openai.com/blog/authors/long\n", + "https://openai.com/blog/authors/lowe\n", + "https://openai.com/blog/learning-to-cooperate-compete-and-communicate\n", + "https://openai.com/blog/authors/jean\n", + "https://openai.com/blog/authors/igor\n", + "https://openai.com/blog/neural-mmo\n", + "https://openai.com/blog/authors/phillip\n", + "https://openai.com/blog/evolved-policy-gradients\n", + "https://openai.com/blog/authors/richard\n", + "https://openai.com/blog/better-exploration-with-parameter-noise\n", + "https://openai.com/blog/authors/xi\n", + "https://openai.com/blog/authors/matthias\n", + "https://openai.com/blog/solving-rubiks-cube\n", + "https://openai.com/blog/authors/ilge\n", + "https://openai.com/blog/vpt\n", + "https://openai.com/blog/authors/brandon\n", + "https://openai.com/blog/authors/raul\n", + "https://openai.com/blog/authors/bowen\n", + "https://openai.com/blog/authors/jie\n", + "https://openai.com/blog/tags/five\n", + "https://openai.com/blog/openai-five-benchmark-results\n", + "https://openai.com/blog/openai-five/#rapid\n", + "https://openai.com/blog/authors/henrique\n", + "https://openai.com/blog/authors/susan\n", + "https://openai.com/blog/authors/brooke\n", + "https://openai.com/blog/authors/michael-petrov\n", + "https://openai.com/blog/multimodal-neurons\n", + "https://openai.com/blog/authors/shan\n", + "https://openai.com/blog/authors/daniela\n", + "https://openai.com/blog/authors/nick\n", + "https://openai.com/blog/authors/chris\n", + "https://openai.com/blog/introducing-activation-atlases\n", + "https://openai.com/blog/authors/ludwig-schubert\n", + "https://openai.com/blog/authors/justin\n", + "https://openai.com/blog/authors/gabriel\n", + "https://openai.com/blog/microscope\n", + "https://openai.com/blog/authors/przemyslaw\n", + "https://openai.com/blog/authors/david\n", + "https://openai.com/blog/authors/jakub-pachocki\n", + "https://openai.com/blog/authors/christy\n", + "https://openai.com/blog/improving-language-model-behavior\n", + "https://openai.com/blog/authors/irene\n", + "https://openai.com/blog/gpt-2-1-5b-release\n", + "https://openai.com/blog/authors/jack-clark\n", + "https://openai.com/blog/cooperation-on-safety\n", + "https://openai.com/blog/authors/amanda\n", + "https://openai.com/blog/ai-safety-needs-social-scientists\n", + "https://openai.com/blog/adversarial-example-research\n", + "https://openai.com/blog/authors/sandy\n", + "https://openai.com/blog/authors/ian\n", + "https://openai.com/blog/machine-learning-unconference\n", + "https://openai.com/events/code-of-conduct.txt\n", + "https://openai.com/blog/authors/rocky\n", + "https://openai.com/blog/authors/nicolas\n", + "https://openai.com/blog/preparing-for-malicious-uses-of-ai\n", + "https://openai.com/blog/authors/michael\n", + "https://openai.com/blog/spam-detection-in-the-physical-world\n", + "https://openai.com/blog/authors/rachel\n", + "https://openai.com/blog/authors/alex-ray\n", + "https://openai.com/blog/generalizing-from-simulation\n", + "https://openai.com/blog/authors/lerrel\n", + "https://openai.com/blog/authors/xue\n", + "https://openai.com/blog/faster-robot-simulation-in-python\n", + "https://openai.com/blog/safety-gym\n", + "https://openai.com/blog/authors/joshua\n", + "https://openai.com/blog/spinning-up-in-deep-rl\n", + "https://openai.com/blog/spinning-up-in-deep-rl-workshop-review\n", + "https://openai.com/blog/hackathon-follow-up\n", + "https://openai.com/blog/authors/parnian\n", + "https://openai.com/blog/openai-hackathon\n", + "https://openai.com/events/hackathon.txt\n", + "https://openai.com/blog/authors/josh-tobin\n", + "https://openai.com/blog/report-from-the-self-organizing-conference\n", + "https://openai.com/blog/faulty-reward-functions\n", + "https://openai.com/blog/authors/miles\n", + "https://openai.com/blog/language-model-safety-and-misuse\n", + "https://openai.com/blog/authors/tyna\n", + "https://openai.com/blog/webgpt\n", + "https://openai.com/blog/authors/jacob-hilton\n", + "https://openai.com/blog/measuring-goodharts-law\n", + "https://openai.com/careers/research-engineer\n", + "https://openai.com/blog/authors/leo\n", + "https://openai.com/blog/learning-to-summarize-with-human-feedback/#optimizingtherewardmodel\n", + "https://openai.com/blog/procgen-benchmark\n", + "https://openai.com/blog/first-retro-contest-retrospective\n", + "https://openai.com/blog/authors/oleg\n", + "https://openai.com/blog/roboschool\n", + "https://openai.com/blog/gym-retro\n", + "https://openai.com/blog/authors/vicki\n", + "https://openai.com/blog/retro-contest\n", + "https://openai.com/blog/authors/alex\n", + "https://openai.com/blog/reptile\n", + "https://openai.com/blog/dall-e-2-pre-training-mitigations\n", + "https://openai.com/blog/authors/larissa\n", + "https://openai.com/blog/openai-scholars-2018-final-projects\n", + "https://openai.com/blog/authors/karl\n", + "https://openai.com/blog/grade-school-math\n", + "https://openai.com/blog/authors/vineet\n", + "https://openai.com/blog/authors/christopher\n", + "https://openai.com/blog/quantifying-generalization-in-reinforcement-learning\n", + "https://openai.com/blog/authors/reiichiro\n", + "https://openai.com/blog/authors/suchir\n", + "https://openai.com/blog/authors/katie\n", + "https://openai.com/blog/authors/sandhini\n", + "https://openai.com/blog/authors/pamela\n", + "https://openai.com/blog/authors/steven\n", + "https://openai.com/blog/authors/gretchen\n", + "https://openai.com/blog/authors/jan\n", + "https://openai.com/blog/critiques\n", + "https://openai.com/blog/authors/william-saunders\n", + "https://openai.com/blog/authors/catherine\n", + "https://openai.com/blog/our-approach-to-alignment-research\n", + "https://openai.com/blog/best-practices-for-deploying-language-models\n", + "https://openai.com/blog/instruction-following/#limitations\n", + "https://openai.com/blog/economic-impacts\n", + "https://openai.com/blog/authors/sam-manning\n", + "https://openai.com/scholars\n", + "https://openai.com/blog/openai-scholars-2021-final-projects\n", + "https://openai.com/blog/openai-scholars-2020-final-projects\n", + "https://openai.com/resources\n", + "https://openai.com/blog/openai-scholars-spring-2020\n", + "https://openai.com/blog/openai-scholars-class-of-19\n", + "https://openai.com/blog/openai-scholars-2019-final-projects\n", + "https://openai.com/blog/authors/jonathan\n", + "https://openai.com/blog/discovering-types-for-entity-disambiguation\n", + "https://openai.com/blog/openai-five-benchmark\n", + "https://openai.com/blog/openai-five-defeats-dota-2-world-champions/#arena\n", + "https://openai.com/blog/openai-five/#ourapproach\n", + "https://openai.com/blog/more-on-dota-2/#botexploits\n", + "https://openai.com/blog/openai-five-benchmark-results/#training\n", + "https://openai.com/blog/openai-five-finals\n", + "https://openai.com/five/#overview\n", + "https://openai.com/blog/dota-2\n", + "https://openai.com/the-international\n", + "https://openai.com/blog/more-on-dota-2\n", + "https://openai.com/blog/the-international-2018-results\n", + "https://openai.com/blog/openai-five-defeats-dota-2-world-champions/#cooperativemode\n", + "https://openai.com/blog/openai-five-defeats-dota-2-world-champions\n", + "https://openai.com/blog/authors/jeff\n", + "https://openai.com/blog/authors/adrien\n", + "https://openai.com/blog/authors/joost\n", + "https://openai.com/blog/authors/peter-zhokhov\n", + "https://openai.com/blog/authors/glenn\n", + "https://openai.com/blog/authors/peter\n", + "https://openai.com/blog/authors/raphael\n", + "https://openai.com/blog/authors/lilian\n", + "https://openai.com/blog/techniques-for-training-large-neural-networks\n", + "https://openai.com/blog/authors/alex-paino\n", + "https://openai.com/blog/authors/nikolas\n", + "https://openai.com/blog/openai-five\n", + "https://openai.com/blog/authors/bob\n", + "https://openai.com/blog/authors/qiming\n", + "https://openai.com/blog/authors/wojciech\n", + "https://openai.com/blog/authors/arthur\n", + "https://openai.com/blog/authors/mateusz\n", + "https://openai.com/blog/authors/maciek\n", + "https://openai.com/blog/authors/jerry\n", + "https://openai.com/blog/authors/lei\n", + "https://openai.com/blog/how-to-train-your-openai-five\n", + "https://openai.com/blog/authors/jonas-schneider\n", + "https://openai.com/jobs/#robotics\n", + "https://openai.com/interview-guide\n", + "https://openai.com/blog/learning-dexterity\n", + "https://openai.com/blog/authors/rafal\n", + "https://openai.com/blog/ingredients-for-robotics-research\n", + "https://openai.com/blog/authors/vikash\n", + "https://openai.com/blog/authors/marcin\n", + "https://openai.com/blog/authors/prafulla\n", + "https://openai.com/blog/authors/szymon-sidor\n", + "https://openai.com/blog/openai-baselines-dqn\n", + "https://openai.com/blog/authors/tamim\n", + "https://openai.com/blog/learning-montezumas-revenge-from-a-single-demonstration\n", + "https://openai.com/blog/authors/bradly\n", + "https://openai.com/blog/authors/rein\n", + "https://openai.com/blog/authors/jonathan-ho\n", + "https://openai.com/blog/learning-a-hierarchy\n", + "https://openai.com/blog/authors/peter-chen\n", + "https://openai.com/blog/authors/kevin\n", + "https://openai.com/blog/authors/filip\n", + "https://openai.com/five\n", + "https://openai.com/blog/authors/yilun\n", + "https://openai.com/blog/authors/joseph\n", + "https://openai.com/blog/interpretable-machine-learning-through-teaching\n", + "https://openai.com/blog/authors/smitha\n", + "https://openai.com/blog/learning-to-model-other-minds\n", + "https://openai.com/blog/authors/shimon\n", + "https://openai.com/blog/authors/maruan\n", + "https://openai.com/blog/authors/jakob-foerster\n", + "https://openai.com/blog/nonlinear-computation-in-linear-networks\n", + "https://openai.com/blog/energy-based-models\n", + "https://openai.com/blog/emergent-tool-use\n", + "https://openai.com/blog/authors/ingmar\n", + "https://openai.com/blog/authors/todor\n", + "https://openai.com/blog/learning-concepts-with-energy-functions\n", + "https://openai.com/blog/authors/yi\n", + "https://openai.com/blog/authors/pieter\n", + "https://openai.com/blog/authors/aviv\n", + "https://openai.com/blog/instruction-following\n", + "https://openai.com/blog/learning-to-communicate\n", + "https://openai.com/blog/authors/jon\n", + "https://openai.com/blog/summarizing-books\n", + "https://openai.com/blog/authors/chelsea\n", + "https://openai.com/blog/gathering_human_feedback\n", + "https://openai.com/blog/authors/dario-amodei\n", + "https://openai.com/blog/science-of-ai\n", + "https://openai.com/blog/authors/jared\n", + "https://openai.com/blog/authors/sam\n", + "https://openai.com/blog/gpt-2-6-month-follow-up\n", + "https://openai.com/blog/better-language-models/#update\n", + "https://openai.com/blog/authors/david-luan\n", + "https://openai.com/blog/authors/danny\n", + "https://openai.com/blog/ai-and-efficiency\n", + "https://openai.com/blog/authors/david-lansky\n", + "https://openai.com/blog/authors/tom\n", + "https://openai.com/blog/testing-robustness\n", + "https://openai.com/blog/authors/jacob\n", + "https://openai.com/blog/authors/yi-sun\n", + "https://openai.com/blog/authors/daniel\n", + "https://openai.com/blog/authors/dan\n", + "https://openai.com/blog/deep-reinforcement-learning-from-human-preferences\n", + "https://openai.com/blog/authors/geoffrey\n", + "https://openai.com/blog/debate\n", + "https://openai.com/blog/authors/jeffrey\n", + "https://openai.com/blog/authors/nisan\n", + "https://openai.com/blog/amplifying-ai-training\n", + "https://openai.com/blog/authors/daniel-ziegler\n", + "https://openai.com/blog/baselines-acktr-a2c\n", + "https://openai.com/blog/authors/yuhuai\n", + "https://openai.com/blog/authors/shun\n", + "https://openai.com/blog/authors/elman\n", + "https://openai.com/blog/openai-baselines-ppo\n", + "https://openai.com/blog/language-unsupervised\n", + "https://openai.com/blog/tags/baselines\n", + "https://openai.com/blog/authors/scott\n", + "https://openai.com/blog/sparse-transformer\n", + "https://openai.com/blog/authors/rewon\n", + "https://openai.com/blog/glow\n", + "https://openai.com/blog/authors/john\n", + "https://openai.com/blog/openai-gym-beta\n", + "https://openai.com/blog/authors/tim\n", + "https://openai.com/jobs\n", + "https://openai.com/blog/formal-math\n", + "https://openai.com/blog/authors/stanislas\n", + "https://openai.com/blog/authors/jesse\n", + "https://openai.com/blog/generative-models\n", + "https://openai.com/blog/authors/andrej\n", + "https://openai.com/blog/distill\n", + "https://openai.com/blog/authors/vicki-cheung\n", + "https://openai.com/blog/jukebox\n", + "https://openai.com/projects/five\n", + "https://openai.com/blog/authors/christine\n", + "https://openai.com/blog/authors/jong\n", + "https://openai.com/blog/authors/heewoo\n", + "https://openai.com/blog/musenet\n", + "https://openai.com/blog/better-language-models\n", + "https://openai.com/blog/robots-that-learn\n", + "https://openai.com/blog/authors/ankur\n", + "https://openai.com/blog/authors/erika-reinhardt\n", + "https://openai.com/blog/deep-double-descent\n", + "https://openai.com/blog/authors/tristan\n", + "https://openai.com/blog/authors/preetum\n", + "https://openai.com/blog/authors/boaz\n", + "https://openai.com/blog/authors/yamini\n", + "https://openai.com/blog/authors/gal\n", + "https://openai.com/blog/tags/gpt-2\n", + "https://openai.com/blog/clip\n", + "https://openai.com/blog/ai-and-compute\n", + "https://openai.com/blog/authors/girish\n", + "https://openai.com/blog/special-projects\n", + "https://openai.com/blog/authors/sam-altman\n", + "https://openai.com/blog/unsupervised-sentiment-neuron\n", + "https://openai.com/blog/dall-e\n", + "https://openai.com/blog/authors/aditya\n", + "https://openai.com/blog/authors/mark\n", + "https://openai.com/blog/authors/mikhail\n", + "https://openai.com/blog/authors/vedant\n", + "https://openai.com/blog/competitive-self-play\n", + "https://openai.com/blog/authors/trapit\n", + "https://openai.com/blog/meta-learning-for-wrestling\n", + "https://openai.com/blog/authors/yura\n", + "https://openai.com/blog/reinforcement-learning-with-prediction-based-rewards\n", + "https://openai.com/blog/authors/harri\n", + "https://openai.com/blog/image-gpt\n", + "https://openai.com/blog/evolution-strategies\n", + "https://openai.com/blog/infrastructure-for-deep-learning\n", + "https://openai.com/blog/generative-models/#gan\n", + "https://openai.com/blog/generative-models#improving-gans\n", + "https://openai.com/blog/tags/multimodal\n", + "https://openai.com/gpt-3\n", + "https://openai.com/javascript:setMathjaxCookie()\n", + "HTTP Error 404: Not Found\n", + "https://openai.com/abs/2005.14165v1\n", + "HTTP Error 404: Not Found\n", + "https://openai.com/list/cs.CL/new\n", + "HTTP Error 404: Not Found\n", + "https://openai.com/abs/2005.14165v3\n", + "HTTP Error 404: Not Found\n", + "https://openai.com/auth/show-endorsers/2005.14165\n", + "HTTP Error 404: Not Found\n", + "https://openai.com/list/cs/recent\n", + "HTTP Error 404: Not Found\n", + "https://openai.com/abs/2005.14165?context=cs\n", + "HTTP Error 404: Not Found\n", + "https://openai.com/{url_path('ignore_me')}\n", + "HTTP Error 404: Not Found\n", + "https://openai.com/abs/2005.14165v2\n", + "HTTP Error 404: Not Found\n", + "https://openai.com/show-email/b5cb66e9/2005.14165\n", + "HTTP Error 404: Not Found\n", + "https://openai.com/prevnext?id=2005.14165&function=next&context=cs.CL\n", + "HTTP Error 404: Not Found\n", + "https://openai.com/format/2005.14165\n", + "HTTP Error 404: Not Found\n", + "https://openai.com/prevnext?id=2005.14165&function=prev&context=cs.CL\n", + "HTTP Error 404: Not Found\n", + "https://openai.com/pdf/2005.14165\n", + "HTTP Error 404: Not Found\n", + "https://openai.com/tb/2005.14165\n", + "HTTP Error 404: Not Found\n", + "https://openai.com/list/cs.CL/2005\n", + "HTTP Error 404: Not Found\n", + "https://openai.com/list/cs.CL/recent\n", + "HTTP Error 404: Not Found\n", + "https://openai.com/blog/dall-e-2\n", + "https://openai.com/blog/authors/openai\n", + "https://openai.com/blog/improving-verifiability\n", + "https://openai.com/blog/dall-e-2-extending-creativity\n", + "https://openai.com/blog/the-international\n", + "https://openai.com/blog/symposium-2019\n", + "https://openai.com/blog/tags/culture\n", + "https://openai.com/blog/learning-day\n", + "https://openai.com/blog/openai-fellows-fall-2018\n", + "https://openai.com/blog/neurips-2020\n", + "https://openai.com/blog/tags/community\n", + "https://openai.com/blog/universe\n", + "https://openai.com/blog/openai-gym-beta/#rl\n", + "https://openai.com/blog/openai-technical-goals/#goal4\n", + "https://openai.com/blog/authors/elon\n", + "https://openai.com/blog/scaling-kubernetes-to-7500-nodes\n", + "https://openai.com/blog/scaling-kubernetes-to-2500-nodes\n", + "https://openai.com/blog/authors/christopher-berner\n", + "https://openai.com/blog/authors/bchess\n", + "https://openai.com/blog/authors/eric\n", + "https://openai.com/blog/forecasting-misuse\n", + "https://openai.com/forecasting-misuse-paper\n", + "https://openai.com/prevnext?id=2301.04246&function=prev&context=cs.CY\n", + "HTTP Error 404: Not Found\n", + "https://openai.com/auth/show-endorsers/2301.04246\n", + "HTTP Error 404: Not Found\n", + "https://openai.com/format/2301.04246\n", + "HTTP Error 404: Not Found\n", + "https://openai.com/pdf/2301.04246\n", + "HTTP Error 404: Not Found\n", + "https://openai.com/show-email/64c5c6bd/2301.04246\n", + "HTTP Error 404: Not Found\n", + "https://openai.com/list/cs.CY/recent\n", + "HTTP Error 404: Not Found\n", + "https://openai.com/prevnext?id=2301.04246&function=next&context=cs.CY\n", + "HTTP Error 404: Not Found\n", + "https://openai.com/list/cs.CY/new\n", + "HTTP Error 404: Not Found\n", + "https://openai.com/list/cs.CY/2301\n", + "HTTP Error 404: Not Found\n", + "https://openai.com/abs/2301.04246?context=cs\n", + "HTTP Error 404: Not Found\n", + "https://openai.com/blog/authors/greg\n", + "https://openai.com/blog/dall-e-api-now-available-in-public-beta\n", + "https://openai.com/blog/api-no-waitlist\n", + "https://openai.com/blog/dall-e-introducing-outpainting\n", + "https://openai.com/blog/team-update\n", + "https://openai.com/blog/chatgpt-plus\n", + "https://openai.com/blog/openai-api\n", + "https://openai.com/jobs/#applied-ai\n", + "https://openai.com/blog/authors/mira\n", + "https://openai.com/join\n", + "Unable to parse page https://openai.com/join due to JavaScript being required\n", + "HTTP Error 403: Forbidden\n", + "https://openai.com/blog/tags/residency\n", + "https://openai.com/blog/openai-licenses-gpt-3-technology-to-microsoft\n", + "https://openai.com/blog/microsoft\n", + "https://openai.com/blog/team-update-august\n", + "https://openai.com/blog/new-ai-classifier-for-indicating-ai-written-text\n", + "https://openai.com/blog/authors/lama\n", + "https://openai.com/blog/authors/scott-aaronson\n", + "https://openai.com/blog/authors/jan-hendrik-kirchner\n", + "https://openai.com/blog/tags/api\n", + "https://openai.com/blog/openai-fellows\n", + "https://openai.com/blog/tags/scholars\n", + "https://openai.com/blog/openai-and-microsoft-extend-partnership\n", + "https://openai.com/blog/dall-e-now-available-without-waitlist\n", + "https://openai.com/blog/helen-toner-joins\n", + "https://openai.com/blog/team-update-january\n", + "https://openai.com/blog/team-plus-plus#interns\n", + "https://openai.com/blog/openai-codex\n", + "https://openai.com/blog/openai-scholars-2019\n", + "https://openai.com/blog/authors/ashley\n", + "https://openai.com/blog/openai-scholars\n", + "https://openai.com/blog/dall-e-now-available-in-beta\n", + "https://openai.com/blog/new-and-improved-embedding-model\n", + "https://openai.com/blog/authors/ryan\n", + "https://openai.com/blog/authors/arvind\n", + "https://openai.com/blog/authors/ted\n", + "https://openai.com/blog/dall-e-2-update\n", + "https://openai.com/blog/authors/joanne\n", + "https://openai.com/blog/tags/fellows\n", + "https://openai.com/blog/openai-summer-fellows-2018\n", + "https://openai.com/blog/authors/maddie\n", + "https://openai.com/blog/codex-apps\n", + "https://openai.com/blog/codex\n", + "HTTP Error 404: Not Found\n", + "https://openai.com/blog/new-and-improved-content-moderation-tooling\n", + "https://openai.com/blog/authors/teddy\n", + "https://openai.com/blog/authors/angela\n", + "https://openai.com/blog/authors/chong\n", + "https://openai.com/blog/welcome-pieter-and-shivon\n", + "https://openai.com/blog/openai-technical-goals\n", + "https://openai.com/blog/procgen-minerl-competitions\n", + "https://openai.com/blog/will-hurd-joins\n", + "https://openai.com/blog/fund\n", + "https://openai.com/news\n", + "HTTP Error 404: Not Found\n", + "https://openai.com/news/introducing-our-first-investments\n", + "HTTP Error 404: Not Found\n", + "https://openai.com/blog/introducing-text-and-code-embeddings\n", + "https://openai.com/blog/authors/boris\n", + "https://openai.com/blog/openai-scholars-2018-meet-our-scholars\n", + "https://openai.com/blog/team-plus-plus\n", + "https://openai.com/blog/gpt-3-apps\n", + "https://openai.com/jobs/#open\n", + "https://openai.com/blog/customized-gpt-3\n", + "https://openai.com/blog/authors/luke\n", + "https://openai.com/blog/authors/rachel-lim\n", + "https://openai.com/blog/authors/michael-wu\n", + "https://openai.com/blog/openai-supporters\n", + "https://openai.com/blog/openai-residency\n", + "https://openai.com/blog/leadership-team-update\n", + "https://openai.com/blog/organizational-update\n", + "https://openai.com/blog/openai-fellows-interns-2019\n", + "https://openai.com/blog/openai-scholars-2020\n", + "https://openai.com/blog/gpt-3-edit-insert\n", + "https://openai.com/blog/authors/mo\n", + "https://openai.com/blog/openai-pytorch\n", + "https://openai.com/blog/openai-scholars-2019-meet-our-scholars\n", + "https://openai.com/blog/openai-charter\n", + "https://openai.com/blog/openai-and-microsoft\n", + "https://openai.com/blog/openai-lp\n", + "https://openai.com/blog/reducing-bias-and-improving-safety-in-dall-e-2\n", + "https://openai.com/terms\n", + "https://openai.com/api/policies/service-terms\n", + "https://openai.com/api/policies/sharing-publication\n", + "https://openai.com/api/policies/terms\n", + "https://openai.com/security/disclosure\n", + "https://openai.com/blog/whisper\n", + "https://openai.com/blog/authors/tao\n", + "https://openai.com/research\n", + "https://openai.com/api/docs\n", + "Unable to parse page https://openai.com/api/docs due to JavaScript being required\n", + "HTTP Error 403: Forbidden\n", + "https://openai.com/dall-e-2\n", + "https://openai.com/privacy\n", + "https://openai.com/api\n", + "https://openai.com/blog\n", + "https://openai.com/blog/triton\n", + "https://openai.com/blog/authors/philippe\n", + "https://openai.com/jobs/#acceleration\n", + "https://openai.com/blog/robust-adversarial-inputs\n", + "https://openai.com/blog/authors/anish-athalye\n", + "https://openai.com/blog/tags/milestones\n", + "https://openai.com/alignment\n", + "https://openai.com\n", + "https://openai.com/publications\n", + "https://openai.com/charter\n", + "https://openai.com/blog/tags/research\n", + "https://openai.com/fund\n", + "https://openai.com/about\n", + "https://openai.com/timeline\n", + "https://openai.com/careers\n", + "https://openai.com/api/examples\n", + "Unable to parse page https://openai.com/api/examples due to JavaScript being required\n", + "HTTP Error 403: Forbidden\n", + "https://openai.com/api/login\n", + "Unable to parse page https://openai.com/api/login due to JavaScript being required\n", + "HTTP Error 403: Forbidden\n", + "https://openai.com/newsroom\n", + "https://openai.com/api/policies\n", + "https://openai.com/api/pricing\n", + "https://openai.com/contact-sales\n", + "https://openai.com/api/pricing/#faq-fine-tuning-pricing-calculation\n", + "https://openai.com/blog/tags/events\n", + "https://openai.com/blog/chatgpt\n" + ] + } + ], + "source": [ + "import requests\n", + "import re\n", + "import urllib.request\n", + "from bs4 import BeautifulSoup\n", + "from collections import deque\n", + "from html.parser import HTMLParser\n", + "from urllib.parse import urlparse\n", + "import os\n", + "\n", + "# Regex pattern to match a URL\n", + "HTTP_URL_PATTERN = r'^http[s]*://.+'\n", + "\n", + "# Define root domain to crawl\n", + "domain = \"openai.com\"\n", + "full_url = \"https://openai.com/\"\n", + "\n", + "# Create a class to parse the HTML and get the hyperlinks\n", + "class HyperlinkParser(HTMLParser):\n", + " def __init__(self):\n", + " super().__init__()\n", + " # Create a list to store the hyperlinks\n", + " self.hyperlinks = []\n", + "\n", + " # Override the HTMLParser's handle_starttag method to get the hyperlinks\n", + " def handle_starttag(self, tag, attrs):\n", + " attrs = dict(attrs)\n", + "\n", + " # If the tag is an anchor tag and it has an href attribute, add the href attribute to the list of hyperlinks\n", + " if tag == \"a\" and \"href\" in attrs:\n", + " self.hyperlinks.append(attrs[\"href\"])\n", + "\n", + "# Function to get the hyperlinks from a URL\n", + "def get_hyperlinks(url):\n", + " \n", + " # Try to open the URL and read the HTML\n", + " try:\n", + " # Open the URL and read the HTML\n", + " with urllib.request.urlopen(url) as response:\n", + "\n", + " # If the response is not HTML, return an empty list\n", + " if not response.info().get('Content-Type').startswith(\"text/html\"):\n", + " return []\n", + " \n", + " # Decode the HTML\n", + " html = response.read().decode('utf-8')\n", + " except Exception as e:\n", + " print(e)\n", + " return []\n", + "\n", + " # Create the HTML Parser and then Parse the HTML to get hyperlinks\n", + " parser = HyperlinkParser()\n", + " parser.feed(html)\n", + "\n", + " return parser.hyperlinks\n", + "\n", + "# Function to get the hyperlinks from a URL that are within the same domain\n", + "def get_domain_hyperlinks(local_domain, url):\n", + " clean_links = []\n", + " for link in set(get_hyperlinks(url)):\n", + " clean_link = None\n", + "\n", + " # If the link is a URL, check if it is within the same domain\n", + " if re.search(HTTP_URL_PATTERN, link):\n", + " # Parse the URL and check if the domain is the same\n", + " url_obj = urlparse(link)\n", + " if url_obj.netloc == local_domain:\n", + " clean_link = link\n", + "\n", + " # If the link is not a URL, check if it is a relative link\n", + " else:\n", + " if link.startswith(\"/\"):\n", + " link = link[1:]\n", + " elif link.startswith(\"#\") or link.startswith(\"mailto:\"):\n", + " continue\n", + " clean_link = \"https://\" + local_domain + \"/\" + link\n", + "\n", + " if clean_link is not None:\n", + " if clean_link.endswith(\"/\"):\n", + " clean_link = clean_link[:-1]\n", + " clean_links.append(clean_link)\n", + "\n", + " # Return the list of hyperlinks that are within the same domain\n", + " return list(set(clean_links))\n", + "\n", + "\n", + "def crawl(url):\n", + " # Parse the URL and get the domain\n", + " local_domain = urlparse(url).netloc\n", + "\n", + " # Create a queue to store the URLs to crawl\n", + " queue = deque([url])\n", + "\n", + " # Create a set to store the URLs that have already been seen (no duplicates)\n", + " seen = set([url])\n", + "\n", + " # Create a directory to store the text files\n", + " if not os.path.exists(\"text/\"):\n", + " os.mkdir(\"text/\")\n", + "\n", + " if not os.path.exists(\"text/\"+local_domain+\"/\"):\n", + " os.mkdir(\"text/\" + local_domain + \"/\")\n", + "\n", + " # Create a directory to store the csv files\n", + " if not os.path.exists(\"processed\"):\n", + " os.mkdir(\"processed\")\n", + "\n", + " # While the queue is not empty, continue crawling\n", + " while queue:\n", + "\n", + " # Get the next URL from the queue\n", + " url = queue.pop()\n", + " print(url) # for debugging and to see the progress\n", + "\n", + " # Save text from the url to a .txt file\n", + " with open('text/'+local_domain+'/'+url[8:].replace(\"/\", \"_\") + \".txt\", \"w\") as f:\n", + "\n", + " # Get the text from the URL using BeautifulSoup\n", + " soup = BeautifulSoup(requests.get(url).text, \"html.parser\")\n", + "\n", + " # Get the text but remove the tags\n", + " text = soup.get_text()\n", + "\n", + " # If the crawler gets to a page that requires JavaScript, it will stop the crawl\n", + " if (\"You need to enable JavaScript to run this app.\" in text):\n", + " print(\"Unable to parse page \" + url + \" due to JavaScript being required\")\n", + " \n", + " # Otherwise, write the text to the file in the text directory\n", + " f.write(text)\n", + "\n", + " # Get the hyperlinks from the URL and add them to the queue\n", + " for link in get_domain_hyperlinks(local_domain, url):\n", + " if link not in seen:\n", + " queue.append(link)\n", + " seen.add(link)\n", + "\n", + "crawl(full_url)" + ] + }, + { + "cell_type": "code", + "execution_count": 1, + "metadata": {}, + "outputs": [], + "source": [ + "def remove_newlines(serie):\n", + " serie = serie.str.replace('\\n', ' ')\n", + " serie = serie.str.replace('\\\\n', ' ')\n", + " serie = serie.str.replace(' ', ' ')\n", + " serie = serie.str.replace(' ', ' ')\n", + " return serie" + ] + }, + { + "cell_type": "code", + "execution_count": null, + "metadata": {}, + "outputs": [], + "source": [ + "import pandas as pd\n", + "\n", + "# Create a list to store the text files\n", + "texts=[]\n", + "\n", + "# Get all the text files in the text directory\n", + "for file in os.listdir(\"text/\" + domain + \"/\"):\n", + "\n", + " # Open the file and read the text\n", + " with open(\"text/\" + domain + \"/\" + file, \"r\") as f:\n", + " text = f.read()\n", + "\n", + " # Omit the first 11 lines and the last 4 lines, then replace -, _, and #update with spaces.\n", + " texts.append((file[11:-4].replace('-',' ').replace('_', ' ').replace('#update',''), text))\n", + "\n", + "# Create a dataframe from the list of texts\n", + "df = pd.DataFrame(texts, columns = ['fname', 'text'])\n", + "\n", + "# Set the text column to be the raw text with the newlines removed\n", + "df['text'] = df.fname + \". \" + remove_newlines(df.text)\n", + "df.to_csv('processed/scraped.csv')\n", + "df.head()" + ] + }, + { + "cell_type": "code", + "execution_count": 26, + "metadata": {}, + "outputs": [ + { + "data": { + "text/plain": [ + "" + ] + }, + "execution_count": 26, + "metadata": {}, + "output_type": "execute_result" + }, + { + "data": { + "image/png": "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", + "text/plain": [ + "
" + ] + }, + "metadata": {}, + "output_type": "display_data" + } + ], + "source": [ + "import tiktoken\n", + "\n", + "# Load the cl100k_base tokenizer which is designed to work with the ada-002 model\n", + "tokenizer = tiktoken.get_encoding(\"cl100k_base\")\n", + "\n", + "df = pd.read_csv('processed/scraped.csv', index_col=0)\n", + "df.columns = ['title', 'text']\n", + "\n", + "# Tokenize the text and save the number of tokens to a new column\n", + "df['n_tokens'] = df.text.apply(lambda x: len(tokenizer.encode(x)))\n", + "\n", + "# Visualize the distribution of the number of tokens per row using a histogram\n", + "df.n_tokens.hist()" + ] + }, + { + "cell_type": "code", + "execution_count": 27, + "metadata": {}, + "outputs": [], + "source": [ + "max_tokens = 500\n", + "\n", + "# Function to split the text into chunks of a maximum number of tokens\n", + "def split_into_many(text, max_tokens = max_tokens):\n", + "\n", + " # Split the text into sentences\n", + " sentences = text.split('. ')\n", + "\n", + " # Get the number of tokens for each sentence\n", + " n_tokens = [len(tokenizer.encode(\" \" + sentence)) for sentence in sentences]\n", + " \n", + " chunks = []\n", + " tokens_so_far = 0\n", + " chunk = []\n", + "\n", + " # Loop through the sentences and tokens joined together in a tuple\n", + " for sentence, token in zip(sentences, n_tokens):\n", + "\n", + " # If the number of tokens so far plus the number of tokens in the current sentence is greater \n", + " # than the max number of tokens, then add the chunk to the list of chunks and reset\n", + " # the chunk and tokens so far\n", + " if tokens_so_far + token > max_tokens:\n", + " chunks.append(\". \".join(chunk) + \".\")\n", + " chunk = []\n", + " tokens_so_far = 0\n", + "\n", + " # If the number of tokens in the current sentence is greater than the max number of \n", + " # tokens, go to the next sentence\n", + " if token > max_tokens:\n", + " continue\n", + "\n", + " # Otherwise, add the sentence to the chunk and add the number of tokens to the total\n", + " chunk.append(sentence)\n", + " tokens_so_far += token + 1\n", + "\n", + " return chunks\n", + " \n", + "\n", + "shortened = []\n", + "\n", + "# Loop through the dataframe\n", + "for row in df.iterrows():\n", + "\n", + " # If the text is None, go to the next row\n", + " if row[1]['text'] is None:\n", + " continue\n", + "\n", + " # If the number of tokens is greater than the max number of tokens, split the text into chunks\n", + " if row[1]['n_tokens'] > max_tokens:\n", + " shortened += split_into_many(row[1]['text'])\n", + " \n", + " # Otherwise, add the text to the list of shortened texts\n", + " else:\n", + " shortened.append( row[1]['text'] )" + ] + }, + { + "cell_type": "code", + "execution_count": 28, + "metadata": {}, + "outputs": [ + { + "data": { + "text/plain": [ + "" + ] + }, + "execution_count": 28, + "metadata": {}, + "output_type": "execute_result" + }, + { + "data": { + "image/png": "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", + "text/plain": [ + "
" + ] + }, + "metadata": {}, + "output_type": "display_data" + } + ], + "source": [ + "df = pd.DataFrame(shortened, columns = ['text'])\n", + "df['n_tokens'] = df.text.apply(lambda x: len(tokenizer.encode(x)))\n", + "df.n_tokens.hist()" + ] + }, + { + "cell_type": "code", + "execution_count": 29, + "metadata": {}, + "outputs": [ + { + "data": { + "text/html": [ + "
\n", + "\n", + "\n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + "
textn_tokensembeddings
0blog authors maddie. Maddie Hall - OpenAI ...175[-0.012958061881363392, -0.006103983614593744,...
1blog authors tom. Tom Brown - OpenAI ...228[-0.0053874170407652855, -0.009962032549083233...
2blog openai scholars 2019 final projects. Op...492[0.0019150723237544298, -0.0070442273281514645...
3In this project, I used curiosity-driven explo...478[-0.0067560747265815735, 0.0004431474662851542...
4Results revealed that the optimal RL policies ...499[-0.012868616729974747, 0.0029640409629791975,...
\n", + "
" + ], + "text/plain": [ + " text n_tokens \\\n", + "0 blog authors maddie. Maddie Hall - OpenAI ... 175 \n", + "1 blog authors tom. Tom Brown - OpenAI ... 228 \n", + "2 blog openai scholars 2019 final projects. Op... 492 \n", + "3 In this project, I used curiosity-driven explo... 478 \n", + "4 Results revealed that the optimal RL policies ... 499 \n", + "\n", + " embeddings \n", + "0 [-0.012958061881363392, -0.006103983614593744,... \n", + "1 [-0.0053874170407652855, -0.009962032549083233... \n", + "2 [0.0019150723237544298, -0.0070442273281514645... \n", + "3 [-0.0067560747265815735, 0.0004431474662851542... \n", + "4 [-0.012868616729974747, 0.0029640409629791975,... " + ] + }, + "execution_count": 29, + "metadata": {}, + "output_type": "execute_result" + } + ], + "source": [ + "import openai\n", + "\n", + "df['embeddings'] = df.text.apply(lambda x: openai.Embedding.create(input=x, engine='text-embedding-ada-002')['data'][0]['embedding'])\n", + "df.to_csv('processed/embeddings.csv')\n", + "df.head()" + ] + }, + { + "cell_type": "code", + "execution_count": 30, + "metadata": {}, + "outputs": [ + { + "data": { + "text/html": [ + "
\n", + "\n", + "\n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + "
textn_tokensembeddings
0blog authors maddie. Maddie Hall - OpenAI ...175[-0.012958061881363392, -0.006103983614593744,...
1blog authors tom. Tom Brown - OpenAI ...228[-0.0053874170407652855, -0.009962032549083233...
2blog openai scholars 2019 final projects. Op...492[0.0019150723237544298, -0.0070442273281514645...
3In this project, I used curiosity-driven explo...478[-0.0067560747265815735, 0.0004431474662851542...
4Results revealed that the optimal RL policies ...499[-0.012868616729974747, 0.0029640409629791975,...
\n", + "
" + ], + "text/plain": [ + " text n_tokens \\\n", + "0 blog authors maddie. Maddie Hall - OpenAI ... 175 \n", + "1 blog authors tom. Tom Brown - OpenAI ... 228 \n", + "2 blog openai scholars 2019 final projects. Op... 492 \n", + "3 In this project, I used curiosity-driven explo... 478 \n", + "4 Results revealed that the optimal RL policies ... 499 \n", + "\n", + " embeddings \n", + "0 [-0.012958061881363392, -0.006103983614593744,... \n", + "1 [-0.0053874170407652855, -0.009962032549083233... \n", + "2 [0.0019150723237544298, -0.0070442273281514645... \n", + "3 [-0.0067560747265815735, 0.0004431474662851542... \n", + "4 [-0.012868616729974747, 0.0029640409629791975,... " + ] + }, + "execution_count": 30, + "metadata": {}, + "output_type": "execute_result" + } + ], + "source": [ + "from openai.embeddings_utils import distances_from_embeddings\n", + "\n", + "df['embeddings'] = df.text.apply(lambda x: openai.Embedding.create(input=x, engine='text-embedding-ada-002')['data'][0]['embedding']))\n", + "\n", + "df.to_csv('processed/embeddings.csv')\n", + "df.head()" + ] + }, + { + "cell_type": "code", + "execution_count": 31, + "metadata": {}, + "outputs": [ + { + "data": { + "text/html": [ + "
\n", + "\n", + "\n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + "
textn_tokensembeddings
0blog authors maddie. Maddie Hall - OpenAI ...175[-0.012958061881363392, -0.006103983614593744,...
1blog authors tom. Tom Brown - OpenAI ...228[-0.0053874170407652855, -0.009962032549083233...
2blog openai scholars 2019 final projects. Op...492[0.0019150723237544298, -0.0070442273281514645...
3In this project, I used curiosity-driven explo...478[-0.0067560747265815735, 0.0004431474662851542...
4Results revealed that the optimal RL policies ...499[-0.012868616729974747, 0.0029640409629791975,...
\n", + "
" + ], + "text/plain": [ + " text n_tokens \\\n", + "0 blog authors maddie. Maddie Hall - OpenAI ... 175 \n", + "1 blog authors tom. Tom Brown - OpenAI ... 228 \n", + "2 blog openai scholars 2019 final projects. Op... 492 \n", + "3 In this project, I used curiosity-driven explo... 478 \n", + "4 Results revealed that the optimal RL policies ... 499 \n", + "\n", + " embeddings \n", + "0 [-0.012958061881363392, -0.006103983614593744,... \n", + "1 [-0.0053874170407652855, -0.009962032549083233... \n", + "2 [0.0019150723237544298, -0.0070442273281514645... \n", + "3 [-0.0067560747265815735, 0.0004431474662851542... \n", + "4 [-0.012868616729974747, 0.0029640409629791975,... " + ] + }, + "execution_count": 31, + "metadata": {}, + "output_type": "execute_result" + } + ], + "source": [ + "import pandas as pd\n", + "import numpy as np\n", + "from openai.embeddings_utils import distances_from_embeddings, cosine_similarity\n", + "\n", + "df=pd.read_csv('processed/embeddings.csv', index_col=0)\n", + "df['embeddings'] = df['embeddings'].apply(eval).apply(np.array)\n", + "\n", + "df.head()" + ] + }, + { + "cell_type": "code", + "execution_count": 32, + "metadata": {}, + "outputs": [ + { + "data": { + "text/plain": [ + "'No, you are not allowed to publish model outputs to Twitter without a human review. You must manually review each generation before sharing or while streaming, and indicate that the content is AI-generated in a way no user could reasonably miss or misunderstand.'" + ] + }, + "execution_count": 32, + "metadata": {}, + "output_type": "execute_result" + } + ], + "source": [ + "def create_context(\n", + " question, df, max_len=1800, size=\"ada\"\n", + "):\n", + " \"\"\"\n", + " Create a context for a question by finding the most similar context from the dataframe\n", + " \"\"\"\n", + "\n", + " # Get the embeddings for the question\n", + " q_embeddings = openai.Embedding.create(input=question, engine='text-embedding-ada-002')['data'][0]['embedding']\n", + "\n", + " # Get the distances from the embeddings\n", + " df['distances'] = distances_from_embeddings(q_embeddings, df['embeddings'].values, distance_metric='cosine')\n", + "\n", + "\n", + " returns = []\n", + " cur_len = 0\n", + "\n", + " # Sort by distance and add the text to the context until the context is too long\n", + " for i, row in df.sort_values('distances', ascending=True).iterrows():\n", + " \n", + " # Add the length of the text to the current length\n", + " cur_len += row['n_tokens'] + 4\n", + " \n", + " # If the context is too long, break\n", + " if cur_len > max_len:\n", + " break\n", + " \n", + " # Else add it to the text that is being returned\n", + " returns.append(row[\"text\"])\n", + "\n", + " # Return the context\n", + " return \"\\n\\n###\\n\\n\".join(returns)\n", + "\n", + "def answer_question(\n", + " df,\n", + " model=\"text-davinci-003\",\n", + " question=\"Am I allowed to publish model outputs to Twitter, without a human review?\",\n", + " max_len=1800,\n", + " size=\"ada\",\n", + " debug=False,\n", + " max_tokens=150,\n", + " stop_sequence=None\n", + "):\n", + " \"\"\"\n", + " Answer a question based on the most similar context from the dataframe texts\n", + " \"\"\"\n", + " context = create_context(\n", + " question,\n", + " df,\n", + " max_len=max_len,\n", + " size=size,\n", + " )\n", + " # If debug, print the raw model response\n", + " if debug:\n", + " print(\"Context:\\n\" + context)\n", + " print(\"\\n\\n\")\n", + "\n", + " try:\n", + " # Create a completions using the question and context\n", + " response = openai.Completion.create(\n", + " prompt=f\"Answer the question based on the context below, and if the question can't be answered based on the context, say \\\"I don't know\\\"\\n\\nContext: {context}\\n\\n---\\n\\nQuestion: {question}\\nAnswer:\",\n", + " temperature=0,\n", + " max_tokens=max_tokens,\n", + " top_p=1,\n", + " frequency_penalty=0,\n", + " presence_penalty=0,\n", + " stop=stop_sequence,\n", + " model=model,\n", + " )\n", + " return response[\"choices\"][0][\"text\"].strip()\n", + " except Exception as e:\n", + " print(e)\n", + " return \"\"" + ] + }, + { + "cell_type": "code", + "execution_count": 33, + "metadata": {}, + "outputs": [ + { + "data": { + "text/plain": [ + "\"I don't know.\"" + ] + }, + "execution_count": 33, + "metadata": {}, + "output_type": "execute_result" + } + ], + "source": [ + "answer_question(df, question=\"What day is it?\", debug=False)" + ] + }, + { + "cell_type": "code", + "execution_count": 34, + "metadata": {}, + "outputs": [ + { + "data": { + "text/plain": [ + "'The newest embeddings model is text-embedding-ada-002.'" + ] + }, + "execution_count": 34, + "metadata": {}, + "output_type": "execute_result" + } + ], + "source": [ + "answer_question(df, question=\"What is our newest embeddings model?\")" + ] + } + ], + "metadata": { + "kernelspec": { + "display_name": "env", + "language": "python", + "name": "python3" + }, + "language_info": { + "codemirror_mode": { + "name": "ipython", + "version": 3 + }, + "file_extension": ".py", + "mimetype": "text/x-python", + "name": "python", + "nbconvert_exporter": "python", + "pygments_lexer": "ipython3", + "version": "3.9.6" + }, + "orig_nbformat": 4, + "vscode": { + "interpreter": { + "hash": "05f34a34d73b71652304030c1097be3a5720ea2447153dd6542d145a26b73181" + } + } + }, + "nbformat": 4, + "nbformat_minor": 2 +} diff --git a/solutions/web_crawl_Q&A/web-qa.py b/solutions/web_crawl_Q&A/web-qa.py new file mode 100644 index 00000000..7fc7e723 --- /dev/null +++ b/solutions/web_crawl_Q&A/web-qa.py @@ -0,0 +1,382 @@ +################################################################################ +### Step 1 +################################################################################ + +import requests +import re +import urllib.request +from bs4 import BeautifulSoup +from collections import deque +from html.parser import HTMLParser +from urllib.parse import urlparse +import os +import pandas as pd +import tiktoken +import openai +from openai.embeddings_utils import distances_from_embeddings +import pandas as pd +import numpy as np +from openai.embeddings_utils import distances_from_embeddings, cosine_similarity + +# Regex pattern to match a URL +HTTP_URL_PATTERN = r'^http[s]*://.+' + +# Define root domain to crawl +domain = "openai.com" +full_url = "https://openai.com/" + +# Create a class to parse the HTML and get the hyperlinks +class HyperlinkParser(HTMLParser): + def __init__(self): + super().__init__() + # Create a list to store the hyperlinks + self.hyperlinks = [] + + # Override the HTMLParser's handle_starttag method to get the hyperlinks + def handle_starttag(self, tag, attrs): + attrs = dict(attrs) + + # If the tag is an anchor tag and it has an href attribute, add the href attribute to the list of hyperlinks + if tag == "a" and "href" in attrs: + self.hyperlinks.append(attrs["href"]) + +################################################################################ +### Step 2 +################################################################################ + +# Function to get the hyperlinks from a URL +def get_hyperlinks(url): + + # Try to open the URL and read the HTML + try: + # Open the URL and read the HTML + with urllib.request.urlopen(url) as response: + + # If the response is not HTML, return an empty list + if not response.info().get('Content-Type').startswith("text/html"): + return [] + + # Decode the HTML + html = response.read().decode('utf-8') + except Exception as e: + print(e) + return [] + + # Create the HTML Parser and then Parse the HTML to get hyperlinks + parser = HyperlinkParser() + parser.feed(html) + + return parser.hyperlinks + +################################################################################ +### Step 3 +################################################################################ + +# Function to get the hyperlinks from a URL that are within the same domain +def get_domain_hyperlinks(local_domain, url): + clean_links = [] + for link in set(get_hyperlinks(url)): + clean_link = None + + # If the link is a URL, check if it is within the same domain + if re.search(HTTP_URL_PATTERN, link): + # Parse the URL and check if the domain is the same + url_obj = urlparse(link) + if url_obj.netloc == local_domain: + clean_link = link + + # If the link is not a URL, check if it is a relative link + else: + if link.startswith("/"): + link = link[1:] + elif link.startswith("#") or link.startswith("mailto:"): + continue + clean_link = "https://" + local_domain + "/" + link + + if clean_link is not None: + if clean_link.endswith("/"): + clean_link = clean_link[:-1] + clean_links.append(clean_link) + + # Return the list of hyperlinks that are within the same domain + return list(set(clean_links)) + + +################################################################################ +### Step 4 +################################################################################ + +def crawl(url): + # Parse the URL and get the domain + local_domain = urlparse(url).netloc + + # Create a queue to store the URLs to crawl + queue = deque([url]) + + # Create a set to store the URLs that have already been seen (no duplicates) + seen = set([url]) + + # Create a directory to store the text files + if not os.path.exists("text/"): + os.mkdir("text/") + + if not os.path.exists("text/"+local_domain+"/"): + os.mkdir("text/" + local_domain + "/") + + # Create a directory to store the csv files + if not os.path.exists("processed"): + os.mkdir("processed") + + # While the queue is not empty, continue crawling + while queue: + + # Get the next URL from the queue + url = queue.pop() + print(url) # for debugging and to see the progress + + # Save text from the url to a .txt file + with open('text/'+local_domain+'/'+url[8:].replace("/", "_") + ".txt", "w") as f: + + # Get the text from the URL using BeautifulSoup + soup = BeautifulSoup(requests.get(url).text, "html.parser") + + # Get the text but remove the tags + text = soup.get_text() + + # If the crawler gets to a page that requires JavaScript, it will stop the crawl + if ("You need to enable JavaScript to run this app." in text): + print("Unable to parse page " + url + " due to JavaScript being required") + + # Otherwise, write the text to the file in the text directory + f.write(text) + + # Get the hyperlinks from the URL and add them to the queue + for link in get_domain_hyperlinks(local_domain, url): + if link not in seen: + queue.append(link) + seen.add(link) + +crawl(full_url) + +################################################################################ +### Step 5 +################################################################################ + +def remove_newlines(serie): + serie = serie.str.replace('\n', ' ') + serie = serie.str.replace('\\n', ' ') + serie = serie.str.replace(' ', ' ') + serie = serie.str.replace(' ', ' ') + return serie + + +################################################################################ +### Step 6 +################################################################################ + +# Create a list to store the text files +texts=[] + +# Get all the text files in the text directory +for file in os.listdir("text/" + domain + "/"): + + # Open the file and read the text + with open("text/" + domain + "/" + file, "r") as f: + text = f.read() + + # Omit the first 11 lines and the last 4 lines, then replace -, _, and #update with spaces. + texts.append((file[11:-4].replace('-',' ').replace('_', ' ').replace('#update',''), text)) + +# Create a dataframe from the list of texts +df = pd.DataFrame(texts, columns = ['fname', 'text']) + +# Set the text column to be the raw text with the newlines removed +df['text'] = df.fname + ". " + remove_newlines(df.text) +df.to_csv('processed/scraped.csv') +df.head() + +################################################################################ +### Step 7 +################################################################################ + +# Load the cl100k_base tokenizer which is designed to work with the ada-002 model +tokenizer = tiktoken.get_encoding("cl100k_base") + +df = pd.read_csv('processed/scraped.csv', index_col=0) +df.columns = ['title', 'text'] + +# Tokenize the text and save the number of tokens to a new column +df['n_tokens'] = df.text.apply(lambda x: len(tokenizer.encode(x))) + +# Visualize the distribution of the number of tokens per row using a histogram +df.n_tokens.hist() + +################################################################################ +### Step 8 +################################################################################ + +max_tokens = 500 + +# Function to split the text into chunks of a maximum number of tokens +def split_into_many(text, max_tokens = max_tokens): + + # Split the text into sentences + sentences = text.split('. ') + + # Get the number of tokens for each sentence + n_tokens = [len(tokenizer.encode(" " + sentence)) for sentence in sentences] + + chunks = [] + tokens_so_far = 0 + chunk = [] + + # Loop through the sentences and tokens joined together in a tuple + for sentence, token in zip(sentences, n_tokens): + + # If the number of tokens so far plus the number of tokens in the current sentence is greater + # than the max number of tokens, then add the chunk to the list of chunks and reset + # the chunk and tokens so far + if tokens_so_far + token > max_tokens: + chunks.append(". ".join(chunk) + ".") + chunk = [] + tokens_so_far = 0 + + # If the number of tokens in the current sentence is greater than the max number of + # tokens, go to the next sentence + if token > max_tokens: + continue + + # Otherwise, add the sentence to the chunk and add the number of tokens to the total + chunk.append(sentence) + tokens_so_far += token + 1 + + return chunks + + +shortened = [] + +# Loop through the dataframe +for row in df.iterrows(): + + # If the text is None, go to the next row + if row[1]['text'] is None: + continue + + # If the number of tokens is greater than the max number of tokens, split the text into chunks + if row[1]['n_tokens'] > max_tokens: + shortened += split_into_many(row[1]['text']) + + # Otherwise, add the text to the list of shortened texts + else: + shortened.append( row[1]['text'] ) + +################################################################################ +### Step 9 +################################################################################ + +df = pd.DataFrame(shortened, columns = ['text']) +df['n_tokens'] = df.text.apply(lambda x: len(tokenizer.encode(x))) +df.n_tokens.hist() + +################################################################################ +### Step 10 +################################################################################ + +df['embeddings'] = df.text.apply(lambda x: openai.Embedding.create(input=x, engine='text-embedding-ada-002')['data'][0]['embedding']) +df.to_csv('processed/embeddings.csv') +df.head() + +################################################################################ +### Step 11 +################################################################################ + +df=pd.read_csv('processed/embeddings.csv', index_col=0) +df['embeddings'] = df['embeddings'].apply(eval).apply(np.array) + +df.head() + +################################################################################ +### Step 12 +################################################################################ + +def create_context( + question, df, max_len=1800, size="ada" +): + """ + Create a context for a question by finding the most similar context from the dataframe + """ + + # Get the embeddings for the question + q_embeddings = openai.Embedding.create(input=question, engine='text-embedding-ada-002')['data'][0]['embedding'] + + # Get the distances from the embeddings + df['distances'] = distances_from_embeddings(q_embeddings, df['embeddings'].values, distance_metric='cosine') + + + returns = [] + cur_len = 0 + + # Sort by distance and add the text to the context until the context is too long + for i, row in df.sort_values('distances', ascending=True).iterrows(): + + # Add the length of the text to the current length + cur_len += row['n_tokens'] + 4 + + # If the context is too long, break + if cur_len > max_len: + break + + # Else add it to the text that is being returned + returns.append(row["text"]) + + # Return the context + return "\n\n###\n\n".join(returns) + +def answer_question( + df, + model="text-davinci-003", + question="Am I allowed to publish model outputs to Twitter, without a human review?", + max_len=1800, + size="ada", + debug=False, + max_tokens=150, + stop_sequence=None +): + """ + Answer a question based on the most similar context from the dataframe texts + """ + context = create_context( + question, + df, + max_len=max_len, + size=size, + ) + # If debug, print the raw model response + if debug: + print("Context:\n" + context) + print("\n\n") + + try: + # Create a completions using the questin and context + response = openai.Completion.create( + prompt=f"Answer the question based on the context below, and if the question can't be answered based on the context, say \"I don't know\"\n\nContext: {context}\n\n---\n\nQuestion: {question}\nAnswer:", + temperature=0, + max_tokens=max_tokens, + top_p=1, + frequency_penalty=0, + presence_penalty=0, + stop=stop_sequence, + model=model, + ) + return response["choices"][0]["text"].strip() + except Exception as e: + print(e) + return "" + +################################################################################ +### Step 13 +################################################################################ + +print(answer_question(df, question="What day is it?", debug=False)) + +print(answer_question(df, question="What is our newest embeddings model?")) \ No newline at end of file