Computer Game Pajama Sam

[Emigdio Binet]

InPajama Sam: No Need to Hide When it's Dark Outside, you take on the role of Pajama Sam himself, finally confronting the darkness hiding away in his closet. As a kid, he's scared of the stuff, but it's not as easy as stowing away the concept of darkness into a lunchbox like Sam hopes. Once in the closet, the kid trips and falls into a Pee-Wee- Hermanesque realm of darkness where everything has a face, from trees to meat grinders. After the transdimensional fall, Sam loses some key darkness banishing items, so your task is to find them and march up to the tip-top of a spooky treehouse to finish the fight off.

From here it plays like an abbreviated LucasArts point-and-click adventure, and for good reason. Ron Gilbert, of Maniac Mansion and The Secret of Monkey Island fame, teamed up with producer Shelley Day to create Humongous Games in 1992, a name contributed by none other than LucasArts cohort Tim Schaefer. Writer of The Secret of Monkey Island and its sequel LeChuck's Revenge, Dave Grossman, came on to pen Pajama Sam, so it retains a familiar mix of visual gags propped up by a simpler, but equally goofy form of adventure game logic in the puzzles.

The art and animation still stand up, though I remember it looking nicer on a fuzzy CRT. They had a way of blending the low-res pixelated scenes into something that more mirrored what I was watching on TV every Saturday morning.

Pajama Sam knows kids aren't going to make it through the space between narrative beats, the time in every adventure game where you're stuck staring at an idle screen, pixel-hunting for a missed object or interactive piece of scenery. Click on nearly any part of a screen and something will happen. A tree root will grow a face and moan about how they didn't have videogames back in their day. A pizza will fall out of the sky onto a tennis racket. Flowers will play Patty Cake. A gorilla will appear in a smudge of darkness, whip out a banana, and have a quick snack.

Sam's a lovable kid, too, in large part thanks to Pamela Adlon's enthused performance. You'd probably recognise the voice instantly. Adlon's also behind one of the greatest animated characters of all time: Bobby from King of the Hill. More recently you might've seen her on the excellent series Better Things, where she writes, directs and plays the lead. Coincidentally, I started watching it the same week I returned to Pajama Sam. An odd thing, to hear the same voice go from being scared of the dark to making jokes about going through menopause. Humongously entertaining, indeed.

Pajama Sam had to be. A single adventure game had to sustain four kids, minimum, for a couple weeks. I met most of my friends at Julie's, someone that kindly watched half the kids in the neighbourhood. Miraculously, she gave us free rein on the computer. We played together, bunched around the computer after school almost every day. Back at school we discussed theories in the sandbox and spouted epiphanies when our swings passed. We should use the doorknob on the door without a doorknob! Baby geniuses.

Revisiting Pajama Sam has been a heartwarming reminder of my own PC gaming origins, and I'm sure I'm not alone here, which makes it an easy recommendation to pick up and play again no matter your age.

We never did crack the whole case on our own, stuck trying to find where to read the water meter in the mines in order to win a quiz show hosted by living doors. While Pajama Sam never got to rubber-with-a- pulley-in-the-middle levels of convoluted, it was still quite difficult to parse for our developing minds. Kids are pretty dumb, turns out.

Luckily, most Humongous Entertainment games are easy to find and play today, thanks to Nightdive Studios, the studio responsible for keeping dozens of old games working on modern machines, including the upcoming remake of System Shock. I expected to get frustrated and bored with Pajama Sam, to play purely for the nostalgia, but I was delighted throughout, even if the narrative is literally: the dark isn't that bad. Believe it or not, I've already overcome my fear of the dark. I can sleep in total darkness and not wet the bed once all night. Sure, I have to pay bills now and my brain eats itself in new ways, like fearing the death of my parents or the collapse of a state's entire infrastructure or the incoming eco-apocalypse, but Pajama Sam helped me nip that whole dark thing in the bud for me 25 years ago. I'm actually very grateful.

The only major caveat weighing down my hot air balloon ride into the past are a couple bugs present in the Steam version of Pajama Sam. Some scenes don't always load in crucial interactive objects properly, or at all. A late game puzzle requires Sam to find some oars in order to fish something out of the water. They're normally propped up on a wall in a room with an organ and a talking bust, but in my instance? Nothing. No oars, even though I could climb the organ and leap on the chandelier to initiate an interactive swinging sequence in which the oars snap into existence.

Worse, though, is that even though I could see the oars, I couldn't ever finish the swinging minigame. Sam would reach out, I would time my clicks to give him momentum, but the oars were forever just out of reach. A quick Google search confirms little game-breaking bugs like this are somewhat common, so keep that in mind before dipping back in. If it's any assurance, it's a short game already, and you can jam escape to skip literally any animation. I got back to the same sequence, unbugged, in about ten minutes.

James is stuck in an endless loop, playing the Dark Souls games on repeat until Elden Ring and Silksong set him free. He's a truffle pig for indie horror and weird FPS games too, seeking out games that actively hurt to play. Otherwise he's wandering Austin, identifying mushrooms and doodling grackles. "}), " -0-10/js/authorBio.js"); } else console.error('%c FTE ','background: #9306F9; color: #ffffff','no lazy slice hydration function available'); James DavenportSocial Links NavigationJames is stuck in an endless loop, playing the Dark Souls games on repeat until Elden Ring and Silksong set him free. He's a truffle pig for indie horror and weird FPS games too, seeking out games that actively hurt to play. Otherwise he's wandering Austin, identifying mushrooms and doodling grackles.

In addition to the data, we are also releasing the tools we built to create SlimPajama. Applying MinHashLSH (Leskovec et al. 2014) deduplication to trillion token datasets like RedPajama was not possible with off-the-shelf open-source code. We made several improvements to existing solutions to produce an infrastructure that can perform MinHashLSH deduplication on trillion token datasets in a distributed, multi-threaded, and memory efficient fashion. Today we are open-sourcing this infrastructure to enable the community to easily create higher quality, extensively deduplicated datasets in the future.

The latest research (Penedo et al. 2023) has shown that data quality is as important as data quantity. While training on more than one data epoch can be beneficial, this should be a choice rather than a side-effect of duplicates in the dataset. We decided to extensively deduplicate RedPajama to produce a dataset with higher information density. This means when using SlimPajama, you can achieve higher accuracy with the same compute budget when compared to other datasets.

Our original intention was to use the RedPajama data as-is, but upon analysis we discovered some corpora contained missing files while others had a large percentage of duplicates. RedPajama followed the deduplication guidelines in the LLaMA paper, which was less strict and only operated within each data source, not between them. To improve compute efficiency and data quality, we decided to produce a cleaned and extensively deduplicated version of the data, which led to the development of SlimPajama.

To produce SlimPajama, we first removed short, low quality documents from RedPajama. After removing punctuation, space symbols, newlines and tabs, we filtered out documents with less than 200 characters. These documents typically contain only meta data and no useful information. Low-length filter was applied to every corpora other than Books and GitHub where we found useful short documents. In total this removed 1.86% of documents from RedPajama.

Table 2: Document low-length filter rates.Training on deduplicated data makes language models better by improving training compute efficiency and reducing the chance of models producing text memorized from the training data (Penedo et al., 2023; Abbas et al., 2023; Lee et al., 2021; Holtzman, 2019; Face, 2023). Every corpus contained duplicates with the most significant duplication found in CommonCrawl and GitHub. Penedo et al. (2023) found similar rates of duplication in CommonCrawl data. In total, we were able to prune 49.6% of bytes from RedPajama, leaving us with the 627B token SlimPajama dataset.

To perform deduplication we used MinHashLSH Leskovec et al. (2014) with a Jaccard similarity threshold of 0.8. We construct document signatures on top of pre-processed lower-cased 13-grams. Pre-processing includes removing punctuation, consecutive spaces, newlines and tabs. We also strip the documents to remove any leading or trailing escape characters. Deduplication was performed both within and between data sources.

Table 4: Dataset source proportions for SlimPajama and related datasets.Similar to RedPajama, there can be distributional bias in the SlimPajama dataset that can manifest in various forms in the downstream model deployment. There are other risks associated with large language models such as amplifying social stereotypes, memorizing training data, or revealing private or secure information.

The existing open-source infrastructure for pre-processing text datasets did not scale to datasets with 1 Trillion tokens. The most significant bottlenecks occurred within the interleaving, shuffling and deduplication steps. Our optimizations are inspired by producer-consumer patterns that we implemented using a standard multiprocessing library available in Python. We also had to rewrite the datasketch (Zhu 2023) implementation to reduce both required memory and make the code more efficient in a distributed setting. Details of our implementation are available at _processing/slimpajama. The end-to-end pre-processing took 2.5 days with a 64 core CPU. The largest memory consumption observed was 1.4 TB.

3a8082e126