Three Formerly Enslaved Artists Created Beautiful Pottery 150 Years Ago, and Now Their Wares Are Coveted Around the World The stunning vessels from the H. Wilson & Company were forgotten for generations, only to gain new appreciation for the craftsmanship that went into them Photographs by DeSean McClinton-Holland Jacoba Urist June 2025 A jar made by H. Wilson & Company in Capote, Texas. Right, the muddy banks of Salt Creek, a tributary of the Guadalupe River about 50 miles northeast of San Antonio, where Wilson’s pottery company sourced its fine red clay.  The Museum of Fine Arts, Houston; DeSean McClinton-Holland In 1856, the Reverend John McKamie Wilson Jr., a Presbyterian minister and entrepreneur interested in clay science, relocated from North Carolina to Texas. There in the Capote Hills—a rural, sparse enclave in Guadalupe County, 12 miles from the town of Seguin—Wilson opened a business called Guadalupe Pottery. Wilson mainly sold jugs, churns, crocks and cemetery flower jars. The pots featured crescent handles and a deep chocolate-colored interior of liquefied clay. Before refrigerators and iceboxes, high-fire, nonporous pottery was essential to life—the Tupperware or Ziploc bags of the 19th century. Clay pots preserved everything from grains, beef and butter to whiskey and even drinking water. The potters who worked under Wilson’s direction mostly used alkaline glaze, one of the oldest methods in ceramics, to create a glassy exterior from a slurry of wood ash, sand and clay. An arduous process in the 1850s, it took days to stoke underground wood-burning kilns to a high enough temperature for a successful firing. Three of Wilson’s potters were enslaved servants who traveled to Texas with him—Hiram, James and Wallace. For more than a decade, these men were responsible for nearly every facet of Guadalupe Pottery’s production: from mixing clay and expertly “throwing” the pots on a kick wheel to glazing and stacking vessels and meticulously controlling the temperature and duration of the kiln’s flames. The men kept working for the reverend even after Abraham Lincoln issued the Emancipation Proclamation on January 1, 1863. In Texas, slavery didn’t end until June 19, 1865, when Major General Gordon Granger arrived in Galveston and issued General Order No. 3, informing the people of Texas that all who’d been enslaved were free—the event commemorated as Juneteenth.  Paula King Harper, a descendant of Hiram Wilson, sits on the steps of the Sebastopol House in Seguin, Texas, with Wilson vessels from her own collection. Harper leads the Wilson Pottery Foundation and coordinates the Wilson Pottery Show each October. DeSean McClinton-Holland Subscribe to Smithsonian magazine now for just $19.99 This article is a selection from the June 2025 issue of Smithsonian magazine In 1869, Hiram founded his own stoneware business at a new site with James and Wallace, H. Wilson & Company. (Though not biologically related, all three took their enslaver’s surname, the practice of some Black Americans at the time.) Some scholars believe that H. Wilson & Company was the first business in Texas founded and owned by formerly enslaved people.  Pots created by the hands of Hiram Wilson and his colleagues have been passed down through households and sold at garage sales over the course of many generations. Today, their rich history, as well as their distinctive styles and glazes, make the jugs sought-after acquisitions for museum collections. Left to right: A preserve jar, a three-gallon butter churn and a large jar with a lid from the Museum of Fine Arts, Houston; a stoneware jar from the Smithsonian’s National Museum of American History.  The Museum of Fine Arts, Houston (3); NMAH A century and a half later, the pots they made are celebrated in the ceramics world. In silvery grays and greens, with uneven salt drips and textured glazes that resemble the moon’s surface or, sometimes, an orange peel, Wilson wares are coveted both as objets d’art and for their extraordinary story of Black self-determination in the postwar South. Wilson pottery now resides in museums across the United States, from the Smithsonian’s National Museum of American History to the Museum of Fine Arts, Houston—an indelible part of the country’s entrepreneurial and creative heritage.   The story of Hiram, James and Wallace Wilson is woven through with threads of folklore, given the sparse records kept at that time, particularly for people born into slavery. But over the last 50 years, researchers have uncovered more details about the Wilsons. Georgeanna Greer—a San Antonio pediatrician and ceramics collector-aficionado—was passionate about locating abandoned kilns, and she left a trove of Wilson information behind, including a 1973 taped interview with James Wilson’s son, James Wilson Jr.  Capote Baptist Church has been in the midst of a restoration since 2012. After a new pastor, Terry Williams, arrived, its membership surged from 4 congregants to 85. DeSean McClinton-Holland A wooden cross on the property of the Capote Baptist Church that Hiram Wilson founded in 1872 DeSean McClinton-Holland Emancipation offered the Wilsons an opportunity none had ever expected. Establishing a pottery business was complex and costly. It meant finding a suitable site, testing clay, constructing a kiln and hiring proficient workers. And the men did all this in an “intensely dangerous environment,” said Ashley Williams, a recent fellow at the Smithsonian American Art Museum. In the late 1860s, Black residents of Guadalupe County “filed almost 200 complaints with the Freedmen’s Bureau for unpaid wages and violent crimes” perpetrated against them. A Baptist missionary from Maine, the Reverend Leonard Ilsley, helped Hiram buy 600 acres of land for $500 and likely either gave Hiram a personal loan or served as a guarantor in the transaction. When Hiram died in 1884­—survived by his wife, Senia, and 11 children—Ilsley served as administrator of his estate. One of Hiram’s great-great-granddaughters, LaVerne Lewis Britt, first discovered her connection to the potters after her retirement, when she became interested in genealogy. Over the course of five years, she researched and wrote a book called In Praise of Hiram Wilson, which describes how her ancestor created a thriving post-slavery community in Capote. Hiram set aside 10 of his 600 acres for a Baptist church and became its minister. The white steepled chapel remains active today, beside a cemetery with cedar and crape myrtle trees. Many Wilsons have been laid to rest there, including Hiram, whose grave is marked by a tall obelisk.  Pastor Terry Williams has been leading the Capote Baptist Church since 2015. When he arrived, he says, he felt the spirit of Hiram Wilson and other former slaves who founded the church in 1872—“a cold breeze coming up in the summertime.”  DeSean McClinton-Holland A Bible rests on the piano at Capote Baptist Church. Right, the building has been in the midst of a restoration since 2012. After Pastor Williams arrived, its membership surged from 4 congregants to 85. Hiram also founded a one-room schoolhouse where Paula King Harper, another of his descendants, says her grandmother once taught. King Harper is the current president of the Wilson Pottery Foundation. On the phone, she described how interest in her celebrated ancestor’s technique has spread in collector circles since the organization’s founding in 1999. A pottery collector in San Antonio was at a garage sale and noticed the green salt glaze drip from underneath dirt that had dried on a gallon jug. “They purchased it for less than $10, went home, cleaned it up, and guess what it was? A stamped H. Wilson.” Finding that stamp is now the equivalent of finding a signed painting by a renowned artist. King Harper once sat at an auction that included several pieces of Wilson pottery from someone’s private estate. “There was a beautiful, pristine five-gallon jug I watched sell for $12,000,” she said. Traditionally, clay artists, who made quotidian jugs and jars rather than purely aesthetic works, have been considered second-tier makers, ranking below sculptors. But over the last few decades, contemporary and historic ceramicists like the Wilsons are receiving new scholarly and art-world attention. Williams, the recent Smithsonian fellow, is researching the Wilsons as part of her doctorate at Columbia University. She notes that most enslaved potters wouldn’t have been able to inscribe their names or initials on a pot. But once the Wilson potters were free, they added a maker’s mark, which made all the difference. Future generations were able to dive into the records and find out more about the remarkable potters. “Because we can tie the makers to these objects, which is so rare, it allows us to see the story about the resilience of the Wilson potters during slavery, and their extreme success and survival,” Williams said. Elmer Joe Brackner Jr. was a graduate student in anthropology at the University of Texas when he unearthed buried ceramic shards and located the original kilns at two different H. Wilson & Company sites. DeSean McClinton-Holland Ashley Williams, a recent pre-doctoral fellow at the Smithsonian American Art Museum, has been researching the Wilson potters as part of her PhD dissertation on crafts made by enslaved and imprisoned artists. DeSean McClinton-Holland The Wilsons’ style may have been influenced by the works coming out of the Edgefield district in South Carolina, a place then known as a kind of pottery Mecca. It was famous for massive, bulbous jars and glazes, features that the early Wilson pots share. Edgefield has its own significant history of highly skilled Black craftsmen, including David Drake, the earliest known enslaved potter to inscribe his work. CollectorsDeSean McClinton-Holland Later, the Wilsons’ pottery took on more distinctive features. The men started out using alkaline glaze, but once they opened their own business they switched to salt glaze, likely for its strength and waterproofing properties. Salt glaze also produces more consistent colors and textures, such as the bumpy orange-peel finish admired by H. Wilson & Company collectors. At the time, the method was uncommon in Texas, according to Michelle Johnson, project manager of the William J. Hill Texas Artisans and Artists Archive at the Bayou Bend Collection and Gardens, part of the Museum of Fine Arts, Houston. “And it was dangerous,” she said, because salt glaze produces highly toxic chlorine gas from the wet salt thrown into the kiln during firing. (For this reason, contemporary ceramicists avoid the method altogether.) The Wilsons may have learned the technique from Isaac Suttles, an Ohio-born potter known for salt glazing, who is listed in the United States census as living in Seguin at the time.  The Sebastopol House, a Greek Revival home, was built in the 1850s by enslaved laborers who had advanced skills in working with concrete. Today, the building houses the Wilson Pottery Museum.  DeSean McClinton-Holland Lid rims are another notable element of H. Wilson & Company’s jars. At the first pottery-making site, during their enslavement, the men made more tie-down rims—jars without lids that require paper or cloth as a cover, which were easier and faster to make. Later, at their own shop, they crafted rims fitted for lids. Those lid rims allowed for more watertightstorage, but they also required more sophisticated expertise.If you think you may have stumbled on one of these treasures at a garage or estate sale, look for a horseshoe-like handle—the most significant visual identifier of a Wilson pot. It’s still unclear why Hiram, James and Wallace switched to these thicker, rolled handles from their original Edgefield-inspired crescents. It may have been because horseshoe handles are sturdier, offering more surface connection to the pot’s body. It’s another puzzling question that Williams is pursuing and hopes to answer in the future.  It took a long time for the art world to discover the Wilsons’ creations. After Hiram’s death in 1884, James and Wallace Wilson went on to work at another pottery site run by Marion Durham, a potter from South Carolina who had moved to Texas with John Chandler, who was most likely his enslaved servant at the time. By the time the Wilsons joined them, Durham and Chandler were in business together. That site closed in 1903. These pots remained in circulation, but many details of their story were lost.  Modern replicas of the Wilson pottery logo created by Earline Green. DeSean McClinton-Holland Earline Green, an educator and a member of the Wilson Pottery Foundation’s board of directors. Green helps promote the potters’ legacy by demonstrating their innovative 19th-century techniques.  DeSean McClinton-Holland Decades later, an anthropology graduate student named Elmer Joe Brackner Jr. conducted a magnetometry survey to find buried pottery. (Clay contains iron oxides that can become magnetized.) At the original Guadalupe Pottery site he found what’s known as a “groundhog kiln,” a uniquely Southern, semi-subterranean kiln used in the 19th century for firing alkaline-glazed pottery. At the second site, home of the independent H. Wilson & Company, Brackner found ceramic-glazed shards and another groundhog kiln. His research, and that of Georgeanna Greer, the ceramic historian, helped piece together the Wilsons’ unusual story. After many years of advocacy by the Wilson Pottery Foundation, art historians and curators began scouring land deeds, court documents and handwritten capacity marks (numbers that denote the volume of each jug) to present a fuller picture of the Wilsons, sometimes correcting previous theories about their methodology and timeline. Texas ceramic artist Earline Green researches the Wilsons’ past while creating her own pottery to honor their legacy. In 2018, she interviewed Wilson descendants and collectors and visited historical societies in Texas for archival information. Two years later, she curated an exhibition on her campus in Fort Worth; the show also displayed ceramic pieces she created based on the work of the Wilson potters. Deacon Willie Hightower Sr., a descendant of Hiram Wilson and one of the current deacons at Capote Baptist Church. DeSean McClinton-Holland Hiram Wilson’sDeSean McClinton-Holland The narrative continues to evolve, due in no small part to proud descendants of Hiram, James and Wallace. Every three years, hundreds of people gather in Seguin, Texas, for a jubilant three-day Wilson family reunion that is open to the public. In 2023, festivities included the Wilson Pottery Foundation gala and a tenth anniversary celebration at the Wilson Pottery Museum. The last reunion fell on the weekend before Juneteenth, and the reunion committee arranged a kickoff concert Sunday evening followed by Juneteenth events in Seguin’s downtown square. “I was fortunate to grow up in Texas in the 1980s, around the time when our cousin LaVerne Lewis Britt had uncovered the history,” explains DeSean McClinton-Holland, the photographer of this story and a Wilson descendant through his maternal grandmother. “I grew up knowing bits of the story and visiting the church.” He describes the process of documenting Hiram Wilson’s legacy as a healing journey. “To think, he was born into slavery and was able to accomplish so much shortly thereafter,” he says. “It’s inspiring and motivating to think of all the freedoms we have now, and that we really have to push a bit harder.”  Get the latest Travel & Culture stories in your inbox.

A federal judge created a path for app makers like Spotify and Patreon to avoid paying Apple hefty commissions. Is this a win for consumers? It’s complicated.

Like other famous structures of similar dimensions, the 48-story Transamerica Pyramid, a revolutionary ‘70s modernist skyscraper and San Francisco icon, has a bit of history buried beneath its ground floor. [Photo: Nils Huenerfuerst/Unsplash] A recently unearthed time capsule, buried in 1974 and discovered during a recent round of renovations, offers a picture of San Francisco’s past. The site of the structure—then a parking lot—was initially part of the original shoreline of the city that reeked of historical significance, from the city’s growth as a shipping and banking capital. The capsule even contains a recipe for Pisco Punch, a cocktail that was invented at the nearby Bank Exchange Saloon, site of the city’s original stock exchange.  [Photo: courtesy SHVO] Part of an exhibit in the building lobby opening May 18, the time capsule’s contents are timeless: pictures of the building’s steel frame beginning to stretch skyward, or vintage news clippings and images of the city after its last ’60s flowering. But within the cylindrical steel capsule, which looks a bit like a large propane tank, there’s also a narrative about building in America, and how that’s radically changed in the last 50 years.  [Photo: courtesy SHVO] The battle over the permitting and construction of the Transamerica Pyramid in San Francisco from 1969 to 1972 offers a flashback to a different time in development, real estate, and construction. The tower was proposed and built in just three years, a sprint compared to the time it takes today to build a signature part of a city skyline. Construction alone for the One World Trade in New York City took eight years; the Comcast Tech Center in Philadelphia, which had issues with cracks in some of the steel frame, took five years; and the St. Regis in Chicago took four years. An analysis of high-rise buildings by Construction Physics found building speeds decreased significantly over the past century, in many cases extending the time it takes to finish by roughly 50%. [Photo: courtesy SHVO] Buildings are more complex and require more permitting today, including complicated environmental review processes. This time-consuming process of development has led to backlash against what opponents call stifling building regulations. It has also led to more engagement from architects around code reform issues including elevator rules and exit stairs, and the formation of the abundance agenda, a center-left push by pundits like Ezra Klein to get the nation building fast again.  “The pace of the approval and the construction here is unbelievable,” says developer Michael Shvo, who paid $650 million to acquire the Transamerica Pyramid in 2020, at the depths of the COVID office freeze.. “The ​​Mayor was very determined to get this thing approved, and Transamerica was very determined to get a building built, and with all the controversy, once they got the green light, they ran as fast as possible. They built it in two years, we couldn’t do that today.” [Photo: courtesy SHVO] A more humane debate Transamerica was then a massive business conglomerate with interests in banking, financial services, and insurance. According to former public relations staffer John Krizek, who worked for Transamerica during the pyramid’s construction and ultimately created the time capsule, the back-and-forth between protestors and developers at the time was more humane, more respectable, and more amusing.  The conversation around the Transamerica Pyramid was, at the time, a larger debate about images, architecture, and aesthetics. The tower was not just a unique shape, but would tower above the skyline. It was to be the city’s tallest building, and wouldn’t be surpassed until 2018’s Salesforce Tower.  [Photo: courtesy SHVO] Artists and community members protested the building for aesthetic reasons, and general distrust of large corporations. Posters passed around the city at the time proclaimed “San Francisco Gets the Shaft” or “Artists Against the Icicle.” The city’s then planning director called the pyramid, designed by architect William Pereira, “inhumane.”  [Photo: courtesy SHVO] During early street protests in front of the company’s office, Transamerica execs sent secretaries to bring ice tea to the protestors lining up outside. During another protest, Krizek and his colleagues printed up fake fortune cookies at a nearby Chinatown bakery, frantically stuffing messages like “Transamerica–Not a square outfit” or “People who protest pyramid seek Che-ops publicity.”  Krizek recalled that the company was determined to break ground in December 1969. The building plan was announced in January of that year, and there was a tax break worth approximately $750,000 expiring at the end of December. Since Krizek and his coworkers knew that as soon as the company was given approval to build, there would be an appeal, they planned to move fast and break ground before paperwork was filed. To head off any challenges, they staged a tractor and truck near the site and sent someone to pick up the approval during the midday lunch break; they were able to get a time-stamped photo of someone digging at site while those opposing the project saw their appeal delayed as staffer enjoyed their lunch. “The emotions around this building, I’ve never seen this for any other building in the world,” says Shvo. “The debates today are more practical; this structure will block my view or cast a shadow. You can’t say that about this building, it was a pyramid designed to let the light down to the street level. It didn’t block views, the only thing people could complain about was this idea of the Manhattanization of San Francisco.” Originally, Pereira’s design was meant for a new building for ABC in New York City. The network passed on the project, deeming the design too futuristic, and went with another architect’s vision. Today, the Transamerica Pyramid stands as an icon in San Francisco, with 80% of the space leased in a challenging office market. The building ABC picked instead? It’s since been demolished. 

Quantization is one of the key techniques for reducing the memory footprint of large language models (LLMs). It works by converting the data type of model parameters from higher-precision formats such as 32-bit floating point (FP32) or 16-bit floating point (FP16/BF16) to lower-precision integer formats, typically INT8 or INT4. For example, quantizing a model to 4-bit means each parameter uses only 0.5 bytes, compared to 4 bytes in FP32. Post-training quantization methods like GPTQ and AWQ can dramatically reduce the size of large models. A model like Llama 3 with 70 billion parameters can occupy around 140 GB in FP16, but this can be reduced to approximately 40 GB using 4-bit quantization, while still maintaining strong performance on downstream tasks. However, despite this substantial reduction, such models still exceed the memory capacity of most consumer-grade GPUs, which typically offer 24 GB to 32 GB of VRAM. To make these models truly accessible, quantization to even lower bitwidths, such as 2-bit, is required. While recent advances in low-bit quantization are promising, achieving stable and accurate 2-bit quantization remains a significant challenge. In this article, we review a technique called EoRA that helps compensate for quantization-induced errors. EoRA is a training-free method, meaning it can be applied quickly and efficiently to any model, even the largest ones. We’ll check how EoRA works and demonstrate how it can significantly improve the performance of 2-bit quantized models, bringing them close to the accuracy of their full-precision counterparts while being up to 5.5x smaller. We’ll analyze experimental results obtained using large models such as Qwen3-32B and Qwen2.5-72B, both quantized to 2-bit using state-of-the-art quantization techniques, to illustrate the effectiveness of EoRA. Diving into the Eigenspace in Search of an Adapter Post-training quantization or, more generally, compression aims to reduce model size or inference cost by minimizing the output difference between the original weights Wl​ and compressed weights Ŵl  using only a small calibration dataset. Most quantization methods are framed layer-wise, but the choice of compression formats is rigid and limits flexibility across diverse deployment needs. To bypass format constraints and improve accuracy, previous work, such as QLoRA [1] and HQQ+ [2], directly fine-tuned a Lora adapter on top of the frozen quantized models. It is also possible to reframe compression as a compensation problem: given a compressed model, introduce low-rank residual paths that specifically correct compression errors. A straightforward method uses SVD to decompose the compression error: \[\Delta W_l = W_l – \hat{W}_l\] into \[U_l \Sigma_l V_l^T\] forming low-rank approximations via two matrices: \[B_l = U_l \Sigma_l \] \[A_l = V_l^T\] where Al and Bl are the standard tensors of a LoRA adapter. However, plain SVD has two limitations: it does not minimize the original layerwise compression loss directly, and it allocates capacity uniformly across all error components, ignoring the varying importance of different parts of the model. To address this, NVIDIA proposes EoRA [3]. EoRA: Training-free Compensation for Compressed LLM with Eigenspace Low-Rank Approximation EoRA first projects the compression error into the eigenspace defined by the input activation covariance: \[\tilde{X} \tilde{X}^T\] where X̃ is the average activation over the calibration set. Then, by performing eigendecomposition, we get: \[\tilde{X} \tilde{X}^T = Q \Lambda Q^T\] The compression error ΔW is projected as: \[\Delta W’ = \Delta W Q’\] where Q′=QΛ. Then SVD is applied on ΔW′ to produce a low-rank approximation, and the result is projected back to the original space, adjusting the low-rank factors accordingly. This eigenspace projection changes the optimization objective: it weights the importance of different error components according to their contribution to the layerwise output (via eigenvalues), making the approximation more efficient. It can be computed quickly without any training, requires only calibration activations, and does not introduce extra inference latency. Moreover, the derivation shows that this approach leads to a direct minimization of the layerwise compression loss, not just the raw weight error. Analytically, truncating a singular value in the projected space corresponds to minimizing the true compression error under reasonable assumptions about the calibration activations. In their paper, NVIDIA presents a wide range of strong results showing that EoRA can significantly boost the accuracy of quantized models. However, their experiments focus mostly on older Quantization methods like GPTQ and are limited to mid-sized LLMs, up to 13B parameters, at 3-bit and 4-bit precisions. This leaves an open question: can EoRA still be effective for much larger models, using more modern quantization techniques, and even pushing down to 2-bit precision? Let’s find out. Calibrating an EoRA Adapter Suppose we have quantized models that show significantly degraded performance compared to their full-precision counterparts on certain tasks. Our goal is to reduce this performance gap using EoRA. For the experiments, I used Qwen2.5-72B Instruct and Qwen3-32B, both quantized to 2-bit using AutoRound (Apache 2.0 license), a state-of-the-art quantization algorithm developed by Intel. AutoRound leverages SignSGD optimization to fine-tune quantization, and is particularly effective for low-bit settings. All the models I made are available here (Apache 2.0 license): Quantized Qwen3 Quantized Qwen2.5 The 2-bit models were quantized with a group size of 32, except for which used a group size of 128. A larger group size reduces model size by storing less quantization metadata, but it introduces greater quantization error. I evaluated the models on IFEval, a benchmark that measures instruction-following capabilities. Results showed a noticeable drop in performance for the quantized versions. Image by the author To compensate for this degradation, I applied an EoRA adapter using the implementation provided in the GPTQModel library (licensed under Apache 2.0). The integration is straightforward. If you’re curious about how it’s implemented in PyTorch, the codebase is compact, clean, and easy to follow: GPTQModel’s EoRA implementation: eora.py EoRA requires a calibration dataset. Ideally, this dataset should reflect the model’s intended use case. However, since we don’t have a specific target task in this context and aim to preserve the model’s general capabilities, I used 1,024 randomly sampled examples from the C4 dataset (licensed under ODC-BY). Another key parameter is the LoRA rank, which greatly influences the effectiveness of the EoRA adapter. Its optimal value depends on the model architecture, the target task, and the calibration data. A higher rank may yield better performance but risks overfitting to the calibration set. It also increases the size of the adapter, counterproductive when the overall goal of quantization is to reduce memory usage. Conversely, a lower rank keeps the adapter lightweight but might not capture enough information to effectively compensate for quantization errors. In my experiments, I tested LoRA ranks of 32, 64, and 256. Below is the code used to create the EoRA adapter with GPTQModel: from gptqmodel import GPTQModel from gptqmodel.adapter.adapter import Lora from datasets import load_dataset calibration_dataset = load_dataset( "allenai/c4", data_files="en/c4-train.00001-of-01024.json.gz", split="train", download_mode="force_redownload" ).select(range(1024))["text"] eora_adapter_path = "Qwen3-32B-autoround-2bit-gptq-r256" model_path = "kaitchup/Qwen3-32B-autoround-2bit-gptq" eora = Lora( path=eora_adapter_path, rank=256, ) GPTQModel.adapter.generate( adapter=eora, model_id_or_path="Qwen/Qwen3-32B", quantized_model_id_or_path=model_path, calibration_dataset=calibration_dataset, calibration_dataset_concat_size=0, auto_gc=False) Using an NVIDIA A100 GPU on RunPod (referral link), it took approximately 4 hours to generate the EoRA adapter for the model Qwen3-32B-autoround-2bit-gptq. All EoRA adapters created for these models are publicly available (Apache 2.0 license): EoRA Adapters for Qwen2.5 and Qwen3 Evaluating EoRA Adapters for 2-bit LLMs Let’s evaluate the effect of the EoRA adapters. Do they improve the accuracy of the 2-bit models? Image by the author It works! The improvements are particularly notable for Qwen3-14B and Qwen3-32B. For instance, applying EoRA to Qwen3-32B, quantized to 2-bit with a group size of 128, resulted in an accuracy gain of nearly 7.5 points. Increasing the LoRA rank, from 32 to 64, also led to improvements, highlighting the impact of rank on performance. EoRA is also effective on larger models like Qwen2.5-72B, though the gains are more modest. Lower-rank adapters showed little to no benefit on this model; it wasn’t until I increased the rank to 256 that significant improvements began to appear. Memory Consumption of EoRA Using the EoRA adapter during inference results in the following increase in memory consumption: Image by the author The overhead is generally negligible. For instance for 2-bit Qwen3-14B, the adapters only add 257 MB and 514 MB to the total model size, with ranks of 32 and 64. With larger ranks, using an EoRA adapter becomes questionable as the total memory consumption may surpass the memory consumption of the same model quantized at a higher precision. For instance, 2-bit Qwen2.5 72B with an EoRA adapter of rank 256 is larger than 3-bit Qwen2.5 72B. Note: This estimate includes only the memory consumed by the adapter’s parameters. For completeness, we could also account for the memory used by adapter activations during inference. However, these are extremely small relative to other tensors (such as the model’s attention and MLP layers) and can safely be considered negligible. Conclusion EoRA works. We’ve confirmed that it’s a simple yet effective method for compensating quantization errors, even at 2-bit precision. It’s intuitive, training-free, and delivers meaningful performance gains. That said, there are a few trade-offs to consider: Rank search: Finding the optimal LoRA rank requires experimentation. It’s difficult to predict in advance whether a rank of 32 will be sufficient or whether a higher rank, like 256, will cause overfitting. The optimal value depends on the model, calibration data, and target task. Increased memory consumption: The goal of quantization is to reduce memory usage, often for highly constrained environments. While EoRA adapters are relatively lightweight at lower ranks, they do slightly increase memory consumption, particularly at higher ranks, reducing the overall efficiency of 2-bit quantization. Looking ahead, NVIDIA’s paper also demonstrates that EoRA adapters make excellent starting points for QLoRA fine-tuning. In other words, if you plan to fine-tune a 2-bit model using QLoRA, initializing from an EoRA-adapted model can lead to better results with less training effort. I’ve written about fine-tuning adapters for GPTQ model last year, in my newsletter: QLoRA with AutoRound: Cheaper and Better LLM Fine-tuning on Your GPU The main difference is that instead of initializing the adapter from scratch, we would load the EoRA adapter. This adapter will be fine-tuned. References [1] Dettmers et al, QLoRA: Efficient Finetuning of Quantized LLMs (2023), arXiv [2] Badri and Shaji, Towards 1-bit Machine Learning Models (2024), Mobius Labs’ Blog [3] Liu et al., EoRA: Training-free Compensation for Compressed LLM with Eigenspace Low-Rank Approximation (2024), arXiv The post Boost 2-Bit LLM Accuracy with EoRA appeared first on Towards Data Science.

The next year for Ubisoft could be a quiet one. The French publisher announced that it’s pushing back some of its biggest games to allow them more time for development. The move comes after an internal review that took place last fall and could see upcoming sequels like Far Cry 7, Assassin’s Creed Hexe, and the next Ghost Recon all delayed into 2026 and beyond. Suggested ReadingWhy This Under-the-Radar AAA Title Is More Than Just A Far Cry Clone Share SubtitlesOffEnglishview videoSuggested ReadingWhy This Under-the-Radar AAA Title Is More Than Just A Far Cry Clone Share SubtitlesOffEnglish“After a thorough review of its pipeline that took place from October to December, the Group decided to provide additional development time to some of its biggest productions in order to create the best conditions for success,” Ubisoft announced in a press release on Wednesday. “This decision has already been beneficial to the quality of Assassin’s Creed Shadows. As a consequence, FY2026-27 and FY2027-28 will see significant growth vs. FY2025-26 on the back of strong content coming from the Group’s largest brands.”While the company didn’t call out its “biggest productions” by name, it’s likely safe to assume it’s talking about entries in the Far Cry, Assassin’s Creed, and Tom Clancy universes. 2025 marks four years since Far Cry 6, with Far Cry 7 well into development but still not officially announced amid the longest break in sequels in the franchise’s recent history (Insider Gaming reports it is slated for late 2026 at the moment). Assassin’s Creed Hexe, meanwhile, has already been teased as the next major entry in the stealth RPG series, but looks unlikely to arrive before 2027 at this rate.Lots of other projects at Ubisoft also remain up in the air, including a Splinter Cell remake and the perennially MIA Beyond Good & Evil 2. We did get some sense of what’s next up in the publisher’s pipeline today, however. Anno 117: Pax Romana, the Prince of Persia: The Sands of Time remake, Rainbow Six Mobile and The Division Resurgence are all expected to come out by March 31, 2026, alongside other still-unannounced projects. That leaves a 12-month gap with no major heavy hitters for one of the biggest publishers in the world, followed by a potential onslaught of tentpole releases. In the meantime, upheaval at Ubisoft will continue as it “crystalizes the true economic value” of a new Tencent-funded shell company containing its most valuable assets and an additional $100-million cost-cutting program that will likely result in additional layoffs over the next two years. .

Senior Level Artist - remotePeople Can FlyMontréal QC ca2 hours agoApplyJob DescriptionCréation de la géométrie et de la composition du monde pour les niveaux dans Unreal Engine.Création efficace de niveaux jouables en fonction de la vision du directeur ou de la directrice artistique et des besoins liés à la conception des niveaux, en collaboration avec le concepteur ou la conceptrice principal(e).Réaménagement d’éléments existants dans de nouvelles formes.Blocage des modules et des niveaux, création des premiers matériaux.Maintien des pipelines de production et des flux de travail dans les délais impartis.Au moins 3 ans d’expérience dans la création de la géométrie de monde pour les niveaux.Expérience de travail avec Unreal Engine 5.Capacité à utiliser quotidiennement des nuanceurs en PBR et la cartographie normales.Expérience dans la création d’esthétiques créatives et visuelles de haute qualité et connaissances techniques en la matière.Capacité à s’adapter rapidement à un style artistique en particulier.Capacité à s’adapter à partir de directives et à améliorer de nouveaux styles, techniques et flux de travail.Compréhension de base de la théorie des couleurs.Bonnes compétences en communication écrite et orale en anglais.Esprit d’équipe et ouverture.Forte éthique de travail et motivation personnelle.Passion pour les jeux vidéo et leur création.Creation of world geometry and composition for levels in Unreal Engine.Effective creation of playable levels based on the art director's vision and level design needs, in collaboration with the lead designer.Rearrange existing elements into new forms.Blocking of modules and levels, creation of initial materials.Maintain production pipelines and workflows on schedule.At least 3 years' experience creating world geometry for levels.Experience working with Unreal Engine 5.Ability to use PBR shaders and normal mapping on a daily basis.Experience and technical knowledge of creating high-quality creative and visual aesthetics.Ability to adapt quickly to a particular artistic style.Ability to adapt from guidelines and improve new styles, techniques and workflows.Basic understanding of color theory.Good written and oral communication skills in English.Team spirit and openness.Strong work ethic and self-motivation.Passion for video games and their creation.QualificationsPortefeuille solide présentant des œuvres d'art de niveau AAA réalisées dans un moteur (Unreal, Unity ou moteur propriétaire).Excellente connaissance des logiciels standard de modélisation et de sculpture (Maya, Max, ZBrush, Photoshop ou leurs équivalents).Solides compétences en matière de modélisation, de sculpture et de texturation.Interpréter artistiquement des concepts ou pousser la créativité au-delà de l'idée initiale.Solide compréhension des compétences artistiques traditionnelles et des compétences fondamentales en matière de conception (forme, texture, valeur et couleur).Connaissance avancée des flux de travail et des pratiques actuelles en matière d'art de niveau.Expérience de travail avec Unreal Engine.Compréhension totale des flux de travail modulaires.Compétence dans l'utilisation quotidienne des shaders PBR et des cartes normales.Expérience des logiciels standard de modélisation 3D.Compétence en matière d'équilibre entre la qualité artistique et les performances du moteur de jeu.Démonstration de l'habillage de décors narratifs.Capacité à apprendre rapidement et à s'adapter à de nouveaux styles, techniques et flux de travail avec des conseils de base.Capacité à collaborer avec les chefs de département et les différents départements pour résoudre les défis artistiques et techniques.Solides compétences en communication orale et écrite en anglais.Bonnes compétences en matière de mentorat.Strong Portfolio showcasing AAA quality level art done in engine (Unreal, Unity or Proprietary Engine).Excellent knowledge of industry-standard software packages for modeling, high-poly sculpting (Maya, Max, ZBrush, Photoshop, or their equivalents).Strong modeling, sculpting and texturing skills.Artistically interpret concepts or creatively push beyond an initial ideation.A solid understanding of traditional art skills along with foundational design competencies (shape, form, texture, value and color).Advanced knowledge of current level art workflows and practices.Experience working with Unreal Engine.Full understanding of Modular Workflows.Skilled in using PBR shaders and normal maps on a daily basis.Experience with industry-standard software packages for 3D modeling.Skilled at balancing art quality and performance in the game engine.Demonstration of narrative set dressing.Adaptive.Ability to quickly learn and adapt to new styles, techniques and workflows with basic guidance.Ability to collaborate with department leads, and different departments to solve artistic and technical challenges.Strong verbal and written communication skills in English.Good mentoring skills.Additional InformationAu CanadaUn ensemble d’avantages sociaux payés à 100 % par PCF. La compagnie d’assurance rembourse les réclamations à 100 % (jusqu’à 500 $ de services par année, en plus d’une couverture familiale ou individuelle).Une couverture dentaire complète, y compris les soins dentaires majeurs et l’orthodontie.La cotisation égale au REER à hauteur de 4 % avant déductions fiscales, 100 % acquis dès le premier jour.Une semaine de congé payé pendant les vacances d’hiver.20 jours de congé payé et 5 jours de congé de maladie payé.Des séances gratuites de soins virtuels de santé et de bien-être mental, incluses dans le régime pour les membres et leurs personnes à charge.Un salaire compétitifDes occasions de développement personnel et la possibilité d’œuvrer dans un environnement mondial.L’occasion de travailler au sein d’une équipe créative avec des personnes passionnées par ce qu’elles font.What we offer:CanadaBenefit package 100% paid by PCF. Insurance company reimburses 100% of claims (Up to $500 per service a year, as well as individual family coverage).Full Dental coverage, including major dental and orthodontics.4% RRSP matching before tax deductions, 100% vested on day 1.Paid week off during Winter Holidays.20 paid vacation days and 5 paid sick days.Free virtual health and mental wellbeing sessions included in the plan for members and their dependents.A competitive salaryPersonal development opportunities and ability to work in a global environment.Work in a creative team with people full of passion for what they do.Company DescriptionPeople Can Fly, un des principaux studios indépendants de développement de jeux AAA, est composé d’une équipe internationale de centaines de personnes talentueuses travaillant depuis des bureaux situés en Pologne, au Royaume-Uni, aux États-Unis et au Canada, mais aussi de partout dans le monde grâce à nos programmes de télétravail.Fondée en 2002, PCF a marqué le genre des jeux de tir avec des titres comme Painkiller, Bulletstorm, Gears of War: Judgment et Outriders. Nous sommes l’un des studios Unreal Engine les plus expérimentés de l’industrie et nous renforçons nos capacités au moyen de solutions internes, à savoir l’infrastructure PCF Framework.Nos équipes créatives travaillent actuellement sur plusieurs titres passionnants : Gemini est notre nouveau projet en développement avec Square Enix ; Maverick est un jeu Triple-A développé en collaboration avec Microsoft Corporation ; Bifrost et Victoria sont des projets que nous développons en modèle d'auto-édition. Nous travaillons également sur un projet de réalité virtuelle et d'autres projets non dévoilés, plus d'informations à venir plus tard.Forte de plus de 20 ans d’expérience, PCF se lance à la découverte de nouveaux horizons. Nous avons pour but de combiner notre expertise avec la créativité des meilleurs visionnaires de l’industrie afin de travailler collectivement à la nouvelle génération de jeux d’action destinés à la communauté mondiale de joueurs.Si vous choisissez d’entreprendre cette aventure avec nous, vous aurez l’occasion de perfectionner votre art et d’approfondir vos connaissances, en travaillant aux côtés de chefs de file de l’industrie pour offrir une expérience unique et toute nouvelle aux joueurs du monde entier._____People Can Fly is one of the leading independent AAA games development studios with an international team of hundreds of talented individuals working from offices located in Poland, UK, US, and Canada, and from all over the world thanks to our remote work programs.Founded in 2002, we made our mark on the shooter genre with titles such as Painkiller, Bulletstorm, Gears of War: Judgment, and Outriders. We are one of the most experienced Unreal Engine studios in the industry and we are expanding it with in-house solutions called PCF Framework.Our creative teams are currently working on several exciting titles: Gemini is our new project being developed with Square Enix; Maverick is a Triple-A game developed in collaboration with Microsoft Corporation; Bifrost & Victoria are projects we're growing in the self-publishing model. We are also busy working on a VR and undisclosed projects, more information on those to come later.With over 20 years of experience, PCF sets out to explore new horizons. We aim to combine our expertise with creativity of the best and most forward-thinking talents in the industry to work together on the new generation of action games for the global gaming community.If you decide to accompany us on this journey, you’ll have a chance to perfect your craft and expand your knowledge, working alongside leaders in the industry on bringing a brand-new unique experience to the players worldwide. Create Your Profile — Game companies can contact you with their relevant job openings. Apply

RT NVIDIA AI PC𝘾𝙤𝙣𝙩𝙧𝙤𝙡 + 𝘼𝙡𝙩 + 𝘾𝙧𝙚𝙖𝙩𝙚 🎛️🧠@DonDiablo’s latest music video for his song BLACKOUT was created using a generative AI hybrid workflow powered by a GeForce RTX 5090 GPU. Watch + learn in #RTXAIGarage 👉 https://nvda.ws/42Wf8Fq

If you’ve stayed in a bath or pool long enough, you’ve likely noticed your skin, especially your fingers, become wrinkly or pruny. This is caused by your blood vessels contracting. When blood vessels narrow, the skin’s area is reduced, and your skin forms wrinkles. While your pruny fingers may seem off-putting, this evolutionary advantage actually helps you grip things easier underwater, compared to non-pruny hands. However, as your fingers wrinkle underwater, do they always wrinkle in the same pattern? That was the question on one young student’s mind. And their question inspired a new study, published in the Journal of the Mechanical Behavior of Biomedical Materials. Why Our Skin Gets WrinklyImages of topographical skin wrinkles overlaid on each other from the two different time points, 24 h apart, with opacity levels of (A) 0%, (B) 50%, (C) 100%. (Image Credit: Guy German) Guy German, an associate professor at Binghamton University and a faculty member at the Thomas J. Watson College of Engineering and Applied Science’s Department of Biomedical Engineering had published several studies on human skin, including a 2020 study that focuses on what happens to the skin when it’s exposed to water for long periods of time. In a 2023 article with The Conversation, German answered a young reader's question about why our skin gets wrinkly in water. The answer to that question, however, sparked another. “A student asked, ‘Yeah, but do the wrinkles always form in the same way?’ And I thought: I haven’t the foggiest clue!” said German in a press release. “So it led to this research to find out.”Read More: A Damaging Hand Disease Has Neanderthal RootsSame Wrinkles on Pruny Fingers  Representative wrinkle pairs established and numbered across (A) Day 0 and (B) 24+ hr later on a subject's ring finger. Black curves denote clearly paired wrinkle morphologies. Red curves denote inconsistent wrinkles between the two time points. (For interpretation of the references to colour in this figure legend, the reader is referred to the Web version of this article.) (Image Credit: Guy German)To answer the question, German and Rachel Laytin, a biomedical engineering graduate from Binghamton University, submerged research participants' fingers in water for 30 minutes and took photos of the pruny fingers. After 24 hours, they repeated the same process with the same participants. After analyzing the photos, the research team found that the participant has the same wrinkly pattern after each submersion. “Blood vessels don’t change their position much — they move around a bit, but in relation to other blood vessels, they’re pretty static,” German said in a press release. “That means the wrinkles should form in the same manner, and we proved that they do.”During this study, German and Laytin made an additional discovery. “We’ve heard that wrinkles don’t form in people who have median nerve damage in their fingers,” German said in a press release. “One of my students told us, ‘I’ve got median nerve damage in my fingers.’ So we tested him — no wrinkles!”Advancing Forensic ScienceWith this new research, not only does that student have an answer to their question, but there may be a way to use this data to help assist in forensic science. This new advancement could help forensic teams better identify crime scene fingerprints and bodies with prolonged water exposure. German can relate to the possibility of these new methods, as his father is a retired police officer in the U.K. who experienced some of these forensic setbacks. “Biometrics and fingerprints are built into my brain,” he said in a press release. “I always think about this sort of stuff, because it’s fascinating.”German is excited to continue his research and answer more of his students’ questions. “I feel like a kid in a candy store, because there’s so much science here that I don’t know. We thank the people at The Conversation and the wonderful question they asked us, because it does create cool new science,” German concluded in a press release. This article is not offering medical advice and should be used for informational purposes only.Read More: Why Do We Have Fingerprints?Article SourcesOur writers at Discovermagazine.com use peer-reviewed studies and high-quality sources for our articles, and our editors review for scientific accuracy and editorial standards. Review the sources used below for this article:Journal of the Mechanical Behavior of Biomedical Materials. On the repeatability of wrinkling topography patterns in the fingers of water immersed human skinExtreme Mechanics Letters. Mechanical, compositional, and microstructural changes caused by human skin macerationAs the marketing coordinator at Discover Magazine, Stephanie Edwards interacts with readers across Discover's social media channels and writes digital content. Offline, she is a contract lecturer in English & Cultural Studies at Lakehead University, teaching courses on everything from professional communication to Taylor Swift, and received her graduate degrees in the same department from McMaster University. You can find more of her science writing in Lab Manager and her short fiction in anthologies and literary magazine across the horror genre.

Lapland Central Hospital Extension / Verstas ArchitectsSave this picture!© Niclas Mäkelä•Rovaniemi, Finland Architects: Verstas Architects Area Area of this architecture project Area:  52000 m² Year Completion year of this architecture project Year:  2023 Photographs Photographs:Niclas Mäkelä Lead Architects: Riina Palva, Jussi Palva, Väinö Nikkilä More SpecsLess Specs Save this picture! Text description provided by the architects. Originally completed in the 1980s, the Lapland Central Hospital was designed by one of Finland's most renowned hospital architects, Reino Koivula. The extensions double the current premises of the hospital, including e.g., emergency outpatient clinic, observation ward, operation rooms, psychiatric wards, and facilities for several supporting functions. The extensions create conditions for extensive operations synergy, transforming the hospital campus into a more integrated and efficient whole.Save this picture!Save this picture!Save this picture!The new section of the hot hospital and parking building is positioned to the east of the existing hospital. Wooden surfaces under the entrance canopy and interior create a warm and inviting atmosphere. The psychiatric hospital is located west of the central hospital building. Its patient wards and outdoor yard are situated next to a small forest, bringing nature into the treatment process. The patient is at the center, with good customer experience, safety, effective treatment, and staff well-being as key goals.Save this picture!The architecture of the new buildings is consistent and recognizable. The primary facade material for the psychiatric hospital is wood with a translucent finish, while the facades of the hot hospital feature white concrete and aluminum. Vertical lamellas and floor-level overhangs provide weather protection and reduce the need for mechanical cooling. The modular facade solution allows for the flexible placement of spaces of various sizes along the facade, thereby supporting the adaptability of the interior.Save this picture!Save this picture!Save this picture!Save this picture!The revised layout of the hospital campus is based on creating a new east-west oriented internal connection, which serves as the spatial and functional unifying backbone of the plan. The architecture conveys humaneness and patient-centered care, while evoking images of Lapland's nature. Materials, spatial solutions, and natural light create a safe, healing environment that serves both patients' effective treatment and the well-being of the staff. Public artworks implemented on the hospital campus enliven the surroundings and promote active rehabilitation.Save this picture! Project gallerySee allShow less Project locationAddress:Rovaniemi, FinlandLocation to be used only as a reference. It could indicate city/country but not exact address.About this officeVerstas ArchitectsOffice••• MaterialsWoodSteelMaterials and TagsPublished on May 15, 2025Cite: "Lapland Central Hospital Extension / Verstas Architects" 15 May 2025. ArchDaily. Accessed . <https://www.archdaily.com/1029946/lapland-central-hospital-extension-verstas-architects&gt ISSN 0719-8884Save世界上最受欢迎的建筑网站现已推出你的母语版本!想浏览ArchDaily中国吗?是否 You've started following your first account!Did you know?You'll now receive updates based on what you follow! Personalize your stream and start following your favorite authors, offices and users.Go to my stream

In context: Google researcher and reverse engineer "LaurieWired" recently posed a thought-provoking thread on X: What would happen after a CPU manufacturing apocalypse? How would the tech world respond to a future without newer, faster processors? Programming and optimization legend John Carmack offered an equally compelling answer. LaurieWired proposes the idea of a "Zero Tape-out Day" (Z-Day), an event causing manufacturers to stop producing new silicon designs. Considering the existing supply, the researcher predicts skyrocketing computer prices, stalled cloud capacity, and a ticking clock on electromigration slowly degrading the most advanced chips built on smaller nodes – all within the first year after Z-Day. Conditions would deteriorate even further in the following years, with a booming black market for processors and Xeon CPUs valued more than gold. Computing technology could regress by decades as older systems built on larger nodes prove far more resilient to electromigration. People would mod classic processors like the Motorola 68000 to operate for thousands of years without significant gate wear. More advanced systems – such as the iMac G3s sold between 1998 and 2003 – would become workstations for the elite, while the proles use repurpose hardware from Gameboys, Macintosh SEs, and Commodore 64s. LaurieWired suggests that 30 years after Z-Day, the world would become a dystopia where computing resembles the 1970s or 1980s. The modern internet would vanish, replaced by sneakernet data exchanges on SSDs and efforts to safeguard valuable desktop hardware from confiscation. Former id Software developer John Carmack decided to weigh in on the thought experiment. Having created the Doom graphics engine in just 28 hours on "vintage hardware," his expertise provided some perspective. Carmack said that a significant part of the modern world could run on outdated hardware if software optimization were a priority for developers. // Related Stories The god-tier coder suggests that developers could transition all interpreted, microservice-based products to monolithic, native codebases. Programmers would abandon modern development patterns and seek more efficient approaches, such as those used during earlier computing eras when there was no internet to push patches. Such a paradigm reset would force post-apocalyptic coders to make ancient hardware hum through software optimization. Carmack also acknowledges that innovative new products would become much rarer without ultra-cheap and scalable computing. While framed within the context of LaurieWired's thought experiment, Carmack's ideas hold practical relevance in today's computing landscape. For example, would Microsoft still impose strict hardware requirements if it prioritized optimizing Windows 11? It's a question worth considering. Similarly, how much could the gaming industry benefit from better optimization?