Connect with us

Artificial Intelligence

Tencent releases versatile open-source Hunyuan AI models

Published

on

Tencent has expanded its family of open-source Hunyuan AI models that are versatile enough for broad use. This new family of models is engineered to deliver powerful performance across computational environments, from small edge devices to demanding, high-concurrency production systems.

The release includes a comprehensive set of pre-trained and instruction-tuned models available on the developer platform Hugging Face. The models come in several sizes, specifically with parameter scales of 0.5B, 1.8B, 4B, and 7B, providing substantial flexibility for developers and businesses.

Tencent has indicated that these models were developed using training strategies similar to its more powerful Hunyuan-A13B model, allowing them to inherit its performance characteristics. This approach enables users to select the optimal model for their needs, whether it’s a smaller variant for resource-constrained edge computing or a larger model for high-throughput production workloads, all while ensuring strong capabilities.

One of the most notable features of the Hunyuan series is its native support for an ultra-long 256K context window. This allows the models to handle and maintain stable performance on long-text tasks, a vital capability for complex document analysis, extended conversations, and in-depth content generation. The models support what Tencent calls “hybrid reasoning,” which allows for both fast and slow thinking modes that users can choose between depending on their specific requirements.

The company has also placed a strong emphasis on agentic capabilities. The models have been optimised for agent-based tasks and have demonstrated leading results on established benchmarks such as BFCL-v3, τ-Bench, and C3-Bench, suggesting a high degree of proficiency in complex, multi-step problem-solving. For instance, on the C3-Bench, the Hunyuan-7B-Instruct model achieves a score of 68.5, while the Hunyuan-4B-Instruct model scores 64.3.

The series’ performance is a focus on efficient inference. Tencent’s Hunyuan models utilise Grouped Query Attention (GQA), a technique known for improving processing speed and reducing computational overhead. This efficiency is further enhanced by advanced quantisation support, a key element of the Hunyuan architecture designed to lower deployment barriers.

Tencent has developed its own compression toolset, AngleSlim, to create a more user-friendly and effective model compression solution. Using this tool, the company offers two main types of quantisation for the Hunyuan series.

The first is FP8 static quantisation, which employs an 8-bit floating-point format. This method uses a small amount of calibration data to pre-determine the quantisation scale without requiring full retraining, converting model weights and activation values into the FP8 format to boost inference efficiency.

The second method is INT4 quantisation, which achieves W4A16 quantisation through the GPTQ and AWQ algorithms:

  • The GPTQ approach processes model weights layer by layer, using calibration data to minimise errors in the quantised weights. This process avoids requiring model retraining and improves inference speed.
  • The AWQ algorithm works by statistically analysing the amplitude of activation values from a small set of calibration data. It then calculates a scaling coefficient for each weight channel, which expands the numerical range of important weights to retain more information during the compression process. 

Developers can either use the AngleSlim tool themselves or download the pre-quantised models directly.

Performance benchmarks confirm the strong capabilities of the Tencent Hunyuan models across a range of tasks. The pre-trained Hunyuan-7B model, for example, achieves a score of 79.82 on the MMLU benchmark, 88.25 on GSM8K, and 74.85 on the MATH benchmark, demonstrating solid reasoning and mathematical skills.

The instruction-tuned variants show impressive results in specialised areas. In mathematics, the Hunyuan-7B-Instruct model scores 81.1 on the AIME 2024 benchmark, while the 4B version scores 78.3. In science, the 7B model reaches 76.5 on OlympiadBench, and in coding, it scores 42 on Livecodebench.

The quantisation benchmarks show minimal performance degradation. On the DROP benchmark, the Hunyuan-7B-Instruct model scores 85.9 in its base B16 format, 86.0 with FP8, and 85.7 with Int4 GPTQ, indicating that efficiency gains do not come at a cost to accuracy.

For deployment, Tencent recommends using established frameworks like TensorRT-LLM, vLLM, or SGLang to serve the Hunyuan models and create OpenAI-compatible API endpoints, ensuring they can be integrated smoothly into existing development workflows. This combination of performance, efficiency, and deployment flexibility positions the Hunyuan series as a continuing powerful contender in open-source AI.

See also: Deep Cogito v2: Open-source AI that hones its reasoning skills

Want to learn more about AI and big data from industry leaders? Check out AI & Big Data Expo taking place in Amsterdam, California, and London. The comprehensive event is co-located with other leading events including Intelligent Automation Conference, BlockX, Digital Transformation Week, and Cyber Security & Cloud Expo.

Explore other upcoming enterprise technology events and webinars powered by TechForge here.

Continue Reading
Click to comment

Leave a Reply

Your email address will not be published. Required fields are marked *

Artificial Intelligence

Sengled’s downfall shows the peril of relying on cloud connections for smart home control

Published

on

After repeatedly leaving customers without smart control of their lights, Sengled has been booted from Amazon’s Works With Alexa program. As first reported by TechHive, beginning August 1st, Sengled’s Alexa skill for controlling its line of LED lights, plugs, switches, and sensors with your voice and routines is no longer available.

In a statement to The Verge explaining the decision, Amazon spokesperson Lauren Raemhild said, “We hold a high bar for the Alexa experience. Sengled has experienced a series of prolonged outages over the past few months that have not been resolved, preventing customers from being able to use Sengled’s Alexa skill to control their light bulbs.”

The future doesn’t look bright for Sengled, which has been silent since the problems started appearing earlier this summer. There appears to have been no communication to customers from the company (Amazon did reach out to its customers about the outages), and no indication of these issues on its website. Repeated attempts by The Verge to contact Sengled have been met with no response.

If your Sengled bulbs were Wi-Fi, you’re out of luck.

There is some good news. If you own Sengled bulbs that use Zigbee, BLE Mesh, or Matter, rather than Wi-Fi, they can still work with Alexa by bypassing Sengled’s spotty servers and connecting to a compatible Echo speaker or Eero Wi-Fi router (this may require setting them up again). Another option is to connect Zigbee bulbs to third-party platform hubs that support the protocol, such as Home Assistant, Hubitat, or the Aeotec SmartThings hub.

But if your bulbs were Sengled’s Wi-Fi ones, you’re out of luck. These won’t connect to Alexa, although they will still work with Sengled’s app, for as long as Sengled’s servers are still running. Users have started reporting problems there, too. All of which goes to show that relying on cloud services to turn your lights on is a fragile solution.

This is a story we’ve seen too often in the smart home. Just last month, Belkin shuttered its WeMo smart home business, and the smart home graveyard is littered with other examples: iHome, Revolv, Staples Connect, Lowes’ Iris, Best Buy’s Insignia, and more.

A common thread with these shutdowns is that the products relied on cloud servers. At one time, it was easier and less expensive for a company to develop a cloud-based controller than a local system, as they don’t require a hub or bridge and can be simpler to set up and use.

However, companies have to maintain those servers, as well as API connections to smart home platforms and voice assistants like Alexa and Google Assistant, which can be costly and resource-intensive. When the business model no longer pans out, history shows us that if they can’t sell it, companies just shut it down.

This brings me to my best piece of advice to anyone buying a smart home device today, especially something as integral as lighting: make sure it has the option of local control. That way, if the company goes under or stops providing the service you signed up for, then your device will still keep working (in some fashion). Plus, locally controlled devices tend to be faster, as they don’t have to wait for a response from a server.

Relying on cloud services to turn your lights on is a fragile solution.

As noted, some Sengled bulbs don’t rely on a cloud connection and instead can work locally in your home. Thanks to a connection via local protocols like Apple’s HomeKit or Zigbee, some products from those companies listed above also still work, even though their servers are gone.

That’s one of the reasons why the new Matter standard is so crucial to the smart home. While it has had its problems, Matter is built on the foundations of HomeKit, Zigbee, and other technologies. It’s an entirely local protocol, communicating to a Matter controller (hub) in your home, not to a company’s cloud.

While Matter ecosystems such as Apple Home, Google Home, Amazon Alexa, and Home Assistant can connect to the cloud to give control when you’re away from home and enable other features like voice assistants, that’s a layer on top of Matter. If a device like a smart bulb supports Matter, either over Wi-Fi or Thread, you don’t need the internet to turn on the lights. And, if the manufacturer’s server dies, your device won’t.

Matter isn’t the only option here. Devices that work over Zigbee, Z-Wave, BLE and BLE Mesh, and local Wi-Fi, also offer local control. But the standardization of Matter, its wide industry support, and its use of non-proprietary IP-based protocols Matter and Thread all broaden its overall compatibility and should make it more futureproof.

The situation with Sengled is just the latest reminder that for a truly reliable smart home, look for local control. While the cloud offers benefits, it should be part of your solution for a smarter home, not the only one.

Developed by Apple, Amazon, Google, Samsung, and others, Matter is an open-sourced, IP-based connectivity software layer for smart home devices. It works over Wi-Fi, Ethernet, and Thread.

Thread is a low-power, wireless mesh protocol. It operates on the same 2.4GHz spectrum as Zigbee and is designed for low-power devices, such as sensors, light bulbs, plugs, and shades. IP-based, Thread devices can communicate directly with each other, the internet, and with other networks using a Thread Border Router.

Today, Matter supports most of the main device types in the home, including lighting, thermostats, locks, robot vacuums, refrigerators, dishwashers, dryers, ovens, smoke alarms, air quality monitors, EV chargers, and more.

A smart home gadget with the Matter logo can be set up and used with any Matter-compatible ecosystem via a Matter controller and controlled by more than one ecosystem with a feature called multi-admin.

Amazon Alexa, Google Home, Samsung SmartThings, Apple Home, Home Assistant, Ikea, and Aqara are among the well-known smart home companies supporting Matter, along with hundreds of device manufacturers.

Follow topics and authors from this story to see more like this in your personalized homepage feed and to receive email updates.


Continue Reading

Artificial Intelligence

IGN hit by layoffs as parent company Ziff Davis cuts costs

Published

on

“The company has told us that the reason for this layoff stems from a Ziff Davis-mandate to cut costs despite several quarters in a row of year-over-year revenue increases, to which IGN Entertainment responded by coming for our members’ jobs,” the guild says. “This is perplexing to us, as we are told again and again that IGN Entertainment has had a tremendously successful year thus far thanks to their hard work.”

An employee in the engineering department was also laid off, Hunter Paniagua, a representative of the Pacific Media Workers Guild, which represents the IGN Creators Guild, tells The Verge.

Ziff Davis didn’t immediately reply to a request for comment. Last year, across all of Ziff Davis, the company employed approximately 3,800 people, according to an SEC filing.

Last year, IGN acquired the website portfolio of Gamer Network, which includes publications like Eurogamer, Gamesindustry.biz, and Rock Paper Shotgun. Terms of that deal weren’t disclosed at the time.

“The company has not responded to the union’s questions about whether its budget for future acquisitions is being reconsidered as a cost-saving measure alongside these other apparently necessary personnel cuts,” the IGN Creators Guild says.

Continue Reading

Artificial Intelligence

Here are the best Apple Watch deals available right now

Published

on

In September, Apple launched its latest smartwatch, introducing the Apple Watch Series 10 alongside a black rendition of the Apple Watch Ultra 2. Each wearable has its own pros and cons, as does the second-gen Apple Watch SE, but the introduction of Apple’s newest wearables also means there are now more Apple Watch models on the market than ever before — and a lot more deals to be had.

But with all of those options, which one should you pick? Generally speaking, you want to buy the newest watch you can afford so that it continues to receive software updates from Apple. The latest update, watchOS 11, launched in September on the Series 6 and newer models, ultimately leaving those still using an older Series 4 or 5 behind. Fortunately, the next iteration of watchOS will continue to support the same devices when it arrives in the fall, even if Apple is revamping its naming convention in a bid to standardize its operating system names.

All that being said, picking up a watch from the latest (or a recent) generation ensures you’re getting a smartwatch with an updated design, a robust number of features, and plenty of sensors.

Now, let’s get to the deals.

The best Apple Watch Series 10 deals

Person looking at music widget on Apple Watch Series 10

The Apple Watch Series 10 represents the latest wearable in Apple’s flagship Series lineup. It features the new S10 SiP chip, faster charging, and a wide-angle OLED display that’s brighter, larger, and easier to read than that of previous models. It weighs half as much as the Ultra, too, and is about 30 percent thinner despite its new 42mm and 46mm sizing. Other upgrades include an added depth and water temperature sensor, as well as a new FDA-cleared sleep apnea detection feature, the latter of which is also available on the last-gen Series 9 and Ultra 2.

While the improvements are welcome, the Series 10 isn’t a vast departure from the prior model, the Series 9. That said, it’s currently on sale at Amazon, Walmart, and Best Buy right now in the 42mm sizing with GPS starting at $299 ($100 off), which is just $20 shy of its recent Prime Day low. The 46mm variant, meanwhile, is on sale at Amazon, Walmart, and Best Buy starting at $329 (also $100 off) — which is the second-best price we’ve seen on the larger configuration — while the LTE models can be had at Amazon, Walmart, and Best Buy starting at around $399 ($100 off).

Read our Apple Watch Series 10 review.

The best Apple Watch SE deals

Woman in floral dress with hand on hip modeling Apple watch SE (2022)Woman in floral dress with hand on hip modeling Apple watch SE (2022)
Woman looking at the app grid on the Apple Watch SE (2022).Woman looking at the app grid on the Apple Watch SE (2022).

The Apple Watch SE received a refresh in late 2022. It has the same chipset as the Series 8, which is great, but with fewer sensors, no always-on display, and a slightly outdated design compared to the Series 8, 9, and 10. Those omissions might take this out of the running for some people, but it still may be exactly what you’re after. Best of all, it starts at $249 for the 40mm Wi-Fi / GPS model, which is $30 less than the previous generation’s baseline cost. Opting for cellular connectivity increases the starting price to $299 for the 40mm size (the 44mm size adds $20 to each configuration).

Right now, the best deal on the latest SE can be found at Amazon, Walmart, and Best Buy, where you can pick up the 40mm model with GPS starting at $169 ($80 off). The 44mm / GPS configuration, meanwhile, is available from the same retailers (Amazon, Walmart, Best Buy) starting at $199 ($80 off). If you want the LTE configuration, the 40mm model is on sale at Amazon, Walmart, and Best Buy starting at $219 (also $80 off), which is about $20 shy of the best price we’ve seen on the entry-level LTE configuration.

Read our Apple Watch SE (second-gen) review.

The best Apple Watch Ultra 2 deals

The Apple Watch Ultra 2 on grass.The Apple Watch Ultra 2 on grass.

The latest Apple Watch Ultra launched at $799 in September with GPS and LTE support, much like the original model. The ultra-capable smartwatch has the most features, sensors, and ruggedness of any Apple Watch model available thus far, along with a display that’s 50 percent brighter than the first Ultra. The 49mm smartwatch also packs Apple’s S9 SiP and second-gen ultra wideband chips, just like the Apple Watch Series 9, while maintaining long-lasting battery life, precise GPS tracking, and a bevy of diving-friendly sensors.

Thanks to Amazon Prime Day, we recently saw the Apple Watch Ultra 2 dip to its lowest price to date in select configurations. Thankfully, it’s still on sale at Amazon with a Trail Loop, an Alpine Loop, or an Ocean Band starting at around $649 ($150 off). You can also pick it up at Best Buy in various configurations for the same price, if you prefer an alternative retailer.

Read our Apple Watch Ultra 2 review.

A note on the more premium models

While all of the Apple Watch models and colorways covered here are encased in aluminum (except the Ultra 2, which has a titanium build), Apple does make a more premium range built out of stainless steel and titanium. These offerings are functionally and aesthetically similar to their aluminum counterparts, with slightly refined colors and finishings — polished for the stainless steel and brushed for the titanium. However, they start at much steeper prices of $699 and above. They, too, can often be found on sale, but they’re never discounted as low as the standard base models, so we don’t include them here.

Follow topics and authors from this story to see more like this in your personalized homepage feed and to receive email updates.


Continue Reading

Trending