Decoding Peaks at 70 Tok/s with Predictable Context.** Apple's Neural Engine on the iPhone 18 is capable of processing 70 tokens per second (tok/s) with a decoding speed of 15-60 tps. This performance is achieved with a 24K context size and both Key/Value pairs stored in integer format (int8). The Apple Neural Engine is noted as fast, indicated by the emoji "🚀." Qwen 3.8-27B on iPhone 18 Apple Neural Engine.** This model runs on the iPhone 18 with the Apple Neural Engine, achieving a decoding speed of 15-60…
Key points
Decoding Peaks at 70 Tok/s with Predictable Context.**
Apple's Neural Engine on the iPh…: Decoding Peaks at 70 Tok/s with Predictable Context.**
Apple's Neural Engine on the iPhone 18 is capable of processing 70 tokens per second (tok/s) with a decoding speed of 15-60 tps.
This performance is achieved with a 24K context size and both Key/Value pairs stored in i…: This performance is achieved with a 24K context size and both Key/Value pairs stored in integer format (int8).
The Apple Neural Engine is noted as fast, indicated by the emoji "🚀."
**Qwen 3.8-27B on…: The Apple Neural Engine is noted as fast, indicated by the emoji "🚀."
**Qwen 3.8-27B on iPhone 18 Apple Neural Engine.**
This model runs on the iPhone 18 with the Apple Neural Engine, achieving a decoding speed of 15-60 tps.
Sources checked by our analysis pipeline: 1
Anemll on X: "Confirmed 24K context on iPhone with both K/V on int8. Delta net stays FP16. Decoding also peaks at 70 tok/s with easy to predict context.
Apple Neural Engine on A20 is fast 🚀"
Confirmed 24K context on iPhone with both K/V on int8. Delta net stays FP16. Decoding also peaks at 70 tok/s with easy to predict context.
Apple Neural Engine on A20 is fast 🚀
Qwen 3.8- 27B on iPhone 18 Apple Neural Engine.
Added draft model , dflash-2 at ~ 2.8 bpw ( mixed 2-4 vector quantization with small drop in acceptance rate)
Base is @PrismML Bonsai-2, 1.585 bpw requantized to 1.5 bpw ( VQ 6 bits per 4 weights)
Decoding speed is 15-60 tps.
Confirmed 24K context on iPhone with both K/V on int8. Delta net stays FP16. Decoding also peaks at 70 tok/s with easy to predict context.
Apple Neural Engine on A20 is fast 🚀
Qwen 3.8- 27B on iPhone 18 Apple Neural Engine.
Added draft model , dflash-2 at ~ 2.8 bpw ( mixed 2-4 vector quantization with small drop in acceptance rate)
Base is @PrismML Bonsai-2, 1.585 bpw requantized to 1.5 bpw ( VQ 6 bits per 4 weights)
TikTok has permanently banned the social media account for Alex Gibney's Elon Musk documentary, following similar bans by YouTube, Meta, and X. The ban comes after TikTok faced backlash for refusing to allow ads for the film, but the platform has not responded to a request for comment.
Four individuals were killed when part of a chimney collapsed at a power plant in Asansol, West Bengal, following an incident that occurred after several workers managed to escape. The cause of the collapse is under investigation.
MapQuest, a popular navigation app, has integrated a surveillance layer into its mobile app. This feature now displays license plate readers and cameras from 10+ companies, including Flock, using data from OpenStreetMap. This addition aims to enhance the user's driving experience by providing real-time traffic updates and safety information.