Books • Learning • Self Growth • Wisdom
Joined July 2022
- Tweets28.9K
- Following24
- Followers193K
- Likes64.7K
Library Path retweeted
This guy used Ray-Ban Meta Smart Glasses, AI, and a MacBook to cheat at chess game.
"When people get used to preferential treatment, equal treatment seems like discrimination."
~ Thomas Sowell
I wish more AI projects would let developers look behind the scenes instead of just showing off big numbers.
That's what caught my attention about EAGLE-4, because beyond the impressive results in its launch paper, there's a GitHub repository where developers can actually explore the implementation and try it for themselves.
I'm genuinely curious to see how well it holds up when people start testing it on their own workloads.
I'd love to hear what developers think about this. Take a look at the project and share your thoughts: github.com/eagle-hq/EAGLE-4
EAGLE-4 is here.
From EAGLE-1 to EAGLE-3, we’ve kept making inference faster.
EAGLE-4 is our fastest yet.
961.6 tokens/s on @Alibaba_Qwen’s Qwen3.8-27B with TokenSpeed (BF16, 4× @NVIDIA B300, TP4, one request, T=0).
Github: github.com/eagle-hq/EAGLE-4
Three AI inference methods went head-to-head, and EAGLE-4 came out ahead in the reported test.
Its launch paper shows 961.6 tokens per second for EAGLE-4, compared to 672.6 for DFlash 2 and 618.6 for DSpark on four NVIDIA B300 GPUs.
That's roughly 1.43x and 1.55x their throughput, respectively, showing what its improved token prediction method can deliver under the same benchmark conditions.
The numbers are worth a closer look. Explore the full benchmarks and GitHub repository: github.com/eagle-hq/EAGLE-4
EAGLE-4 is here.
From EAGLE-1 to EAGLE-3, we’ve kept making inference faster.
EAGLE-4 is our fastest yet.
961.6 tokens/s on @Alibaba_Qwen’s Qwen3.8-27B with TokenSpeed (BF16, 4× @NVIDIA B300, TP4, one request, T=0).
Github: github.com/eagle-hq/EAGLE-4
Everyone wants better AI models, but what about fixing the technology running them.
EAGLE-4 is taking a different approach with EAGLE-Harness, which adjusts how tokens are predicted based on the hardware and workload, while its Efficient LM Head cuts draft-model overhead by over 40%.
Instead of just throwing more computing power at the problem, it's finding smarter ways to use what's already there.
Curious about the engineering behind it? Dive into EAGLE-4 and see how it works: github.com/eagle-hq/EAGLE-4
EAGLE-4 is here.
From EAGLE-1 to EAGLE-3, we’ve kept making inference faster.
EAGLE-4 is our fastest yet.
961.6 tokens/s on @Alibaba_Qwen’s Qwen3.8-27B with TokenSpeed (BF16, 4× @NVIDIA B300, TP4, one request, T=0).
Github: github.com/eagle-hq/EAGLE-4
If AI models are the rock stars, the engineers building the infrastructure deserve a standing ovation.
EAGLE-4 just gave us another reason to be bullish, with its launch paper reporting 4x the throughput of regular decoding in its tested setup.
The next wave of AI could get very interesting when improvements like this reach more developers.
See what EAGLE-4 is building: github.com/eagle-hq/EAGLE-4
EAGLE-4 is here.
From EAGLE-1 to EAGLE-3, we’ve kept making inference faster.
EAGLE-4 is our fastest yet.
961.6 tokens/s on @Alibaba_Qwen’s Qwen3.8-27B with TokenSpeed (BF16, 4× @NVIDIA B300, TP4, one request, T=0).
Github: github.com/eagle-hq/EAGLE-4
Could voice become a more natural way to interact with AI?
Imagine remembering an email you need to send or a meeting you need to schedule.
You say what you need, and SpeakON’s personal AI agent can help with supported everyday tasks like these.
That’s an interesting direction for personal AI.
we crushed a keyboard.
its retirement party got a little out of hand.
for years, it was there for every email, every plan, every “just one more thing.” a loyal colleague. a very loud colleague.
but your ideas move faster than your fingers.
and somewhere between knowing what you want to do and typing it all out, a quick task becomes another thing taking up your afternoon.
meet SpeakON AI Agent.
start with what you need, in your own words. less time behind a blinking cursor. more room to move things forward.
the keyboard had a good run.
now, let’s give your voice a turn.
THE LAST KEYSTROKE is here. watch the film, then tell us: what’s the first task you’d hand over?
speakon.app/x
Disclaimer: The calling feature is still in beta and will be available soon. All other Agent features are available now, with ongoing improvements.
