Joined December 2017
- Tweets 9,067
- Following 2,539
- Followers 1,730
- Likes 34,453
609 Photos and videos
Jun 2
Radeon Ai Pro R9600D
✅ Single slot 👀
✅ 32GB 👀
⚠️ Passive cooling
⚠️ 640 GB/s
49
Jun 2
Looks like it's Holo yolo time.
Jun 2
🌞This is big Local AI news! A new open-source Computer-Use LLM has just launched.
Holo 3.1 is H Company’s (🇫🇷) new local computer-use agent model that beats Qwen3.5-397B, Kimi-K2.5, and Sonnet 4.6!
Since it is built for local deployment →
⬩ Runs fully on your machine (MacBook, Windows PC, DGX Spark, RTX Spark)
⬩ Based on Qwen architecture, specialized for GUI understanding & computer control
⬩ Optimized checkpoints: NVFP4, FP8 & Q4 GGUF (0.8B to 35B sizes)
⬩ Strong gains: 79.3% on AndroidWorld benchmark (35B model)
💻 Comparison to Qwen3.5:
Holo 3.1 is fine-tuned specifically for computer-use agents (screen understanding, planning, clicking, navigation). Better at real GUI tasks than general-purpose Qwen3.5, especially when running locally.⚡
24
May 20
What sets Qwen3.7-Max apart is the breadth and depth of its agent capabilities. It excels as a coding agent, from frontend prototyping to complex multi-file engineering.
..weeks of specialized team effort can now be completed end-to-end within hours.
qwen.ai/blog?id=qwen3.7
76
May 19
Let's go!
LLMs coming this week :
> Qwen3.7 max ✅
> Qwen3.7 27b/35b
> Minimax M3.0
> Gemini 3.5 Pro/Flash
> GPT-5.6
> Sonnet-4.8❓
> Kimi/GLM ❓
It’s a big week.
41
May 12
Check out this incredible project from @brrrkyle.
This is for anyone interested in GPU programming or performance engineering.
brrrviz.com/
this is how i wish i learned GPU fundamentals
not a lengthy textbook. not a static image. every concept is an interactive visualization.
covering the SM architecture, memory coalescing, synchronization, and more.
what concepts do you want to see next?
brrrviz.com
1
1
77
May 12
Ultra-large models can run on a single PCIe card.
Powered by six HTX301 chips and 384GB of memory, enterprises can now execute 700B-parameter LLM inference locally at just ~240W.
Source:
skymizer.ai/skymizer-announc…
43
May 6
Local level up
Today we're releasing ZAYA1-8B, a reasoning MoE trained on @AMD and optimized for intelligence density.
With <1B active params, it outperforms open-weight models many times its size on math and reasoning, closing in on DeepSeek-V3.2 and GPT-5-High with test-time compute. 🧵
56
Apr 15
Got m.2, want 20b?
What Unigen has done is taken the SAKURA-II AI accelerator and put it on an E1.S board, along with an impressive memory capacity of up to 32 GB. The module is available in both 16 GB and 32 GB flavors, offering up to 68 GB/s bandwidth.
wccftech.com/turn-your-vacan…
64
Apr 14
Ngl, following AI stuff is exciting.
Apr 14
Introducing Kernels on the Hugging Face Hub ✨
What if shipping a GPU kernel was as easy as pushing a model?
- Pre-compiled for your exact GPU, PyTorch & OS
- Multiple kernel versions coexist in one process
- torch.compile compatible
- 1.7x–2.5x speedups over PyTorch baselines
56
Apr 12
Some people just say the craziest things..
39
Xandar retweeted
💯Are you aware? Charli3's PULL oracle manages up to 1ms operations for requested data. SCALABLE and CHEAP, and decentralized.
🔔Pull oracles function by a call, on demand, for data. This call is made by the user. Perfect for liquidations, payouts, and in the moment betting. Very scalable to be used by the masses
It is very useful for a myriad of operations.
11
79
2,126
Xandar retweeted
🔔Dont be fooled! Up to 1ms updates are already possible with Charli3. It is not new. We do this already with our ODV Pull oracle.
The final attestation of the transaction takes 1 block but the data is usable for the transaction at Up to 1ms.
2
13
87
2,342
Xandar retweeted
💎Charli3 Built the first partnerchain for Cardano.
Built for RealFi applications and real legacy industries.
@BikeIDNumber and @palmeconomy are both interested in using it, among others, for mass data at fixed costs.
✨️Many beleive Midnight to be the first
💎But with the help of @txpipe_tools , our partnerchain has been live since many months ago
📯we also have the only direct transaction bridge back to Cardano.
We plan to create SDKs for EVM networks as well as with Midnight with their seperate infrastructure.
There is still much to be done, and we work towards those goals.
5
24
96
2,196
Xandar retweeted
$NIGHT market is powered by Charli3. The asset is volatile, which is why @FluidTokens has chosen Oracle security to keep its users safe.
10 Dec 2025
JUST IN 🚨
NIGHT market is the second most used on FluidTokens
Lenders, want to get 7% on your $ADA?
Borrowers, do you want to unlock liquidity with your $NIGHT?
app.fluidtokens.com/ada/mark…
2
5
89
2,866
Xandar retweeted
💯NIGHT/ADA oracle is live. $NIGHT
Will you use it ? Watch out on @FluidTokens tomorrow for the lending pools.
4
19
100
2,217
Xandar retweeted
📯NIGHT/ADA oracle feed coming today📯 $NIGHT
ALL sources, CEX and DEX, the most accurate and decentralized price reporting of NIGHT
Available on @FluidTokens soon after.
1
10
99
1,705
Xandar retweeted
💎Oracles Charli3 runs on Cardano.
1. Push Oracle = consistent asset prices updated with price swings and a set time interval. All information on chain, fully auditable. @FluidTokens uses these
2. Pull Oracle = manually triggered by the user. Asset price update in the moment, for bet and liquidation settlements, getting the most accurate data possible on demand. Cheaper, faster, and the better option.
3. Proof of Reserve = for real world assets, tying in to real world banks and institutional holdings to create tokenized management and transparency of assets. @USDMOfficial uses this.
5
13
79
2,256
Xandar retweeted
27 Nov 2025
1
38
Xandar retweeted
💎Pull Based Oracle cost reduction! and security increase!
📚Pull Oracles are the standard, allowing off chain signatures for user consumption.
💚We have just upgraded our pull oracle to reduce costs by ~50% while also increasing security and speed of the data.
1
7
32
1,049