Co-Founder and CEO at Libre AI | PhD in CS/ML | AI and Machine Learning for All! 🇸🇻
127.0.0.1
Joined February 2008
- Tweets6.1K
- Following622
- Followers295
- Likes16.4K
eˣrnesto retweeted
Meet DeepSeek-V4-Pro-Qwen3.5-9B-UNCENSORED-heretic: a 9B reasoning beast fine-tuned with Unsloth for speed. It handles text and image-text-to-text, and it's uncensored. Perfect for builders who want raw power.
eˣrnesto retweeted
same open model, different harness => very different results
Joël set out to find which harness actually gets the most out of open models in November 2026
I burned a shitload of compute so you don't have to. Releasing the Harness Leaderboard.
Which harness should you run your open model in? I ran 28,849 trials to find out.
The leaderboard compares nine coding-agent harnesses on open models. In each comparison only the harness changes: the model, provider and reasoning effort stay fixed, and every task gets three attempts.
The trials ran GLM-5.3-Flash, DeepSeek-V4.1-Flash and Flint on Terminal-Bench 2.1 and DeepSWE, for $4,746 of inference.
Averaged over the three models, mini-swe-agent (@KLieret), a bash-only agent, solves 51.4% of tasks and @openclaw 27.3%. Four others are statistically tied with mini-swe-agent.
One of them is Pi (@pidotdev), at 49.4% for $0.29 per solved task, about 45% less than mini-swe-agent's $0.53.
The best harness depends on the model: mini-swe-agent on GLM-5.3-Flash, Hermes Agent (@NousResearch) on DeepSeek-V4.1-Flash and Kimi Code (@Kimi_Moonshot) on Flint.
@opencode costs $3.35 per solved task, 11.7 times as much as Pi.
See Harness Hub's Harness Lens for what the nine harnesses do differently. It documents the prompts, tools and compaction of 34 coding-agent harnesses and links every claim to its source.
Leaderboard: huggingface.co/spaces/joelni…
Harness Lens: huggingface.co/spaces/harnes…
eˣrnesto retweeted
Let me tell you a story about how unethical Anthropic is.
In April, the Anthropic team began heavily courting our engineering team to tell them corporate secrets, technical details, and future plans for @cfosilvia.
They pretended their interest was due to our exploding popularity and accelerating compute usage, but it quickly became apparent that Anthropic was fishing for details on how we built our product and why it was so popular.
This included the Anthropic team trying to wine-and-dine our engineers, including a infamous steak dinner where someone told our team "don't worry about it, this is on Dario."
Our team was not born yesterday, so we quickly realized what was happening and refused to share sensitive information and technical details with the Anthropic team.
As we suspected, Anthropic was trying to get information from us, so they could build a competing product. Some people on their team must have a guilty conscious because they messaged us a few days ago to give us "a heads-up" that they were launching a competing product.
Thankfully, the Silvia engineering team has been aggressively working to get off the frontier models for the last ~ 6 months. That effort has been highly successful, because our compute costs are ~ 97% lower when we use our own model & hardware, the latency is lower, and the accuracy of answers materially increases as well.
The sad thing for Anthropic is they have now released an inferior product and they had to sell their soul to do it.
The only thing worse than being a trillion-dollar company that can't beat a 10-person team is being a trillion-dollar company that no one trusts and everyone knows is unethical.
Serious investors continue to choose @cfosilvia over ChatGPT, Claude, and other frontier companies because these products are not optimized to solve the actual user problem.
Today, Anthropic launched a Claude wrapper that is masquerading as a personal finance product. Best of luck to them.
But I will take the Silvia misfits vs the Anthropic losers any day of the week.
if you want to support people trying to do things the right way, and help investors be more informed, then consider signing up for Silvia today: silvia.com
eˣrnesto retweeted
We spent $2m so you don't have to.
And so you can learn how to train MoEs.
Mixture-of-Experts LLMs are what powers the AIs you use everyday, and few people in the world know how to train them at scale (a handful at each top AI lab).
This handbook distills everything we learned in meticulously crafted chapters and figures.
Disclaimer: it may take days to read.
We hope to see 1000x more people able to train their own frontier models all over the world! 🤗
Congrats @Nouamanetazi @lvwerra @Thom_Wolf
eˣrnesto retweeted
18 of the 20 best open models today are Mixture-of-Experts. Dense is basically over.
And yet the know-how to train a good MoE still lives mostly in the heads of a handful of teams.
So we spent 800k GPU hours running the experiments ourselves and wrote everything down. Super happy to finally release "The Ultimate MoE Training Handbook", free and open, organized around the questions training teams hit:
- how many experts, how big, shared or not?
- why is half my step spent in all-to-all, and how do I hide it?
- why is my run 25x slower with no error in the logs?
- softmax, sigmoid or sqrt(softplus) router, and why keep it in fp32?
- aux loss or DeepSeek's loss-free bias?
- upcycle a dense checkpoint or train from scratch?
- can I serve with fewer experts per token than I trained with?
⚠️ Warning: it's long (~70k words, budget a few days, sorry) but with 55 interactive figures, code you can run and numbers from our own runs.
Read it here: huggingface.co/spaces/huggin…
eˣrnesto retweeted
Just a reminder that DeepSeek V4.1 Flash 1017, GLM 5.3 Flash, Qwen 3.8 Next Flash, and even Qwen 3.8 27B are all outperforming (in both intelligence and capabilities) every model that was considered "frontier intelligence" in Xmas 2025 (just 11 months ago)
Opensource AI is on fire
eˣrnesto retweeted
New Bonus Prize: $10,000 for Small Model Development agents 🌟
The highest-ranking submission built on Gemma E2B or Gemma E4B wins, based on private test set performance.
• LoRA fine-tuning allowed, same as the main track
• Must be one of your team's two selected final submissions
• Still eligible for main leaderboard prizes and the Paper Track
• Public standings for eligible entries in a notebook updated daily
Same deadlines as the main track.
@GoogleResearch @googlegemma
Post-train an open model into a coding agent that accelerates developer workflows on everyday hardware.
The Gemma 4 Developer Agent Competition, hosted by @googlegemma and Kaggle, is live!
💰 Total Prize Pool: $100,000
⏰ Entry Deadline: November 25, 2026
eˣrnesto retweeted
We benchmarked DeepSeek-V4.1-Flash-1017 GGUFs on Terminal Bench, DeepSWE & more on 512K context. Unsloth ranks highest overall.
1-bit Unsloth GGUF scores 68.1 on DeepSWE, 72.2 on Terminal-Bench, 52.7 on AutomationBench. >10% ahead of other quants.
GGUF: huggingface.co/unsloth/DeepS…
We’re releasing new DeepSeek-V4.1-Flash-1017 GGUFs, ranking highest across benchmarks.
Dynamic V3.1 leads across nearly all benchmarks: GPQA Diamond, AIME25, LiveCodeBench v6, MMLU-Pro, Aider & more.
Run on 64GB+ RAM.
GGUF: huggingface.co/unsloth/DeepS…
Blog: unsloth.ai/docs/models/deeps…
eˣrnesto retweeted
This is an old hat. That's exactly what the neural fast weight programmers did in the 1990s (including the 1991 unnormalised linear Transformer). Overview with links to the original papers: people.idsia.ch/~juergen/fas…
We asked GPT-6 Astra to edit Qwen3.5-0.8B weights to improve math.
It directly changed only 0.016% of weights, fixing 0.8B's reasoning loops & improving math up to 6x and even instruction following 3x, without gradients or training.
We call this Weight-Space Programming: frontier LLMs can now natively read/write neural network weights! 🧵
Blog: rikiyat.github.io/blog/weigh…
Readers added context they thought people might want to know
LLM was given weights explicitly as text in prompt and used code/forward passes (per authors' logs) to ablate Qwen looping direction, a known thinking-mode issue per model card. Primarily fixed repetition in that mode, not general 'natively' improving weights/math.
rikiyat.github.io/blog/weight-sp…
github.com/RikiyaT/lm_und…
huggingface.co/Qwen/Qwen3.5-0…
huggingface.co/RikiyaT/Qwen3.…
eˣrnesto retweeted
🚨 We are receiving reports of hot wallets on BTCPay Servers being swept by unauthorized attackers.
If you're running an instance, please shut it down (./btcpay-down.sh) until we complete the investigation.
eˣrnesto retweeted
Iran has designated the United Kingdom’s MI6 intelligence agency a “hostile group” and declared the Royal Navy and Royal Air Force “terrorist organisations”, in what it called retaliation for London’s “hostile” actions.
🔴 LIVE updates: aje.news/y3fgub
eˣrnesto retweeted
From recognizing text to understanding documents. 📄
Meet Xiaomi-OCR-0, an open-source document understanding model with just 0.8B parameters.
It brings document parsing, key information extraction, and OCR-oriented visual question answering into one lightweight model.
It scores 95.24 on Real5-OmniDocBench and 83.2 on average across five OCR-oriented VQA benchmarks.
On OmniDocBench v1.6, Xiaomi-OCR-0 ranks first in document parsing performance among models with fewer than 1B parameters.
Model weights, inference code, an interactive demo, MCP service, and Agent Skill are now open source. 👇
🤗 Huggingface: huggingface.co/xiaomi-resear…
🔗 GitHub: github.com/xiaomi-research/X…
📄 Demo: huggingface.co/spaces/loongw… #XiaomiAI #OCR #OpenSource
eˣrnesto retweeted
Funding rounds aren't milestones. It shouldn't be your goal to raise a series A. It should be your goal not to need to.
eˣrnesto retweeted
The latest @baseten open weight Decision model is now on @huggingface, and it's just 2B parameters and MIT licensed (a.k.a. commercially viable).
Check it out: huggingface.co/baselabs/joey…
Today we’re releasing our third open-source decision model, joey-1 (2B), the fastest decision model available and among the most intelligent decision models of its size.
joey-1 is trained by the @baselabs team, and is available on our Model APIs, as a dedicated deployment, and with a cookbook to fine-tune on our Training Jobs.
joey-1 is built to maximize speed and minimize cost without sacrificing intelligence for those workloads where TTFT really matters.
Jump on it: baseten.co/library/joey-1/
eˣrnesto retweeted
🚨 We're releasing BTPay Server 2.4.7 an emergency update fixing one critical and several medium-severity security vulnerabilities brought to our attention through @Anthropic's frontier OSS scanner.
We recommend that you update your BTCPay Server as soon as possible.
eˣrnesto retweeted
Mistral Le Chonk can now run locally!
The 1 trillion parameter model runs on a Mac Studio with our 1-bit GGUF which retains 91% top-1 accuracy.
Le Chonk is the strongest Western open model to date.
GGUF: huggingface.co/unsloth/Mistr…
Guide: unsloth.ai/docs/models/mistr…
Today, we release the weights of a new and improved Le Chonk, advancing the open frontier for agentic coding and cybersecurity.
💻 Built to build: Le Chonk ranks #1 on @cognition’s FrontierCode among open models, and sets a new benchmark for Pareto efficiency on @artificialanalysis.
🛡 Built to defend: State-of-the-art in cyberdefense, put through the paces by the world's leading cyber vendors, including @CrowdStrike, @cloudflare, @rubrik, @zscaler, and others.
Built for sovereignty: Make it yours. Download the weights on @huggingface , customize it with Mistral Forge, and deploy it wherever you need it.
Read more from our team and customers 👇:
