🔥The 9th Round of Easy Loan, Earn $40 Reward is in progress❗️ ⏰ Promotion Period: January 15th - Feburary 15th, 2025 👉 Register now and check more details at gate.io/campaigns/358

Training epochs definitely move faster when you stare at the progress bars

LLaMA-1 paper dropped on arxiv 2 years ago today! 🙏

We are hiring on the PyTorch team! 🙏 Partner Engineer is a mix of building applications with real use cases, applied research and software engineering I work on the llama wing but learn so much everytime I speak with PyTorch org. They are some of the smartest and most humble

Lake Tahoe is beautiful! ❤️

Playbook to defeat frontier ai labs without billions of dollars initially: - build an app on top of their models - solve an important, large problem for humanity - resume training on top OSS models to reduce dependency, lower costs for certain tasks, increase performance -

x.com/i/article/1904…

Llama 4 supports 10M Context length! 🙏 Reading AN ENTIRE GitHub repo of 900k tokens and writing a guide on it takes under 3 minutes! We are launching two new models Scout and Maverick: - Upto 10M context length - Scout fits on single H100 with int4 quant - Upto 5 images -

Llama 4 inference in pure JAX! Expert/tensor parallelism with int8 quantization. Contributions welcome!

Llama 4 Intelligence Index Update: We have now replicated Meta’s claimed values for MMLU Pro and GPQA Diamond, pushing our Intelligence Index scores for both Scout and Maverick higher Key update details: ➤ We noted in our first post 48 hours ago that we noticed discrepancies

1.5M tokens to website in 5 minutes 🙏 - Upload an entire repo of apps - Upload multiple sketches website - Use the repo content to populate the template Llama 4 supports 10M context + upto 10 images in a session:

Also note if you're not getting good Llama 4 results, there are a few bugs: 1. QK Norm eps should be 1e-5 - collabed with HF on the fix! github.com/huggingface/tr… 2. RoPE scaling for Scout changed: github.com/ggml-org/llama… 3. vLLM +2% acc shared QK norm fix: github.com/vllm-project/v…

We’re excited to showcase all the amazing ways you’ve been building with Llama + Unsloth at LlamaCon 2025! 🦥🦙 Get ready for surprises and exciting announcements from Meta and us on Apr 29 in SF. 👀 Also big thanks to AI at Meta for the support and awesome merch!

Super excited to launch Synthetic-Data-Kit! 🙏 Fine-tuning LLMs is easy, there are many packages to get started, Unsloth AI is my absolute favorite. However, there is still a BIG HURDLE when working on fine-tuning: Data preparation Today I’m super grateful to be launching a

We partnered with AI at Meta on a free notebook that turns your documents into high-quality synthetic datasets using Llama! Features: • Parses PDFs, websites, videos • Use Llama to generate QA pairs + auto-filter data • Fine-tunes dataset with Llama 🔗colab.research.google.com/github/unsloth…

Llama Synthetic Data Fine-Tuning Guide! 🙏 My favourite thing about this tutorial-everything is powered by 3B model. It covers a step overlooked everywhere-data preparation and generation for fine tuning. Thanks Unsloth AI team for this gem: colab.research.google.com/github/unsloth…

You can now fine-tune TTS models with Unsloth! Train, run and save models like Sesame-CSM and OpenAI's Whisper locally with our free notebooks. Unsloth makes TTS training 1.5x faster with 50% less VRAM. GitHub: github.com/unslothai/unsl… Docs & Notebooks: docs.unsloth.ai/basics/text-to…

We're bringing the Unsloth magic to TTS and audio models! There are multiple free Colab notebooks with free GPUs for Whisper, Sesame, Orpheus, Spark, Llasa & Oute on our docs! docs.unsloth.ai/basics/text-to…

Every birthday, I recap my favourite learnings 🙏 This time, I’ve been laser focussed on simple ideas for fine-tuning and data generation. A roadmap of 8 ideas from papers that I’ve enjoyed the most, along with our implementations: 1. How much data do you need to fine-tune?