OptimalScale

Building Extensible Toolkit for Finetuning and Inference of Large Foundation Models. Large Model for All.

[Rui Pan @rui4research]

NVIDIA Hymba-1.5B Instruction Fine-tuning, powered by LMFlow🚀

[Rui Pan @rui4research]

Excited to share LISA, which enables
- 7B tuning on a 24GB GPU
- 70B tuning on 4x80GB GPUs

and obtains better performance than LoRA in ~50% less time 🚀

[OptimalScale @OptimalScale]

Honesty Alignment🥳

[Shizhe Diao @shizhediao]

Thanks for following our work!
Instructing towards honest LLMs is an interesting topic and it is also important for alignment research.
A surprising discovery: Learning uncertainty improves model calibration (expected) and prediction (surprising)! 🌟

[Hanrong YE @ Nvidia @leoyerrrr]

🌟Read a paper on "Refusal-Aware Instruction Tuning", which is pretty interesting. They identified the knowledge gap between LLMs and instruction tuning data. With R-Tuning, they teach LLMs to recognize when they lack knowledge and resist making things up.

[Hanze Dong @ COLM 2026 @hendrydong]

🚀 Exciting to share our latest TMLR paper — RAFT: Reward rAnked FineTuning for aligning Generative Foundation Models! 🧠💡
https://t.co/3eDv62Qdyl

[OptimalScale @OptimalScale]

Thanks for everyone's attention! We are a group of researchers from the Hong Kong University of Science and Technology, affiliating to the Statistics and Machine Learning Research Group https://t.co/ghMr2WlRXK.

[Omar Sanseviero @osanseviero]

And so many more: @weizmannscience, @wabyking, @polodataclub, @tomgoldsteincs group, @OptimalScale, and so on

[Omar Sanseviero @osanseviero]

And so many more: @weizmannscience, @wabyking, @polodataclub, @tomgoldsteincs group, @OptimalScale, and so on

[OptimalScale @OptimalScale]

LMFlow now supports Speculative Decoding! 🥳
Experience faster model inference without retraining or tweaking the architecture! We're seeing close to a 4x speed boost (gpt2-xl <- gpt2). 🚀
Check out the speculative_inference.py in LMFlow!
https://t.co/muag8bgkAf

[OptimalScale @OptimalScale]

LMFlow now supports training and inference using FlashAttention-2! 🥳
This cutting-edge feature will take your language modeling to the next level. To use it, simply add --use_flash_attention True.
Upgrade to LMFlow now and experience the future of LM! 🚀https://t.co/jNIon94iAQ

[OptimalScale @OptimalScale]

online demo: https://t.co/FPX4rd50wI

[OptimalScale @OptimalScale]

Personalized chatbot toolbox LMFlow now supports image inputs!😆 With LMFlow, you can build "MiniBard" with ease!😉 Currently we support chatbots with minigpt4 + robin 7b/13b inference, the multimodal finetuning service will be available soon~
https://t.co/tb4W2dEb5y

[elvis @omarsar0]

LMFlow: an extensible and lightweight toolkit that simplifies finetuning and inference of general large foundation models.

Also supports continuous pretraining, instruction tuning, parameter-efficient finetuning, alignment tuning, and large model inference.

Looks very promising!

paper: https://t.co/tIatSbfEl0
code: https://t.co/CFhSnr1BT5

[OptimalScale @OptimalScale]

🥳 The LMFlow paper (https://t.co/ojT44dsgrd) has officially dropped! 📚⚡️
⏩ LMFlow is your go-to toolkit for unleashing the power of LLMs. 🌟✨
🔎 Dive into the nitty-gritty of our mind-blowing implementation.💥🔥
💻 Let your creativity flow freely! 🌊💡
#LMFlow #LLM #GitHub

[AK @_akhaliq]

LMFlow-CHAT

demo: https://t.co/nzgbYYWvGQ

fully upgraded Robin Series V2 language model

carried out in-depth fine-tuning based on the entire LLaMA series, including 7b, 13b, 33b, 65b, all of which have achieved pleasing results. Robin-7b scored 51.7 in the OpenLLM standard test, and Robin-13b even reached as high as 59.1, ranking sixth, surpassing many 33b models. The achievements of Robin-33b and Robin-65b are even more surprising, with scores of 64.1 and 65.2 respectively, firmly securing the top positions.

[OptimalScale @OptimalScale]

Robin V2 have carried out in-depth fine-tuning based on the entire LLaMA series, including 7b, 13b, 33b, 65b, all of which have achieved pleasing results.

[OptimalScale @OptimalScale]

Robin-7B scored 51.7 in the OpenLLM standard test, and Robin-13B even reached 59.1, ranking sixth, surpassing some 60B models. The achievements of Robin-33b and Robin-65b are even more surprising, with scores of 64.1 and 65.2 respectively, firmly securing the top positions.

[OptimalScale @OptimalScale]

Use LMFlow (https://t.co/EeVD5Bb72q) for easy, one-stop training & fine-tuning. Dive in, experience the power & richness of Robin V2. Reach out to us anytime, we're here to help. Feel the Robin V2 difference today! #AI #LanguageModel #RobinV2

[OptimalScale @OptimalScale]

Introducing the Robin V2 - a leap in LLM fine-tuning with LMFlow! 🚀Defeating major open-source LLMs like Falcon, LLaMA & more. Robin-7B to 65B boast impressive scores in OpenLLM.🥇Deep-tuned & optimized for accuracy in multiple domains. For results, check out LMFlow-benchmark.📊

[OptimalScale @OptimalScale]

Evaluation dataset is released. We have manually constructed a high-quality evaluation dataset, which prompts entirely come from natural user inquiries. This evaluation dataset has been fully released (see https://t.co/DYcupNw25Y), and everyone is welcome to use it.

[OptimalScale @OptimalScale]

【LMFlow 0515 Update is here! 🚀】
1. New evaluation: LMFlow Benchmark
2. New model: Robin-7b-v2
3. New training data: LMFlow-data
4. New test data: LMFlow-test