I^2

InferenceIllusionist

InferenceIllusionist

AI & ML interests

He deals the cards to find the answer / The sacred geometry of chance / The hidden law of a probable outcome / The numbers lead a dance

Recent Activity

updated a model 27 days ago

Quant-Cartel/Questionable-MN-12B-iMat-GGUF

published a model 27 days ago

Quant-Cartel/Questionable-MN-12B-iMat-GGUF

posted an update about 2 months ago

MilkDropLM-32b-v0.3: Unlocking Next-Gen Visuals ✨ Stoked to release the latest iteration of our MilkDropLM project! This new release is based on the powerful Qwen2.5-Coder-32B-Instruct model using the same great dataset that powered our 7b model. What's new? - Genome Unlocked: Deeper understanding of preset relationships for more accurate and creative generations. - Preset Revival: Breathe new life into old presets with our upgraded model! - Loop-B-Gone: Say goodbye to pesky loops and hello to smooth generation. - Natural Chats: Engage in more natural sounding conversations with our LLM than ever before. Released under Apache 2.0, because sharing is caring! Try it out: https://huggingface.co/InferenceIllusionist/MilkDropLM-32b-v0.3 Shoutout to @superwatermelon for his invaluable insights and collab, and to all those courageous members in the community that have tested and provided feedback before!

View all activity

Organizations

InferenceIllusionist's activity

posted an update about 2 months ago

Post

1995

MilkDropLM-32b-v0.3: Unlocking Next-Gen Visuals ✨

Stoked to release the latest iteration of our MilkDropLM project! This new release is based on the powerful Qwen2.5-Coder-32B-Instruct model using the same great dataset that powered our 7b model.

What's new?

- Genome Unlocked: Deeper understanding of preset relationships for more accurate and creative generations.

- Preset Revival: Breathe new life into old presets with our upgraded model!

- Loop-B-Gone: Say goodbye to pesky loops and hello to smooth generation.

- Natural Chats: Engage in more natural sounding conversations with our LLM than ever before.

Released under Apache 2.0, because sharing is caring!

Try it out: InferenceIllusionist/MilkDropLM-32b-v0.3

Shoutout to @superwatermelon for his invaluable insights and collab, and to all those courageous members in the community that have tested and provided feedback before!

posted an update 2 months ago

Post

1878

Howdy folks, your friendly neighborhood fine-tuner I^2 here coming out of stealth with a thrilling model release announcement for you all.

🎵 Introducing MilkDropLM-v0.3: An LLM for Generating MilkDrop Presets
InferenceIllusionist/MilkDropLM-7b-v0.3

A weekend project that combines two of my favorite things - LLMs and music visualization. MilkDropLM is an experimental model trained to generate MilkDrop preset scripts, aiming to expand the creative possibilities in the visualization space. If you're familiar with WinAmp, MilkDrop was one of the many spiritual successors that spun out of its visualization features.

-Model Details-

Early alpha release (v0.3)
- Trained to generate MilkDrop preset syntax
- Based on qwen-2.5-coder-7b due to its math and programming ability
- Fine-tuned on 10k+ cream of the crop presets, hand-curated by a MilkDrop vet (shout out @superwatermelon !)
- Apache 2.0 licensed

The training process was fascinating - a lot of broader ML questions about this space that you can read all about on the model card.

This is still an early alpha so any feedback is appreciated but I'm excited to share the results with you all.

Feel free to experiment, contribute, or just grab some presets for your next visualization session. PRs and feedback always welcome!
Demo notebook and training details in the repo.

PS: A HUGE thank you to my followers old and new - your support does not go unnoticed. It's my absolute honor to share these experiments with you all!

reacted to clem's post with 🚀 8 months ago

Post

5785

5,000 new repos (models, datasets, spaces) are created EVERY DAY on HF now. The community is amazing!

reacted to dvilasuero's post with 🚀 8 months ago

Post

8159

Today is a huge day in Argilla’s history. We couldn’t be more excited to share this with the community: we’re joining Hugging Face!

We’re embracing a larger mission, becoming part of a brilliant and kind team and a shared vision about the future of AI.

Over the past year, we’ve been collaborating with Hugging Face on countless projects: launching partner of Docker Spaces, empowering the community to clean Alpaca translations into Spanish and other languages, launching argilla/notus-7b-v1 building on Zephyr’s learnings, the Data is Better Together initiative with hundreds of community contributors, or releasing argilla/OpenHermesPreferences, one of the largest open preference tuning datasets

After more than 2,000 Slack messages and over 60 people collaborating for over a year, it already felt like we were part of the same team, pushing in the same direction. After a week of the smoothest transition you can imagine, we’re now the same team.

To those of you who’ve been following us, this won’t be a huge surprise, but it will be a big deal in the coming months. This acquisition means we’ll double down on empowering the community to build and collaborate on high quality datasets, we’ll bring full support for multimodal datasets, and we’ll be in a better place to collaborate with the Open Source AI community. For enterprises, this means that the Enterprise Hub will unlock highly requested features like single sign-on and integration with Inference Endpoints.

As a founder, I am proud of the Argilla team. We're now part of something bigger and a larger team but with the same values, culture, and goals. Grateful to have shared this journey with my beloved co-founders Paco and Amélie.

Finally, huge thanks to the Chief Llama Officer @osanseviero for sparking this and being such a great partner during the acquisition process.

Would love to answer any questions you have so feel free to add them below!

28 replies

reacted to thomwolf's post with 🚀🔥 10 months ago

Post

4898

Is is time for the open-source AI robots revolution 🚀?

With @haixuantao and @Leyo we’ve been playing with a low-cost DJI robot controlled by three local open-source AI models (Whisper, Idefics2, Parler-TTS - all Apache2) and orchestrated by Dora-cs.

Links to find all the hardware/software we used in the demo:
- robot control framework – dora-rs: https://github.com/dora-rs/dora
- speech-to-text model – whisper: openai/whisper-base
- vision-text model – Idefics2: HuggingFaceM4/idefics2-8b-AWQ
- text-to-speech model – ParlerTTS mini: parler-tts/parler_tts_mini_v0.1
- robot: https://dji.com/robomaster-s1
- code gist: https://gist.github.com/haixuanTao/860e1740245dc2c8dd85b496150a9320
- Larger codebase: dora-rs/dora-idefics2
- laptop/pc: any with a recent GPU card (our has a RTX 4090)

Enjoy!

4 replies

reacted to clem's post with 🚀 10 months ago

Post

2434

Fun dataset added last week by @esind from https://huggingface.co/Anthropic to compare persuasiveness between AI and human outputs:
Anthropic/persuasion

2 replies

reacted to VictorSanh's post with 🔥 10 months ago

Post

2541

Can't wait to see multimodal LLama 3!

We released a resource that might come in handy: The Cauldron 🍯

The Cauldron is a massive manually-curated collection of 50 vision-language sets for instruction fine-tuning. 3.6M images, 30.3M query/answer pairs.

It covers a large variety of downstream uses: visual question answering on natural images, OCR, document/charts/figures/tables understanding, textbooks/academic question, reasoning, captioning, spotting differences between 2 images, and screenshot-to-code.

HuggingFaceM4/the_cauldron

1 reply

reacted to yagilb's post with 🔥 11 months ago

Post

4452

Today we're starting a new initiative: LM Studio Community Models! 🤖

@bartowski , a prolific quantizer (both GGUF and EXL2) will be helping to curate notable new models in LM Studio's Community Models page: https://huggingface.co/lmstudio-community.

Our goal is to ensure the community has access to GGUF files for new & noteworthy models as soon as possible. Keep an eye on that page for updates.

If you're unfamiliar with GGUF, it's the de-facto standard for 'compressed' LLM weights. It is the native format of llama.cpp (https://github.com/ggerganov/llama.cpp, an LLM runtime C/C++ library.) This format is supported in LM Studio.

We will also be sharing new models on the LM Studio Discord: https://discord.gg/aPQfnNkxGC

5 replies

reacted to clefourrier's post with 👍 11 months ago

Post

2362

Fun fact about evaluation!

Did you know that, if you evaluate the same model, with the same prompt formatting & the same fixed few-shot examples, only changing
♻️the order in which the few shot examples are added to the prompt ♻️
you get a difference of up to 3 points in evaluation score?

I did a small experiment using some MMLU subsets on the best performing 7B and lower pretrained models from the leaderboard.

I tried 8 different prompting methods (containing more or less information, such as just the question, or Question: question, or Question: question Choices: ..., see the x axis) that are commonly used in evaluation.

I then compared the results for all these methods, in 5-shot, during 2 runs. The *only difference* between the first and second run being that the samples used in few-shot are not introduced in the same order.
For example, run one would have been "A B C D E Current sample", vs, in run 2, "D C E A B Current sample".
All the other experiment parameters stayed exactly the same.

As you can see on the attached picture, you get a difference of up to 3 points between the 2 few-shot samples shuffling.

So, when just changing *the order of the few shot samples* can change your results by several points, what is the impact of all other "minimal" and unreported prompting changes?

-> Any kind of model score, provided without an evaluation script for reproducibility, is basically bullshit (or coms).
-> This is why we need reproducible evaluation in a fair and exactly similar setup, using evaluation suites such as lm_eval from the Harness, lighteval from HF, or the Open LLM Leaderboard.

4 replies

reacted to Locutusque's post with ❤️ 11 months ago

Post

2646

Exciting news! 🎉 I've created the OpenCerebrum datasets, open-source alternatives to Aether Research's proprietary Cerebrum dataset.

The first, OpenCerebrum SFT, is a text-generation and question-answering dataset with ~1.2M examples, curated from sources like Open-Orca, glaiveai, camel-ai, and more! 📚

The second, OpenCerebrum DPO, is a smaller dataset with ~21k examples, focusing on data point optimization. It's curated from sources like jondurbin, argilla, grimulkan, and others. 📊

Both datasets are licensed under Apache-2.0 and are available in English. They're ready for use in your projects, and I welcome any feedback for future improvements! 🚀

Locutusque/OpenCerebrum-dpo
Locutusque/OpenCerebrum-SFT
Locutusque/OpenCerebrum-1.0-7b-SFT
Locutusque/OpenCerebrum-1.0-7b-DPO

5 replies

reacted to shumingma's post with 🚀 11 months ago

Post

2697

The Era of 1-bit LLMs: Training Tips, Code and FAQ

https://github.com/microsoft/unilm/blob/master/bitnet/The-Era-of-1-bit-LLMs__Training_Tips_Code_FAQ.pdf

We present details and tips for training 1-bit LLMs. We also provide additional experiments and results that were not reported and responses to questions regarding the "The-Era-of-1-bit-LLM" paper. Finally, we include the official PyTorch implementation of BitNet (b1.58 and b1) for future research and development of 1-bit LLMs.

2 replies

reacted to osanseviero's post with ❤️🚀 11 months ago

Post

1893

Diaries of Open Source. Part 7!

🔥Sakana releases Evolutionary Model Merge
Blog post: https://sakana.ai/evolutionary-model-merge/
Paper: Evolutionary Optimization of Model Merging Recipes (2403.13187)
Models and demo: https://hf.co/SakanaAI

🍞MixedBread releases new SoTA sentence embedding model
Announcement: https://www.mixedbread.ai/blog/mxbai-embed-large-v1
Model: mixedbread-ai/mxbai-embed-large-v1

🎥VideoMamba, a Mamba-based model for video understanding
Blog: https://hf.co/blog/vladbogo/video-mamba
Demo: OpenGVLab/VideoMamba
Model: OpenGVLab/VideoMamba

🔍 MathVerse, a visual math benchmark for multimodal LLMs
Paper page: https://mathverse-cuhk.github.io/
Dataset: AI4Math/MathVerse
Paper: MathVerse: Does Your Multi-modal LLM Truly See the Diagrams in Visual Math Problems? (2403.14624)

🧠GraphWiz, a family of instruct-tuned LLMs to solve graph problems
Repos: https://hf.co/GraphWiz
Paper: GraphWiz: An Instruction-Following Language Model for Graph Problems (2402.16029)

🪆NLLB-SigLIP-MRL: a combination of NLLB and SigLIP trained with Matryoshka representation learning
Model: visheratin/nllb-siglip-mrl-large
Tweet: https://twitter.com/visheratin/status/1766643219909984734?s=46

🧍HDM and ProciGen: Template-free reconstruction of human-object interactions
Paper page: https://virtualhumans.mpi-inf.mpg.de/procigen-hdm/
Demo: xiexh20/HDM-interaction-recon
Models: xiexh20/HDM-models

🌎Models and data around the world
EagleX 7B, multi-lingual RNN-based model https://hf.co/spaces/recursal/EagleX-7B-1.7T-Gradio-Demo
Tamil LLM mervinpraison/tamil-large-language-model-7b-v1.0

2 replies

reacted to merve's post with 🚀 11 months ago

Post

3327

LLaVA-NeXT is recently merged to Hugging Face transformers and it outperforms many of the closed source models like Gemini on various benchmarks 🤩 Let's take a look!
Demo: merve/llava-next
Notebook: https://colab.research.google.com/drive/1afNudu72SNWZCYtCVrRlb9T9Vj9CFJEK?usp=sharing
LLaVA is essentially a vision-language model that consists of ViT-based CLIP encoder, a MLP projection and Vicuna as decoder ✨
LLaVA 1.5 was released with Vicuna, but LLaVA NeXT (1.6) is released with four different LLMs:
- Nous-Hermes-Yi-34B
- Mistral-7B
- Vicuna 7B & 13B
Mistral and Nous-Hermes-Yi-34B are performing better and have better commercial use.
Moreover, according to authors' findings, the improvements comes from more diverse and high quality data mixture and dynamic high resolution.
LLaVA based on Nous-Hermes-Yi-34B outperforms many other models, including Gemini in various multimodal understanding and generation benchmarks 😊

reacted to loubnabnl's post with 🔥 11 months ago

Post

6417

We've just published a detailed blog post on the creation of Cosmopedia dataset. We hope this will provide insights about generating synthetic data at scale for pre-training.
https://huggingface.co/blog/cosmopedia

Here are some key takeaways:
🎯 Prompt curation is crucial: we want to cover many topics with few duplicates.
📚 You can leverage various resources for diversity: using different seed data, generation formats, and target audiences.
⚙️ The importance of a good technical stack: for scalable generations with tools like llm-swarm and fast model training and evaluation.

Have a good read!

1 reply

reacted to akhaliq's post with 👍 12 months ago

Post

MobileLLM

Optimizing Sub-billion Parameter Language Models for On-Device Use Cases

MobileLLM: Optimizing Sub-billion Parameter Language Models for On-Device Use Cases (2402.14905)

paper addresses the growing need for efficient large language models (LLMs) on mobile devices, driven by increasing cloud costs and latency concerns. We focus on designing top-quality LLMs with fewer than a billion parameters, a practical choice for mobile deployment. Contrary to prevailing belief emphasizing the pivotal role of data and parameter quantity in determining model quality, our investigation underscores the significance of model architecture for sub-billion scale LLMs. Leveraging deep and thin architectures, coupled with embedding sharing and grouped-query attention mechanisms, we establish a strong baseline network denoted as MobileLLM, which attains a remarkable 2.7%/4.3% accuracy boost over preceding 125M/350M state-of-the-art models. Additionally, we propose an immediate block-wise weight sharing approach with no increase in model size and only marginal latency overhead. The resultant models, denoted as MobileLLM-LS, demonstrate a further accuracy enhancement of 0.7%/0.8% than MobileLLM 125M/350M. Moreover, MobileLLM model family shows significant improvements compared to previous sub-billion models on chat benchmarks, and demonstrates close correctness to LLaMA-v2 7B in API calling tasks, highlighting the capability of small models for common on-device use cases.