3
Flux Quantized or Original?
🔎
Generate and compare quantized and non-quantized images
🧨Diffusers
non-profit
AI & ML interests
Solving everything with diffusion models!
Recent Activity
View all activity
diffusers's activity
derekl35
updated
a
Space
3 days ago
Post
2293
Diffusers supports a good variety of quantization backends. It can be challenging to navigate through them, given the complex nature of diffusion pipelines in general.
So, @derekl35 set out to write a comprehensive guide that puts users in the front seat. Explore the different backends we support, learn the trade-offs they offer, and finally, check out the cool space we built that lets you compare quantization results.
Give it a go here:
https://lnkd.in/gf8Pi4-2
So, @derekl35 set out to write a comprehensive guide that puts users in the front seat. Explore the different backends we support, learn the trade-offs they offer, and finally, check out the cool space we built that lets you compare quantization results.
Give it a go here:
https://lnkd.in/gf8Pi4-2
Post
1654
Despite the emergence of combining LLM and DiT architectures for T2I synthesis, its design remains severely understudied.
This was done long ago and got into CVPR25 -- super excited to finally share it now, along with the data and code ♥️
We explore several architectural choices that affect this design. We provide an open & reproducible training recipe that works at scale.
Works like Playground v3 have already explored a deep fusion between an LLM and a DiT, sharing their representations through layerwise attention. They exhibit excellent performance on T2I.
Despite its compelling results and other performance virtues, it remains unexplored, which is what we want to improve in our work. Specifically, we take a pre-trained LLM (Gemma-2B) and trainable DiT, and set out to explore what makes a "good deep fusion" between the two for T2I.
We explore several key questions in the work, such as:
Q1: How should we do attention? We considered several alternatives. PixArt-Alpha like attention (cross-attention) is very promising.
Q2: Should we incorporate additional text modulation?
Q3: Can we eliminate timestep conditioning?
Q4: How do we do positional encodings?
Q5: Do instruction-tuned LLMs help deep fusion?
Q6: Would using a decoder LLM from a multimodal model be helpful?
Q7: Does using a better variant of Gemma help?
Based on the above findings, we arrive at FuseDiT with the following components on top of the base architecture from the findings of our experiments.
* No AdaLN-Zero modules
* 1D + 2D-RoPE
* Gemma 2 2B, adjusting DiT configurations accordingly
We trained FuseDiT on a mixture from CC12M, JourneyDB, & SA (~26M image-text pairs) for 800 steps. While not the best model, it's encouraging to develop something in a guided manner using open datasets.
To know more (code, models, all are available), please check out the paper:
https://lnkd.in/gg6qyqZX.
This was done long ago and got into CVPR25 -- super excited to finally share it now, along with the data and code ♥️
We explore several architectural choices that affect this design. We provide an open & reproducible training recipe that works at scale.
Works like Playground v3 have already explored a deep fusion between an LLM and a DiT, sharing their representations through layerwise attention. They exhibit excellent performance on T2I.
Despite its compelling results and other performance virtues, it remains unexplored, which is what we want to improve in our work. Specifically, we take a pre-trained LLM (Gemma-2B) and trainable DiT, and set out to explore what makes a "good deep fusion" between the two for T2I.
We explore several key questions in the work, such as:
Q1: How should we do attention? We considered several alternatives. PixArt-Alpha like attention (cross-attention) is very promising.
Q2: Should we incorporate additional text modulation?
Q3: Can we eliminate timestep conditioning?
Q4: How do we do positional encodings?
Q5: Do instruction-tuned LLMs help deep fusion?
Q6: Would using a decoder LLM from a multimodal model be helpful?
Q7: Does using a better variant of Gemma help?
Based on the above findings, we arrive at FuseDiT with the following components on top of the base architecture from the findings of our experiments.
* No AdaLN-Zero modules
* 1D + 2D-RoPE
* Gemma 2 2B, adjusting DiT configurations accordingly
We trained FuseDiT on a mixture from CC12M, JourneyDB, & SA (~26M image-text pairs) for 800 steps. While not the best model, it's encouraging to develop something in a guided manner using open datasets.
To know more (code, models, all are available), please check out the paper:
https://lnkd.in/gg6qyqZX.
Post
3445
hey hey
@mradermacher
- VB from Hugging Face here, we'd love to onboard you over to our optimised xet backend! 💥
as you know we're in the process of upgrading our storage backend to xet (which helps us scale and offer blazingly fast upload/ download speeds too): https://huggingface.co/blog/xet-on-the-hub and now that we are certain that the backend can scale with even big models like Llama 4/ Qwen 3 - we;re moving to the next phase of inviting impactful orgs and users on the hub over as you are a big part of the open source ML community - we would love to onboard you next and create some excitement about it in the community too!
in terms of actual steps - it should be as simple as one of the org admins to join hf.co/join/xet - we'll take care of the rest.
p.s. you'd need to have a the latest hf_xet version of huggingface_hub lib but everything else should be the same: https://huggingface.co/docs/hub/storage-backends#using-xet-storage
p.p.s. this is fully backwards compatible so everything will work as it should! 🤗
as you know we're in the process of upgrading our storage backend to xet (which helps us scale and offer blazingly fast upload/ download speeds too): https://huggingface.co/blog/xet-on-the-hub and now that we are certain that the backend can scale with even big models like Llama 4/ Qwen 3 - we;re moving to the next phase of inviting impactful orgs and users on the hub over as you are a big part of the open source ML community - we would love to onboard you next and create some excitement about it in the community too!
in terms of actual steps - it should be as simple as one of the org admins to join hf.co/join/xet - we'll take care of the rest.
p.s. you'd need to have a the latest hf_xet version of huggingface_hub lib but everything else should be the same: https://huggingface.co/docs/hub/storage-backends#using-xet-storage
p.p.s. this is fully backwards compatible so everything will work as it should! 🤗
- 11 replies
sayakpaul
authored
a
paper
16 days ago
Post
3116
FramePack is hands down one of the best OS releases in video generation 🙇🏻♀️🤯
✅ fully open sourced + amazing quality + reduced memory + improved speed
but more even - its gonna facilitate *soooo* many downstream applications
like this version adapted for landscape rotation 👇https://huggingface.co/spaces/tori29umai/FramePack_rotate_landscape
✅ fully open sourced + amazing quality + reduced memory + improved speed
but more even - its gonna facilitate *soooo* many downstream applications
like this version adapted for landscape rotation 👇https://huggingface.co/spaces/tori29umai/FramePack_rotate_landscape
- 2 replies
Post
4719
HOW TO ADD MCP SUPPORT TO ANY 🤗 SPACE
Gradio now supports MCP! If you want to convert an existing Space, like this one hexgrad/Kokoro-TTS, so that you can use it with Claude Desktop / Cursor / Cline / TinyAgents / or any LLM that supports MCP, here's all you need to do:
1. Duplicate the Space (in the Settings Tab)
2. Upgrade the Gradio
3. Set
4. (Optionally) add docstrings to the function so that the LLM knows how to use it, like this:
That's it! Now your LLM will be able to talk to you 🤯
Gradio now supports MCP! If you want to convert an existing Space, like this one hexgrad/Kokoro-TTS, so that you can use it with Claude Desktop / Cursor / Cline / TinyAgents / or any LLM that supports MCP, here's all you need to do:
1. Duplicate the Space (in the Settings Tab)
2. Upgrade the Gradio
sdk_version to 5.28 (in the README.md)3. Set
mcp_server=True in launch()4. (Optionally) add docstrings to the function so that the LLM knows how to use it, like this:
def generate(text, speed=1):
"""
Convert text to speech audio.
Parameters:
text (str): The input text to be converted to speech.
speed (float, optional): Playback speed of the generated speech.That's it! Now your LLM will be able to talk to you 🤯
Post
2681
Hi folks! Excited to share a new feature from the Gradio team along with a tutorial.
If you don't already know, Gradio is an open-source Python library used to build interfaces for machine learning models. Beyond just creating UIs, Gradio also exposes API capabilities and now, Gradio apps can be launched Model Context Protocol (MCP) servers for LLMs.
If you already know how to use Gradio, there are only two additional things you need to do:
* Add standard docstrings to your function (these will be used to generate the descriptions for your tools for the LLM)
* Set
Here's a complete example (make sure you already have the latest version of Gradio installed):
This is a very simple example, but you can add the ability to generate Ghibli images or speak emotions to any LLM that supports MCP. Once you have an MCP running locally, you can copy-paste the same app to host it on [Hugging Face Spaces](https://huggingface.co/spaces/) as well.
All free and open-source of course! Full tutorial: https://www.gradio.app/guides/building-mcp-server-with-gradio
If you don't already know, Gradio is an open-source Python library used to build interfaces for machine learning models. Beyond just creating UIs, Gradio also exposes API capabilities and now, Gradio apps can be launched Model Context Protocol (MCP) servers for LLMs.
If you already know how to use Gradio, there are only two additional things you need to do:
* Add standard docstrings to your function (these will be used to generate the descriptions for your tools for the LLM)
* Set
mcp_server=True in launch()Here's a complete example (make sure you already have the latest version of Gradio installed):
import gradio as gr
def letter_counter(word, letter):
"""Count the occurrences of a specific letter in a word.
Args:
word: The word or phrase to analyze
letter: The letter to count occurrences of
Returns:
The number of times the letter appears in the word
"""
return word.lower().count(letter.lower())
demo = gr.Interface(
fn=letter_counter,
inputs=["text", "text"],
outputs="number",
title="Letter Counter",
description="Count how many times a letter appears in a word"
)
demo.launch(mcp_server=True)This is a very simple example, but you can add the ability to generate Ghibli images or speak emotions to any LLM that supports MCP. Once you have an MCP running locally, you can copy-paste the same app to host it on [Hugging Face Spaces](https://huggingface.co/spaces/) as well.
All free and open-source of course! Full tutorial: https://www.gradio.app/guides/building-mcp-server-with-gradio
- 2 replies
sayakpaul
authored
a
paper
about 1 month ago
Post
2852
We just shipped HiDream Image LoRA fine-tuning to diffusers🧨
HiDream's sota capabilities (and mit license) bring a lot of potential to explore with fine-tunes 🔥
- more upgrades and features soon!
- code, weights and config example 👇
🧶Yarn art lora: linoyts/HiDream-yarn-art-LoRA
code: https://github.com/huggingface/diffusers/blob/main/examples/dreambooth/README_hidream.md
HiDream's sota capabilities (and mit license) bring a lot of potential to explore with fine-tunes 🔥
- more upgrades and features soon!
- code, weights and config example 👇
🧶Yarn art lora: linoyts/HiDream-yarn-art-LoRA
code: https://github.com/huggingface/diffusers/blob/main/examples/dreambooth/README_hidream.md
- 2 replies
nouamanetazi
authored
a
paper
about 2 months ago
anton-l
authored
a
paper
about 2 months ago
pcuenq
authored
a
paper
about 2 months ago
reach-vb
authored
a
paper
about 2 months ago
Post
3827
JOURNEY TO 1 MILLION DEVELOPERS
5 years ago, we launched Gradio as a simple Python library to let researchers at Stanford easily demo computer vision models with a web interface.
Today, Gradio is used by >1 million developers each month to build and share AI web apps. This includes some of the most popular open-source projects of all time, like Automatic1111, Fooocus, Oobabooga’s Text WebUI, Dall-E Mini, and LLaMA-Factory.
How did we get here? How did Gradio keep growing in the very crowded field of open-source Python libraries? I get this question a lot from folks who are building their own open-source libraries. This post distills some of the lessons that I have learned over the past few years:
1. Invest in good primitives, not high-level abstractions
2. Embed virality directly into your library
3. Focus on a (growing) niche
4. Your only roadmap should be rapid iteration
5. Maximize ways users can consume your library's outputs
1. Invest in good primitives, not high-level abstractions
When we first launched Gradio, we offered only one high-level class (gr.Interface), which created a complete web app from a single Python function. We quickly realized that developers wanted to create other kinds of apps (e.g. multi-step workflows, chatbots, streaming applications), but as we started listing out the apps users wanted to build, we realized what we needed to do:
Read the rest here: https://x.com/abidlabs/status/1907886
5 years ago, we launched Gradio as a simple Python library to let researchers at Stanford easily demo computer vision models with a web interface.
Today, Gradio is used by >1 million developers each month to build and share AI web apps. This includes some of the most popular open-source projects of all time, like Automatic1111, Fooocus, Oobabooga’s Text WebUI, Dall-E Mini, and LLaMA-Factory.
How did we get here? How did Gradio keep growing in the very crowded field of open-source Python libraries? I get this question a lot from folks who are building their own open-source libraries. This post distills some of the lessons that I have learned over the past few years:
1. Invest in good primitives, not high-level abstractions
2. Embed virality directly into your library
3. Focus on a (growing) niche
4. Your only roadmap should be rapid iteration
5. Maximize ways users can consume your library's outputs
1. Invest in good primitives, not high-level abstractions
When we first launched Gradio, we offered only one high-level class (gr.Interface), which created a complete web app from a single Python function. We quickly realized that developers wanted to create other kinds of apps (e.g. multi-step workflows, chatbots, streaming applications), but as we started listing out the apps users wanted to build, we realized what we needed to do:
Read the rest here: https://x.com/abidlabs/status/1907886
sayakpaul
authored
a
paper
3 months ago
Post
6879
SmolVLM-2 and SigLIP-2 are now part of
They're added on top of the v4.49.0 release, and can be installed from the following tags:
This marks a new beginning for the release process of transformers. For the past five years, we've been doing monthly releases featuring many models (v4.49.0, the latest release, features 9 new architectures).
Starting with SmolVLM-2 & SigLIP2, we'll now additionally release tags supporting new models on a stable branch. These models are therefore directly available for use by installing from the tag itself. These tags will continue to be updated with fixes applied to these models.
Going forward, continue expecting software releases following semantic versioning: v4.50.0 will have ~10 new architectures compared to v4.49.0, as well as a myriad of new features, improvements and bug fixes. Accompanying these software releases, we'll release tags offering brand new models as fast as possible, to make them accessible to all immediately.
transformers in dedicated releases!They're added on top of the v4.49.0 release, and can be installed from the following tags:
v4.49.0-SmolVLM-2 and v4.49.0-SigLIP-2.This marks a new beginning for the release process of transformers. For the past five years, we've been doing monthly releases featuring many models (v4.49.0, the latest release, features 9 new architectures).
Starting with SmolVLM-2 & SigLIP2, we'll now additionally release tags supporting new models on a stable branch. These models are therefore directly available for use by installing from the tag itself. These tags will continue to be updated with fixes applied to these models.
Going forward, continue expecting software releases following semantic versioning: v4.50.0 will have ~10 new architectures compared to v4.49.0, as well as a myriad of new features, improvements and bug fixes. Accompanying these software releases, we'll release tags offering brand new models as fast as possible, to make them accessible to all immediately.
-
- 1 reply
Post
3857
Inference-time scaling meets Flux.1-Dev (and others) 🔥
Presenting a simple re-implementation of "Inference-time scaling diffusion models beyond denoising steps" by Ma et al.
I did the simplest random search strategy, but results can potentially be improved with better-guided search methods.
Supports Gemini 2 Flash & Qwen2.5 as verifiers for "LLMGrading" 🤗
The steps are simple:
For each round:
1> Starting by sampling 2 starting noises with different seeds.
2> Score the generations w.r.t a metric.
3> Obtain the best generation from the current round.
If you have more compute budget, go to the next search round. Scale the noise pool (
This constitutes the random search method as done in the paper by Google DeepMind.
Code, more results, and a bunch of other stuff are in the repository. Check it out here: https://github.com/sayakpaul/tt-scale-flux/ 🤗
Presenting a simple re-implementation of "Inference-time scaling diffusion models beyond denoising steps" by Ma et al.
I did the simplest random search strategy, but results can potentially be improved with better-guided search methods.
Supports Gemini 2 Flash & Qwen2.5 as verifiers for "LLMGrading" 🤗
The steps are simple:
For each round:
1> Starting by sampling 2 starting noises with different seeds.
2> Score the generations w.r.t a metric.
3> Obtain the best generation from the current round.
If you have more compute budget, go to the next search round. Scale the noise pool (
2 ** search_round) and repeat 1 - 3.This constitutes the random search method as done in the paper by Google DeepMind.
Code, more results, and a bunch of other stuff are in the repository. Check it out here: https://github.com/sayakpaul/tt-scale-flux/ 🤗
reach-vb
authored
a
paper
4 months ago
anton-l
authored
a
paper
4 months ago