Post
1098
the french ministry of culture releases their first conversation datasets on huggingface ๐๐ป
ministere-culture/comparia-conversations
ministere-culture/comparia-conversations
AI & ML interests
None defined yet.
Post
790
COMPUTER CONTROL IS ON-DEVICE !
๐ก๐ค 78 % of EU smart-home owners DONโT trust cloud voice assistants.
So we killed the cloud.
Meet Extรฉ: a palm-sized Android device that sees, hears & speaks your language - 100 % offline, 0 % data sent anywhere.
๐ We submitted our technologies for consideration to the Liquid AI hackathon.
๐ Dataset: 79 k UI-action pairs on Hugging Face (largest Android-control corpus ever) Tonic/android-operator-episodes
โก Model: 98 % task accuracy, 678MB compressed , fits on existing android devices ! Tonic/l-android-control
๐ค๏ธ Experiment Tracker : check out the training on our TrackioApp Tonic/l-android-control
๐ฎ Live Model Demo: Upload an Android Screenshot and instructions to see the model in action ! Tonic/l-operator-demo
Built in a garage, funded by pre-orders, no VC. Now weโre scaling to 1 k installer units.
Weโre giving 50 limited-edition prototypes to investors , installers & researchers who want to co-design the sovereign smart home.
๐ Drop โEUSKERAโ in the comments if you want an invite, tag a friend who still thinks Alexa is โconvenient,โ and smash โฅ๏ธ if AI should belong to people - not servers.
๐ก๐ค 78 % of EU smart-home owners DONโT trust cloud voice assistants.
So we killed the cloud.
Meet Extรฉ: a palm-sized Android device that sees, hears & speaks your language - 100 % offline, 0 % data sent anywhere.
๐ We submitted our technologies for consideration to the Liquid AI hackathon.
๐ Dataset: 79 k UI-action pairs on Hugging Face (largest Android-control corpus ever) Tonic/android-operator-episodes
โก Model: 98 % task accuracy, 678MB compressed , fits on existing android devices ! Tonic/l-android-control
๐ค๏ธ Experiment Tracker : check out the training on our TrackioApp Tonic/l-android-control
๐ฎ Live Model Demo: Upload an Android Screenshot and instructions to see the model in action ! Tonic/l-operator-demo
Built in a garage, funded by pre-orders, no VC. Now weโre scaling to 1 k installer units.
Weโre giving 50 limited-edition prototypes to investors , installers & researchers who want to co-design the sovereign smart home.
๐ Drop โEUSKERAโ in the comments if you want an invite, tag a friend who still thinks Alexa is โconvenient,โ and smash โฅ๏ธ if AI should belong to people - not servers.
- 4 replies
Post
724
๐๐ปโโ๏ธ Hey there folks ,
Just wanted to annouce ๐ญSmolFactory : it's the quickest and best way to finetune SmolLM3 and GPT-OSS-20B on huggingface !
Basicaly it's an app you can run on huggingface by duplicating the space and running your training directly on huggingface GPUs .
It will help you basically select datasets and models, fine tune your model , make an experiment tracker you can use on your mobile phone , push all your model card and even automatically make a demo for you on huggingface so you can directly test it out when it's done !
check out the blog to learn more : https://huggingface.co/blog/Tonic/smolfactory
or just try the app directly :
Tonic/SmolFactory
you can vibe check the cool models I made :
French SmolLM3 : Tonic/Petite-LLM-3
Medical GPT-OSS : Tonic/med-gpt-oss-20b-demo
check out the model cards :
multilingual reasoner (gpt-oss) - Tonic/gpt-oss-20b-multilingual-reasoner
med-gpt-oss : Tonic/med-gpt-oss-20b
petite-elle-l-aime : Tonic/petite-elle-L-aime-3-sft
github repo if you like command line more than gradio : https://github.com/josephrp/smolfactory
drop some likes on these links it's really much appreciated !
feedback and PRs are welcome !
Just wanted to annouce ๐ญSmolFactory : it's the quickest and best way to finetune SmolLM3 and GPT-OSS-20B on huggingface !
Basicaly it's an app you can run on huggingface by duplicating the space and running your training directly on huggingface GPUs .
It will help you basically select datasets and models, fine tune your model , make an experiment tracker you can use on your mobile phone , push all your model card and even automatically make a demo for you on huggingface so you can directly test it out when it's done !
check out the blog to learn more : https://huggingface.co/blog/Tonic/smolfactory
or just try the app directly :
Tonic/SmolFactory
you can vibe check the cool models I made :
French SmolLM3 : Tonic/Petite-LLM-3
Medical GPT-OSS : Tonic/med-gpt-oss-20b-demo
check out the model cards :
multilingual reasoner (gpt-oss) - Tonic/gpt-oss-20b-multilingual-reasoner
med-gpt-oss : Tonic/med-gpt-oss-20b
petite-elle-l-aime : Tonic/petite-elle-L-aime-3-sft
github repo if you like command line more than gradio : https://github.com/josephrp/smolfactory
drop some likes on these links it's really much appreciated !
feedback and PRs are welcome !
ImranzamanMLย
posted
an
update
5 months ago
Post
650
# Runway Aleph: The Future of AI Video Editing
Runwayโs new **Aleph** model lets you *transform*, *edit*, and *generate* video from existing footage using just text prompts.
You can remove objects, change environments, restyle shots, alter lighting, and even create entirely new camera angles, all in one tool.
## Key Links
- ๐ฌ [Introducing Aleph (Runway Research)](https://runwayml.com/research/introducing-runway-aleph)
- ๐ [Aleph Prompting Guide (Runway Help Center)](https://help.runwayml.com/hc/en-us/articles/43277392678803-Aleph-Prompting-Guide)
- ๐ฌ [How to Transform Videos (Runway Academy)](https://academy.runwayml.com/aleph/how-to-transform-videos)
- ๐ฐ [Gadgets360 Coverage](https://www.gadgets360.com/ai/news/runway-aleph-ai-video-editing-generation-model-post-production-unveiled-8965180)
- ๐ฅ [YouTube Demo: ALEPH by Runway](https://www.youtube.com/watch?v=PPerCtyIKwA)
- ๐ฐ [Runway Alpha dataset]( Rapidata/text-2-video-human-preferences-runway-alpha)
## Prompt Tips
1. Be clear and specific (e.g., _โChange to snowy night, keep people unchangedโ_).
2. Use action verbs like _add, remove, restyle, relight_.
3. Add reference images for style or lighting.
Aleph shifts AI video from *text-to-video* to *video-to-video*, making post-production faster, more creative, and more accessible than ever.
Runwayโs new **Aleph** model lets you *transform*, *edit*, and *generate* video from existing footage using just text prompts.
You can remove objects, change environments, restyle shots, alter lighting, and even create entirely new camera angles, all in one tool.
## Key Links
- ๐ฌ [Introducing Aleph (Runway Research)](https://runwayml.com/research/introducing-runway-aleph)
- ๐ [Aleph Prompting Guide (Runway Help Center)](https://help.runwayml.com/hc/en-us/articles/43277392678803-Aleph-Prompting-Guide)
- ๐ฌ [How to Transform Videos (Runway Academy)](https://academy.runwayml.com/aleph/how-to-transform-videos)
- ๐ฐ [Gadgets360 Coverage](https://www.gadgets360.com/ai/news/runway-aleph-ai-video-editing-generation-model-post-production-unveiled-8965180)
- ๐ฅ [YouTube Demo: ALEPH by Runway](https://www.youtube.com/watch?v=PPerCtyIKwA)
- ๐ฐ [Runway Alpha dataset]( Rapidata/text-2-video-human-preferences-runway-alpha)
## Prompt Tips
1. Be clear and specific (e.g., _โChange to snowy night, keep people unchangedโ_).
2. Use action verbs like _add, remove, restyle, relight_.
3. Add reference images for style or lighting.
Aleph shifts AI video from *text-to-video* to *video-to-video*, making post-production faster, more creative, and more accessible than ever.
ImranzamanMLย
posted
an
update
5 months ago
Post
834
OpenAI has launched GPT-5, a significant leap forward in AI technology that is now available to all users. The new model unifies all of OpenAI's previous developments into a single, cohesive system that automatically adapts its approach based on the complexity of the user's request. This means it can prioritize speed for simple queries or engage a deeper reasoning model for more complex problems, all without the user having to manually switch settings.
Key Features and Improvements
Unified System: GPT-5 combines various models into one interface, intelligently selecting the best approach for each query.
Enhanced Coding: It's being hailed as the "strongest coding model to date," with the ability to create complex, responsive websites and applications from a single prompt.
PhD-level Reasoning: According to CEO Sam Altman, GPT-5 offers a significant jump in reasoning ability, with a much lower hallucination rate. It also performs better on academic and human-evaluated benchmarks.
New Personalities: Users can now select from four preset personalitiesโCynic, Robot, Listener and Nerd to customize their chat experience.
Advanced Voice Mode: The voice mode has been improved to sound more natural and adapt its speech based on the context of the conversation.
https://openai.com/index/introducing-gpt-5/
https://openai.com/gpt-5/
Key Features and Improvements
Unified System: GPT-5 combines various models into one interface, intelligently selecting the best approach for each query.
Enhanced Coding: It's being hailed as the "strongest coding model to date," with the ability to create complex, responsive websites and applications from a single prompt.
PhD-level Reasoning: According to CEO Sam Altman, GPT-5 offers a significant jump in reasoning ability, with a much lower hallucination rate. It also performs better on academic and human-evaluated benchmarks.
New Personalities: Users can now select from four preset personalitiesโCynic, Robot, Listener and Nerd to customize their chat experience.
Advanced Voice Mode: The voice mode has been improved to sound more natural and adapt its speech based on the context of the conversation.
https://openai.com/index/introducing-gpt-5/
https://openai.com/gpt-5/
ImranzamanMLย
posted
an
update
5 months ago
Post
303
All key links to OpenAI open sourced GPT OSS models (117B and 21B) which are released under apache 2.0. Here is a quick guide to explore and build with them:
Intro & vision: https://openai.com/index/introducing-gpt-oss
Model specs & license: https://openai.com/index/gpt-oss-model-card/
Dev overview: https://cookbook.openai.com/topic/gpt-oss
How to run via vLLM: https://cookbook.openai.com/articles/gpt-oss/run-vllm
Harmony I/O format: https://github.com/openai/harmony
Reference PyTorch code: https://github.com/openai/gpt-oss?tab=readme-ov-file#reference-pytorch-implementation
Community site: https://gpt-oss.com/
Lets deep dive with OpenAI models now ๐
#OpenSource #AI #GPTOSS #OpenAI #LLM #Python #GenAI
Intro & vision: https://openai.com/index/introducing-gpt-oss
Model specs & license: https://openai.com/index/gpt-oss-model-card/
Dev overview: https://cookbook.openai.com/topic/gpt-oss
How to run via vLLM: https://cookbook.openai.com/articles/gpt-oss/run-vllm
Harmony I/O format: https://github.com/openai/harmony
Reference PyTorch code: https://github.com/openai/gpt-oss?tab=readme-ov-file#reference-pytorch-implementation
Community site: https://gpt-oss.com/
Lets deep dive with OpenAI models now ๐
#OpenSource #AI #GPTOSS #OpenAI #LLM #Python #GenAI
ImranzamanMLย
posted
an
update
5 months ago
Post
3524
Finaly OpenAI is open to share open-source models after GPT2-2019.
gpt-oss-120b
gpt-oss-20b
openai/gpt-oss-120b
#AI #GPT #LLM #Openai
gpt-oss-120b
gpt-oss-20b
openai/gpt-oss-120b
#AI #GPT #LLM #Openai
- 1 reply
Post
3629
๐ซก I am the first and only one to like the French Tax Code Dataset
that's it , that's the post
find the dataset here : louisbrulenaudet/code-impots
follow : @louisbrulenaudet
that's it , that's the post
find the dataset here : louisbrulenaudet/code-impots
follow : @louisbrulenaudet
-
- 3 replies
ImranzamanMLย
posted
an
update
5 months ago
Post
318
Working of Transformer model layers!
I focused on showing the core steps side by side with tokenization, embedding and the transformer model layers, each highlighting the self attention and feedforward parts without getting lost in too much technical depth.
Its showing how these layers work together to understand context and generate meaningful output!
If you are curious about the architecture behind AI language models or want a clean way to explain it, hit me up, Iโd love to share!
#AI #MachineLearning #NLP #Transformers #DeepLearning #DataScience #LLM #AIAgents
I focused on showing the core steps side by side with tokenization, embedding and the transformer model layers, each highlighting the self attention and feedforward parts without getting lost in too much technical depth.
Its showing how these layers work together to understand context and generate meaningful output!
If you are curious about the architecture behind AI language models or want a clean way to explain it, hit me up, Iโd love to share!
#AI #MachineLearning #NLP #Transformers #DeepLearning #DataScience #LLM #AIAgents
ImranzamanMLย
posted
an
update
5 months ago
Post
1682
Hugging Face just made life easier with the new hf CLI!
huggingface-cli to hf
With renaming the CLI, there are new features added like hf jobs. We can now run any script or Docker image on dedicated Hugging Face infrastructure with a simple command. It's a good addition for running experiments and jobs on the fly.
To get started, just run:
pip install -U huggingface_hub
List of hf CLI Commands
Main Commands
hf auth: Manage authentication (login, logout, etc.).
hf cache: Manage the local cache directory.
hf download: Download files from the Hub.
hf jobs: Run and manage Jobs on the Hub.
hf repo: Manage repos on the Hub.
hf upload: Upload a file or a folder to the Hub.
hf version: Print information about the hf version.
hf env: Print information about the environment.
Authentication Subcommands (hf auth)
login: Log in using a Hugging Face token.
logout: Log out of your account.
whoami: See which account you are logged in as.
switch: Switch between different stored access tokens/profiles.
list: List all stored access tokens.
Jobs Subcommands (hf jobs)
run: Run a Job on Hugging Face infrastructure.
inspect: Display detailed information on one or more Jobs.
logs: Fetch the logs of a Job.
ps: List running Jobs.
cancel: Cancel a Job.
hashtag#HuggingFace hashtag#MachineLearning hashtag#AI hashtag#DeepLearning hashtag#MLTools hashtag#MLOps hashtag#OpenSource hashtag#Python hashtag#DataScience hashtag#DevTools hashtag#LLM hashtag#hfCLI hashtag#GenerativeAI
huggingface-cli to hf
With renaming the CLI, there are new features added like hf jobs. We can now run any script or Docker image on dedicated Hugging Face infrastructure with a simple command. It's a good addition for running experiments and jobs on the fly.
To get started, just run:
pip install -U huggingface_hub
List of hf CLI Commands
Main Commands
hf auth: Manage authentication (login, logout, etc.).
hf cache: Manage the local cache directory.
hf download: Download files from the Hub.
hf jobs: Run and manage Jobs on the Hub.
hf repo: Manage repos on the Hub.
hf upload: Upload a file or a folder to the Hub.
hf version: Print information about the hf version.
hf env: Print information about the environment.
Authentication Subcommands (hf auth)
login: Log in using a Hugging Face token.
logout: Log out of your account.
whoami: See which account you are logged in as.
switch: Switch between different stored access tokens/profiles.
list: List all stored access tokens.
Jobs Subcommands (hf jobs)
run: Run a Job on Hugging Face infrastructure.
inspect: Display detailed information on one or more Jobs.
logs: Fetch the logs of a Job.
ps: List running Jobs.
cancel: Cancel a Job.
hashtag#HuggingFace hashtag#MachineLearning hashtag#AI hashtag#DeepLearning hashtag#MLTools hashtag#MLOps hashtag#OpenSource hashtag#Python hashtag#DataScience hashtag#DevTools hashtag#LLM hashtag#hfCLI hashtag#GenerativeAI
- 1 reply
ImranzamanMLย
authored
a
paper
5 months ago
Post
824
๐ Hey there folks,
just submitted my plugin idea to the G-Assist Plugin Hackathon by @nvidia . Check it out, it's a great way to use a local SLA model on a windows machine to easily and locally get things done ! https://github.com/NVIDIA/G-Assist
just submitted my plugin idea to the G-Assist Plugin Hackathon by @nvidia . Check it out, it's a great way to use a local SLA model on a windows machine to easily and locally get things done ! https://github.com/NVIDIA/G-Assist
Post
640
๐๐ปโโ๏ธ Hey there folks ,
Yesterday , Nvidia released a reasoning model that beats o3 on science, math and coding !
Today you can try it out here : Tonic/Nvidia-OpenReasoning
hope you like it !
Yesterday , Nvidia released a reasoning model that beats o3 on science, math and coding !
Today you can try it out here : Tonic/Nvidia-OpenReasoning
hope you like it !
Post
3398
๐๐ปโโ๏ธ Normalize adding compute & runtime traces to your model cards
- 2 replies
Post
543
Who's going to Raise Summit in Paris Tomorrow ?
If you're around , I would love to meet you :-)
If you're around , I would love to meet you :-)
Post
718
๐๐ปโโ๏ธ hey there folks ,
So every bio/med/chem meeting i go to i always the same questions "why are you sharing a gdrive link with me for this?" and "Do you have any plans to publish your model weights and datasets on huggingface?" and finally i got a good answer today which explains everything :
basically there is some kind of government censorship on this (usa, but i'm sure others too) and they are told they are not allowed as it is considered a "dataleak" which is illegal !!!!
this is terrible ! but the good news is that we can do something about it !
so there is this "call for opinions and comments" here from the NIH (usa) , and here we can make our opinion on this topic known : https://osp.od.nih.gov/comment-form-responsibly-developing-and-sharing-generative-artificial-intelligence-tools-using-nih-controlled-access-data/
kindly consider dropping your opinion and thoughts about this censorship of science , and share this post , link or thoughts widely .
Together maybe we can start to share data and model weights appropriately and openly in a good way ๐๐ป๐
cc. @cyrilzakka
So every bio/med/chem meeting i go to i always the same questions "why are you sharing a gdrive link with me for this?" and "Do you have any plans to publish your model weights and datasets on huggingface?" and finally i got a good answer today which explains everything :
basically there is some kind of government censorship on this (usa, but i'm sure others too) and they are told they are not allowed as it is considered a "dataleak" which is illegal !!!!
this is terrible ! but the good news is that we can do something about it !
so there is this "call for opinions and comments" here from the NIH (usa) , and here we can make our opinion on this topic known : https://osp.od.nih.gov/comment-form-responsibly-developing-and-sharing-generative-artificial-intelligence-tools-using-nih-controlled-access-data/
kindly consider dropping your opinion and thoughts about this censorship of science , and share this post , link or thoughts widely .
Together maybe we can start to share data and model weights appropriately and openly in a good way ๐๐ป๐
cc. @cyrilzakka
Post
2570
๐๐ปโโ๏ธ Hey there folks ,
Yesterday the world's first "Learn to Vibe Code" application was released .
As vibe coding is the mainstream paradigm , so now the first educational app is there to support it .
You can try it out already :
https://vibe.takara.ai
and of course it's entirely open source, so i already made my issue and feature branch :-) ๐
Yesterday the world's first "Learn to Vibe Code" application was released .
As vibe coding is the mainstream paradigm , so now the first educational app is there to support it .
You can try it out already :
https://vibe.takara.ai
and of course it's entirely open source, so i already made my issue and feature branch :-) ๐
ImranzamanMLย
posted
an
update
8 months ago
Post
674
Run LLM model Locally using Docker right inside your codebase (No GUI Needed!)
In this project, I did not used the suporting GUI like Open WebUI or LM Studio or any other, so the purpose to use stand alone LLM models with ollama to give you the idea that how you can use it in your project/code instead of running through third party. Everything is containerized with Docker, so setup is clean and repeatable. Its just a fun side project so my connections can learn more about running models locally in their own projects.
Tech stack used:
๐ Docker
๐ฆ LLaMA via Ollama
๐ป HTML/CSS/JS
๐ Python + FastAPI
๐ NGINX
Its still early and a fun side project, but if you are into local model deployment, or just want to see how it works, check it out on the given link!
https://github.com/Imran-ml/llama-chatbot-dockerized
#LLM #Docker #OpenSource #Chatbot #LLaMA #fastapi
In this project, I did not used the suporting GUI like Open WebUI or LM Studio or any other, so the purpose to use stand alone LLM models with ollama to give you the idea that how you can use it in your project/code instead of running through third party. Everything is containerized with Docker, so setup is clean and repeatable. Its just a fun side project so my connections can learn more about running models locally in their own projects.
Tech stack used:
๐ Docker
๐ฆ LLaMA via Ollama
๐ป HTML/CSS/JS
๐ Python + FastAPI
๐ NGINX
Its still early and a fun side project, but if you are into local model deployment, or just want to see how it works, check it out on the given link!
https://github.com/Imran-ml/llama-chatbot-dockerized
#LLM #Docker #OpenSource #Chatbot #LLaMA #fastapi
ImranzamanMLย
authored
a
paper
8 months ago
ImranzamanMLย
posted
an
update
8 months ago
Post
2916
๐ New paper out: "Improving Arabic Multi-Label Emotion Classification using Stacked Embeddings and Hybrid Loss Function"
Improving Arabic Multi-Label Emotion Classification using Stacked Embeddings and Hybrid Loss Function (2410.03979)
In this work, we tackle some major challenges in Arabic multi-label emotion classification especially the issues of class imbalance and label correlation that often hurt model performance, particularly for minority emotions.
Our approach:
Stacked contextual embeddings from fine-tuned ArabicBERT, MarBERT, and AraBERT models.
A meta-learning strategy that builds richer representations.
A hybrid loss function combining class weighting, label correlation matrices, and contrastive learning to better handle class imbalances.
๐ง Model pipeline: stacked embeddings โ meta-learner โ Bi-LSTM โ fully connected network โ multi-label classification.
๐ Extensive experiments show significant improvements across Precision, Recall, F1-Score, Jaccard Accuracy, and Hamming Loss.
๐ The hybrid loss function in particular helped close the gap between majority and minority classes!
We also performed ablation studies to break down each componentโs contribution and the results consistently validated our design choices.
This framework isn't just for Arabic it offers a generalizable path for improving multi-label emotion classification in other low-resource languages and domains.
Big thanks to my co-authors: Muhammad Azeem Aslam, Wang Jun, Nisar Ahmed, Li Yanan, Hu Hongfei, Wang Shiyu, and Xin Liu!
Would love to hear your thoughts on this work! ๐
Improving Arabic Multi-Label Emotion Classification using Stacked Embeddings and Hybrid Loss Function (2410.03979)
In this work, we tackle some major challenges in Arabic multi-label emotion classification especially the issues of class imbalance and label correlation that often hurt model performance, particularly for minority emotions.
Our approach:
Stacked contextual embeddings from fine-tuned ArabicBERT, MarBERT, and AraBERT models.
A meta-learning strategy that builds richer representations.
A hybrid loss function combining class weighting, label correlation matrices, and contrastive learning to better handle class imbalances.
๐ง Model pipeline: stacked embeddings โ meta-learner โ Bi-LSTM โ fully connected network โ multi-label classification.
๐ Extensive experiments show significant improvements across Precision, Recall, F1-Score, Jaccard Accuracy, and Hamming Loss.
๐ The hybrid loss function in particular helped close the gap between majority and minority classes!
We also performed ablation studies to break down each componentโs contribution and the results consistently validated our design choices.
This framework isn't just for Arabic it offers a generalizable path for improving multi-label emotion classification in other low-resource languages and domains.
Big thanks to my co-authors: Muhammad Azeem Aslam, Wang Jun, Nisar Ahmed, Li Yanan, Hu Hongfei, Wang Shiyu, and Xin Liu!
Would love to hear your thoughts on this work! ๐