Farid Hassainia
@fastpath.one
🚀 AI Advisor | Mentor | IceVision Co-Creator | fastpath.one
✨I help companies leverage AI and develop innovative products
🧩I assist mentees in becoming the top 1% of AI practitioners
✨I help companies leverage AI and develop innovative products
🧩I assist mentees in becoming the top 1% of AI practitioners
Reposted by Farid Hassainia
Hot take: shipping BERT-sized models in 2025 will benefit far more people than sharing an LLM overfitted to some saturated leaderboards 🙊
We're already seeing ModernBERT finetunes on the @hf.co Hub. My guess is we'll see hundreds of these by the end of 2025.
We're already seeing ModernBERT finetunes on the @hf.co Hub. My guess is we'll see hundreds of these by the end of 2025.
December 20, 2024 at 9:35 AM
Hot take: shipping BERT-sized models in 2025 will benefit far more people than sharing an LLM overfitted to some saturated leaderboards 🙊
We're already seeing ModernBERT finetunes on the @hf.co Hub. My guess is we'll see hundreds of these by the end of 2025.
We're already seeing ModernBERT finetunes on the @hf.co Hub. My guess is we'll see hundreds of these by the end of 2025.
Reposted by Farid Hassainia
you can now stay up-to-date with big AI research labs' updates on @hf.co easily over org activity page 🥹
I have been looking forward to this feature as I felt most back to back releases are overwhelming and I tend to miss out 🤠
I have been looking forward to this feature as I felt most back to back releases are overwhelming and I tend to miss out 🤠
December 20, 2024 at 1:09 PM
you can now stay up-to-date with big AI research labs' updates on @hf.co easily over org activity page 🥹
I have been looking forward to this feature as I felt most back to back releases are overwhelming and I tend to miss out 🤠
I have been looking forward to this feature as I felt most back to back releases are overwhelming and I tend to miss out 🤠
Reposted by Farid Hassainia
ModernBERT is available as a slot-in replacement for any BERT-like model, with both 139M param and 395M param sizes.
It has a 8192 sequence length, is extremely efficient, is uniquely great at analyzing code, and much more. Read this for details:
huggingface.co/blog/modern...
It has a 8192 sequence length, is extremely efficient, is uniquely great at analyzing code, and much more. Read this for details:
huggingface.co/blog/modern...
December 19, 2024 at 4:45 PM
ModernBERT is available as a slot-in replacement for any BERT-like model, with both 139M param and 395M param sizes.
It has a 8192 sequence length, is extremely efficient, is uniquely great at analyzing code, and much more. Read this for details:
huggingface.co/blog/modern...
It has a 8192 sequence length, is extremely efficient, is uniquely great at analyzing code, and much more. Read this for details:
huggingface.co/blog/modern...
Reposted by Farid Hassainia
Introducing 📐FineMath: the best open math pre-training dataset with 50B+ tokens!
Math remains challenging for LLMs and by training on FineMath we see considerable gains over other math datasets, especially on GSM8K and MATH.
🤗 huggingface.co/datasets/Hug...
Here’s a breakdown 🧵
Math remains challenging for LLMs and by training on FineMath we see considerable gains over other math datasets, especially on GSM8K and MATH.
🤗 huggingface.co/datasets/Hug...
Here’s a breakdown 🧵
December 19, 2024 at 3:55 PM
Introducing 📐FineMath: the best open math pre-training dataset with 50B+ tokens!
Math remains challenging for LLMs and by training on FineMath we see considerable gains over other math datasets, especially on GSM8K and MATH.
🤗 huggingface.co/datasets/Hug...
Here’s a breakdown 🧵
Math remains challenging for LLMs and by training on FineMath we see considerable gains over other math datasets, especially on GSM8K and MATH.
🤗 huggingface.co/datasets/Hug...
Here’s a breakdown 🧵
Reposted by Farid Hassainia
The first Gemini 2.0 Flash "Thinking Mode" model came out today. I've updated my llm-gemini plugin to support it and ran a few test prompts - it's a very interesting new entrant in the o1-inspired "inference scaling" category of models.
simonwillison.net/2024/Dec/19/...
simonwillison.net/2024/Dec/19/...
Gemini 2.0 Flash “Thinking mode”
Those new model releases just keep on flowing. Today it’s Google’s snappily named gemini-2.0-flash-thinking-exp, their first entrant into the o1-style inference scaling class of models. I posted about...
simonwillison.net
December 20, 2024 at 12:17 AM
The first Gemini 2.0 Flash "Thinking Mode" model came out today. I've updated my llm-gemini plugin to support it and ran a few test prompts - it's a very interesting new entrant in the o1-inspired "inference scaling" category of models.
simonwillison.net/2024/Dec/19/...
simonwillison.net/2024/Dec/19/...
Reposted by Farid Hassainia
ColPali is landed at @hf.co transformers and I have just shipped a very lean fine-tuning tutorial in smol-vision 🤠💗
QLoRA fine-tuning with 4-bit with bsz of 4 can be done with 32 GB VRAM and is very fast! ✨
github.com/merveenoyan/...
QLoRA fine-tuning with 4-bit with bsz of 4 can be done with 32 GB VRAM and is very fast! ✨
github.com/merveenoyan/...
December 20, 2024 at 3:53 PM
ColPali is landed at @hf.co transformers and I have just shipped a very lean fine-tuning tutorial in smol-vision 🤠💗
QLoRA fine-tuning with 4-bit with bsz of 4 can be done with 32 GB VRAM and is very fast! ✨
github.com/merveenoyan/...
QLoRA fine-tuning with 4-bit with bsz of 4 can be done with 32 GB VRAM and is very fast! ✨
github.com/merveenoyan/...
Reposted by Farid Hassainia
Introducing FineWeb-C 🌐🎓, a community-built dataset for improving language models in ALL languages.
Inspired by FineWeb-Edu the community is labelling the educational quality of texts for many languages.
318 annotators, 32K+ annotations, 12 languages - and growing!🌍
huggingface.co/datasets/dat...
Inspired by FineWeb-Edu the community is labelling the educational quality of texts for many languages.
318 annotators, 32K+ annotations, 12 languages - and growing!🌍
huggingface.co/datasets/dat...
December 20, 2024 at 4:53 PM
Introducing FineWeb-C 🌐🎓, a community-built dataset for improving language models in ALL languages.
Inspired by FineWeb-Edu the community is labelling the educational quality of texts for many languages.
318 annotators, 32K+ annotations, 12 languages - and growing!🌍
huggingface.co/datasets/dat...
Inspired by FineWeb-Edu the community is labelling the educational quality of texts for many languages.
318 annotators, 32K+ annotations, 12 languages - and growing!🌍
huggingface.co/datasets/dat...
Reposted by Farid Hassainia
Researchers used AI to generate 288 complete academic finance papers predicting stock returns, complete with plausible theoretical frameworks & citations. Each paper looks and reads as legit.
They did this to show how easy it now is to mass produce "credible" research. Academia isn't ready.
They did this to show how easy it now is to mass produce "credible" research. Academia isn't ready.
December 18, 2024 at 3:59 PM
Researchers used AI to generate 288 complete academic finance papers predicting stock returns, complete with plausible theoretical frameworks & citations. Each paper looks and reads as legit.
They did this to show how easy it now is to mass produce "credible" research. Academia isn't ready.
They did this to show how easy it now is to mass produce "credible" research. Academia isn't ready.
Reposted by Farid Hassainia
Aya by Cohere For AI can now see! 👀
C4AI community has built Maya 8B, a new open-source multilingual VLM built on SigLIP and Aya 8B 🌱 works on 8 languages! 🗣️
The authors extend Llava dataset using Aya's translation capabilities with 558k examples!
works very well ⬇️ huggingface.co/spaces/kkr51...
C4AI community has built Maya 8B, a new open-source multilingual VLM built on SigLIP and Aya 8B 🌱 works on 8 languages! 🗣️
The authors extend Llava dataset using Aya's translation capabilities with 558k examples!
works very well ⬇️ huggingface.co/spaces/kkr51...
December 18, 2024 at 11:37 AM
Aya by Cohere For AI can now see! 👀
C4AI community has built Maya 8B, a new open-source multilingual VLM built on SigLIP and Aya 8B 🌱 works on 8 languages! 🗣️
The authors extend Llava dataset using Aya's translation capabilities with 558k examples!
works very well ⬇️ huggingface.co/spaces/kkr51...
C4AI community has built Maya 8B, a new open-source multilingual VLM built on SigLIP and Aya 8B 🌱 works on 8 languages! 🗣️
The authors extend Llava dataset using Aya's translation capabilities with 558k examples!
works very well ⬇️ huggingface.co/spaces/kkr51...
Reposted by Farid Hassainia
Wrote up some notes on Microsoft's new Phi-4 LLM. They trained it on a LOT of synthetic data, and the details of how and why they did that are really interesting.
simonwillison.net/2024/Dec/15/...
simonwillison.net/2024/Dec/15/...
Phi-4 Technical Report
Phi-4 is the latest LLM from Microsoft Research. It has 14B parameters and claims to be a big leap forward in the overall Phi series. From [Introducing Phi-4: Microsoft’s Newest …
simonwillison.net
December 16, 2024 at 12:02 AM
Wrote up some notes on Microsoft's new Phi-4 LLM. They trained it on a LOT of synthetic data, and the details of how and why they did that are really interesting.
simonwillison.net/2024/Dec/15/...
simonwillison.net/2024/Dec/15/...
Reposted by Farid Hassainia
who's fine-tuning LLMs for reasoning? This dataset has been trending for a few weeks and there's a list of models trained on it.
- It has SFT formatted reasoning sequences, like those in o1.
- You could incorporate these into post training to boost reasoning abilities.
- It has SFT formatted reasoning sequences, like those in o1.
- You could incorporate these into post training to boost reasoning abilities.
O1-OPEN/OpenO1-SFT · Datasets at Hugging Face
We’re on a journey to advance and democratize artificial intelligence through open source and open science.
buff.ly
December 12, 2024 at 11:00 AM
who's fine-tuning LLMs for reasoning? This dataset has been trending for a few weeks and there's a list of models trained on it.
- It has SFT formatted reasoning sequences, like those in o1.
- You could incorporate these into post training to boost reasoning abilities.
- It has SFT formatted reasoning sequences, like those in o1.
- You could incorporate these into post training to boost reasoning abilities.
Reposted by Farid Hassainia
Learn how to build a complete multimodal RAG pipeline with
ColQwen2 as retriever, MonoQwen2-VL as reranker, Qwen2-VL as VLM in this notebook that runs on a GPU as small as L4 🔥 huggingface.co/learn/cookbo...
ColQwen2 as retriever, MonoQwen2-VL as reranker, Qwen2-VL as VLM in this notebook that runs on a GPU as small as L4 🔥 huggingface.co/learn/cookbo...
December 12, 2024 at 2:31 PM
Learn how to build a complete multimodal RAG pipeline with
ColQwen2 as retriever, MonoQwen2-VL as reranker, Qwen2-VL as VLM in this notebook that runs on a GPU as small as L4 🔥 huggingface.co/learn/cookbo...
ColQwen2 as retriever, MonoQwen2-VL as reranker, Qwen2-VL as VLM in this notebook that runs on a GPU as small as L4 🔥 huggingface.co/learn/cookbo...
Reposted by Farid Hassainia
Introducing TTS WebGPU: The first ever text-to-speech web app built with WebGPU acceleration! 🔥
High-quality and natural speech generation that runs 100% locally in your browser, powered by OuteTTS and Transformers.js. 🤗 Try it out yourself!
Demo + source code below 👇
High-quality and natural speech generation that runs 100% locally in your browser, powered by OuteTTS and Transformers.js. 🤗 Try it out yourself!
Demo + source code below 👇
December 8, 2024 at 7:38 PM
Introducing TTS WebGPU: The first ever text-to-speech web app built with WebGPU acceleration! 🔥
High-quality and natural speech generation that runs 100% locally in your browser, powered by OuteTTS and Transformers.js. 🤗 Try it out yourself!
Demo + source code below 👇
High-quality and natural speech generation that runs 100% locally in your browser, powered by OuteTTS and Transformers.js. 🤗 Try it out yourself!
Demo + source code below 👇
Reposted by Farid Hassainia
📊 After analyzing real production LLMOps data, here's what actually works for prompt engineering: structured prompts for reliability, systematic versioning for scale, and retrieval-augmented generation for efficiency. No theory—just battle-tested approaches.
December 11, 2024 at 9:55 AM
📊 After analyzing real production LLMOps data, here's what actually works for prompt engineering: structured prompts for reliability, systematic versioning for scale, and retrieval-augmented generation for efficiency. No theory—just battle-tested approaches.
Reposted by Farid Hassainia
The Open LLM Leaderboard got a new front page for Christmas
Check it out at huggingface.co/spaces/open-...
Check it out at huggingface.co/spaces/open-...
December 11, 2024 at 8:16 AM
The Open LLM Leaderboard got a new front page for Christmas
Check it out at huggingface.co/spaces/open-...
Check it out at huggingface.co/spaces/open-...
Reposted by Farid Hassainia
📊 New blog analyzing real-world LLM agent deployments in production. After studying dozens of case studies, the gap between demos and production is stark. Most "successful" implementations? Extremely narrow in scope with heavy human oversight.
December 9, 2024 at 8:57 AM
📊 New blog analyzing real-world LLM agent deployments in production. After studying dozens of case studies, the gap between demos and production is stark. Most "successful" implementations? Extremely narrow in scope with heavy human oversight.
Reposted by Farid Hassainia
Google dropped PaliGemma2. It's a vison language model starting at 2b parameters. Here's some notebooks finetuning it in:
@merve.bsky.social transformers with quantized LoRA: https://buff.ly/3CXfH7s
Jetha Chan brings keras LoRA https://buff.ly/4iBgprp & JAX https://buff.ly/4f5DLCv
@merve.bsky.social transformers with quantized LoRA: https://buff.ly/3CXfH7s
Jetha Chan brings keras LoRA https://buff.ly/4iBgprp & JAX https://buff.ly/4f5DLCv
smol-vision/Fine_tune_PaliGemma.ipynb at main · merveenoyan/smol-vision
Recipes for shrinking, optimizing, customizing cutting edge vision models. 💜 - merveenoyan/smol-vision
buff.ly
December 5, 2024 at 7:29 PM
Google dropped PaliGemma2. It's a vison language model starting at 2b parameters. Here's some notebooks finetuning it in:
@merve.bsky.social transformers with quantized LoRA: https://buff.ly/3CXfH7s
Jetha Chan brings keras LoRA https://buff.ly/4iBgprp & JAX https://buff.ly/4f5DLCv
@merve.bsky.social transformers with quantized LoRA: https://buff.ly/3CXfH7s
Jetha Chan brings keras LoRA https://buff.ly/4iBgprp & JAX https://buff.ly/4f5DLCv
Reposted by Farid Hassainia
I am holding open office hours on LLM Evals. I recorded the first one which was about evaluating multi-turn chats
Notes and recording here:
hamel.dev/notes/llm/of...
Notes and recording here:
hamel.dev/notes/llm/of...
Multi-Turn Chat Evals – Hamel’s Blog
Office hours discussion on multi-turn chat evals
hamel.dev
December 6, 2024 at 6:44 PM
I am holding open office hours on LLM Evals. I recorded the first one which was about evaluating multi-turn chats
Notes and recording here:
hamel.dev/notes/llm/of...
Notes and recording here:
hamel.dev/notes/llm/of...
Reposted by Farid Hassainia
🚀🚀PaliGemma 2 is our updated and improved PaliGemma release using the Gemma 2 models and providing new pre-trained checkpoints for the full cross product of {224px,448px,896px} resolutions and {3B,10B,28B} model sizes.
1/7
1/7
December 5, 2024 at 6:16 PM
🚀🚀PaliGemma 2 is our updated and improved PaliGemma release using the Gemma 2 models and providing new pre-trained checkpoints for the full cross product of {224px,448px,896px} resolutions and {3B,10B,28B} model sizes.
1/7
1/7
Reposted by Farid Hassainia
I can't begin to describe how life-changing this new project, ShellSage, has been for me over the last few weeks.
ShellSage is an LLM that lives in your terminal. It can see what directory you're in, what commands you've typed, what output you got, & your previous AI Q&A's.🧵
ShellSage is an LLM that lives in your terminal. It can see what directory you're in, what commands you've typed, what output you got, & your previous AI Q&A's.🧵
As R&D staff @ answer.ai, I work a lot on boosting productivity with AI. A common theme that always comes up is the combination of human+AI. This combination proved to be powerful in our new project ShellSage, which is an AI terminal buddy that learns and teaches with you. A 🧵
December 5, 2024 at 8:30 PM
I can't begin to describe how life-changing this new project, ShellSage, has been for me over the last few weeks.
ShellSage is an LLM that lives in your terminal. It can see what directory you're in, what commands you've typed, what output you got, & your previous AI Q&A's.🧵
ShellSage is an LLM that lives in your terminal. It can see what directory you're in, what commands you've typed, what output you got, & your previous AI Q&A's.🧵
Reposted by Farid Hassainia
A doctor’s letter to United Healthcare for denying nausea meds for a child on chemo
December 5, 2024 at 3:01 PM
A doctor’s letter to United Healthcare for denying nausea meds for a child on chemo
Reposted by Farid Hassainia
This is a neat new variant on RAG - no vectors, not even full-text search, instead showing the model a header hierarchy and giving it a tool to read the relevant sections
My notes here: simonwillison.net/2024/Dec/6/r...
My notes here: simonwillison.net/2024/Dec/6/r...
December 6, 2024 at 3:04 AM
This is a neat new variant on RAG - no vectors, not even full-text search, instead showing the model a header hierarchy and giving it a tool to read the relevant sections
My notes here: simonwillison.net/2024/Dec/6/r...
My notes here: simonwillison.net/2024/Dec/6/r...
Reposted by Farid Hassainia
Exponentially growing number of open-source AI models over the course of the past 30 months – from a few thousands to over 1 million and more
Interactive data viz: huggingface.co/spaces/huggi...
Interactive data viz: huggingface.co/spaces/huggi...
December 6, 2024 at 8:14 AM
Exponentially growing number of open-source AI models over the course of the past 30 months – from a few thousands to over 1 million and more
Interactive data viz: huggingface.co/spaces/huggi...
Interactive data viz: huggingface.co/spaces/huggi...
Reposted by Farid Hassainia
🔍 Ever wondered how companies like Instacart and eBay are scaling their embedding systems in production?
Here's what we learned from analyzing real-world implementations...
Here's what we learned from analyzing real-world implementations...
December 6, 2024 at 8:59 AM
🔍 Ever wondered how companies like Instacart and eBay are scaling their embedding systems in production?
Here's what we learned from analyzing real-world implementations...
Here's what we learned from analyzing real-world implementations...
Reposted by Farid Hassainia
Day 2 of smol course and the community is building something here.
👷 If you want to get involved, you can do this:
- read (and star) the repo
- check out our new discord channel
- open a PR to submit an exercise on module 1
- open an issue to improve the course
- review another submission
🧵
👷 If you want to get involved, you can do this:
- read (and star) the repo
- check out our new discord channel
- open a PR to submit an exercise on module 1
- open an issue to improve the course
- review another submission
🧵
December 5, 2024 at 8:48 AM
Day 2 of smol course and the community is building something here.
👷 If you want to get involved, you can do this:
- read (and star) the repo
- check out our new discord channel
- open a PR to submit an exercise on module 1
- open an issue to improve the course
- review another submission
🧵
👷 If you want to get involved, you can do this:
- read (and star) the repo
- check out our new discord channel
- open a PR to submit an exercise on module 1
- open an issue to improve the course
- review another submission
🧵