Stephen Genusa's picture
5 12

Stephen Genusa PRO

StephenGenusa
·

AI & ML interests

LFM, LLM, Quantization, Vision, RAG/Hybrid/Graph, Multimodality, NLP (will take us further down the road with existing LLM tech)

Recent Activity

liked a Space 7 days ago
dcarpintero/pangolin
reacted to tomaarsen's post with ❤️ 7 days ago
‼️Sentence Transformers v4.0 is out! You can now train and finetune reranker models with multi-GPU training, bf16 support, loss logging, callbacks & much more. I also prove that finetuning on your domain helps much more than you might think. 1️⃣ Reranker Training Refactor Reranker models can now be trained using an extensive trainer with a lot of powerful features: - MultiGPU Training (Data Parallelism (DP) and Distributed Data Parallelism (DDP)) - bf16 training support; loss logging - Evaluation datasets + evaluation loss - Improved callback support + an excellent Weights & Biases integration - Gradient checkpointing, gradient accumulation - Model card generation - Resuming from a training checkpoint without performance loss - Hyperparameter Optimization and much more! Read my detailed blogpost to learn about the components that make up this new training approach: https://huggingface.co/blog/train-reranker Notably, the release is fully backwards compatible: all deprecations are soft, meaning that they still work but emit a warning informing you how to upgrade. 2️⃣ New Reranker Losses - 11 new losses: - 2 traditional losses: BinaryCrossEntropy and CrossEntropy - 2 distillation losses: MSE and MarginMSE - 2 in-batch negatives losses: MNRL (a.k.a. InfoNCE) and CMNRL - 5 learning to rank losses: Lambda, p-ListMLE, ListNet, RankNet, ListMLE 3️⃣ New Reranker Documentation - New Training Overview, Loss Overview, API Reference docs - 5 new, 1 refactored training examples docs pages - 13 new, 6 refactored training scripts - Migration guides (2.x -> 3.x, 3.x -> 4.x) 4️⃣ Blogpost Alongside the release, I've written a blogpost where I finetune ModernBERT on a generic question-answer dataset. My finetunes easily outperform all general-purpose reranker models, even models 4x as big. Finetuning on your domain is definitely worth it: https://huggingface.co/blog/train-reranker See the full release notes here: https://github.com/UKPLab/sentence-transformers/releases/v4.0.1
View all activity

Organizations

Social Post Explorers's profile picture

StephenGenusa's activity

reacted to bartowski's post with 👍 2 days ago
view post
Post
65612
Switching to author_model-name

I posted a poll on twitter, and others have mentioned the interest in me using the convention of including the author name in the model path when I upload.

It has a couple advantages, first and foremost of course is ensuring clarity of who uploaded the original model (did Qwen upload Qwen2.6? Or did someone fine tune Qwen2.5 and named it 2.6 for fun?)

The second thing is that it avoids collisions, so if multiple people upload the same model and I try to quant them both, I would normally end up colliding and being unable to upload both

I'll be implementing the change next week, there are just two final details I'm unsure about:

First, should the files also inherit the author's name?

Second, what to do in the case that the author name + model name pushes us past the character limit?

Haven't yet decided how to handle either case, so feedback is welcome, but also just providing this as a "heads up"
·
reacted to tomaarsen's post with ❤️ 7 days ago
view post
Post
2073
‼️Sentence Transformers v4.0 is out! You can now train and finetune reranker models with multi-GPU training, bf16 support, loss logging, callbacks & much more. I also prove that finetuning on your domain helps much more than you might think.

1️⃣ Reranker Training Refactor
Reranker models can now be trained using an extensive trainer with a lot of powerful features:
- MultiGPU Training (Data Parallelism (DP) and Distributed Data Parallelism (DDP))
- bf16 training support; loss logging
- Evaluation datasets + evaluation loss
- Improved callback support + an excellent Weights & Biases integration
- Gradient checkpointing, gradient accumulation
- Model card generation
- Resuming from a training checkpoint without performance loss
- Hyperparameter Optimization
and much more!

Read my detailed blogpost to learn about the components that make up this new training approach: https://huggingface.co/blog/train-reranker
Notably, the release is fully backwards compatible: all deprecations are soft, meaning that they still work but emit a warning informing you how to upgrade.

2️⃣ New Reranker Losses
- 11 new losses:
- 2 traditional losses: BinaryCrossEntropy and CrossEntropy
- 2 distillation losses: MSE and MarginMSE
- 2 in-batch negatives losses: MNRL (a.k.a. InfoNCE) and CMNRL
- 5 learning to rank losses: Lambda, p-ListMLE, ListNet, RankNet, ListMLE

3️⃣ New Reranker Documentation
- New Training Overview, Loss Overview, API Reference docs
- 5 new, 1 refactored training examples docs pages
- 13 new, 6 refactored training scripts
- Migration guides (2.x -> 3.x, 3.x -> 4.x)

4️⃣ Blogpost
Alongside the release, I've written a blogpost where I finetune ModernBERT on a generic question-answer dataset. My finetunes easily outperform all general-purpose reranker models, even models 4x as big. Finetuning on your domain is definitely worth it: https://huggingface.co/blog/train-reranker

See the full release notes here: https://github.com/UKPLab/sentence-transformers/releases/v4.0.1
posted an update 3 months ago
view post
Post
1207
I have a pro account and I am logged in. I have duplicated a space due to the error "You have exceeded your GPU quota", I am showing 0 GPU use, yet I am unable to use it "You have exceeded your GPU quota (60s requested vs. 44s left). Create a free account to get more daily usage quota." "Expert Support" is a pitch for consulting.
·
reacted to vincentg64's post with 🔥 3 months ago
view post
Post
2243
LLM 2.0, RAG & Non-Standard Gen AI on GitHub https://mltblog.com/3DsyZSq

In this article, I share my latest Gen AI and LLM advances, featuring innovative approaches radically different from both standard AI and classical ML/NLP. The focus is on doing better with less, using efficient architectures, new algorithms and evaluation metrics. It originates from research that I started long ago. It gained significant momentum in the last two years. See background and history at https://mltblog.com/4g2sKTv.

OpenAI, Perplexity, Anthropic, Llama and others typically follow the trend and implement solutions very similar to mines within 3 to 6 months after I publish new milestones. For instance, multi-tokens, knowledge graph tokens, multi-indexes, real-time fine-tuning, mixtures of experts, LLM routers, small enterprise sub-LLMs, prompt distillation, relevancy scoring engine, deep contextual retrieval, optimum agentic chunking, and modern UI instead of the basic prompt box. I keep adding new features all the time, staying ahead of competition.

➡️ Read full article with links to GitHub, at https://mltblog.com/3DsyZSq
  • 1 reply
·
reacted to m-ric's post with 🚀 6 months ago
view post
Post
1295
𝗔𝗱𝗱 𝘀𝗼𝘂𝗿𝗰𝗲 𝗵𝗶𝗴𝗵𝗹𝗶𝗴𝗵𝘁𝗶𝗻𝗴 𝘁𝗼 𝘆𝗼𝘂𝗿 𝗥𝗔𝗚 𝘀𝘆𝘀𝘁𝗲𝗺! 📄💡

RAG systems are supposed to make your LLM's answer more trustworthy, by inserting in the prompt some supporting documents from a knowledge base : we say that we're "adding some context".

👎 But if you don't know which part of the answer has been generated based on which input tokens, it's hard to tell wether it was effectively grounded in the context knowledge or not!

🤔 I've been working on the question: is it possible to add notes to the answer linking to which part of the context they're generated from?

And I've found a great solution: a great technique called Layer-wise Relevance Propagation (LRP), showcased in a paper at ICML `24 by Reduan Achtibat et al allows, allows to precisely score how important each input token was in generating your output! They've made it into a library called LXT.

📊 For each generated output token, LXT gives you attribution scores for each input token.

⚙️ So I've worked a bit more on aggregating these scores into meaningful spans between successive input and output tokens, and I finally obtained my desired result: RAG with source highlighting!

Try the demo here 👉 m-ric/rag_highlights

Caveats:
- It slows down generation (for now quite a lot, could hopefully be reduced a lot)
- For now it supports only specific models: Llama models and Mixtral

If there's enough interest in this solution, I can improve it further and spin it off into a specific library for RAG! 🚀
reacted to Wauplin's post with 🤗 6 months ago
view post
Post
3118
What a great milestone to celebrate! The huggingface_hub library is slowly becoming a cornerstone of the Python ML ecosystem when it comes to interacting with the @huggingface Hub. It wouldn't be there without the hundreds of community contributions and feedback! No matter if you are loading a model, sharing a dataset, running remote inference or starting jobs on our infra, you are for sure using it! And this is only the beginning so give a star if you wanna follow the project 👉 https://github.com/huggingface/huggingface_hub
  • 1 reply
·
New activity in mattshumer/ref_70_e3 7 months ago
replied to m-ric's post 7 months ago
view reply

I think there will be a big breakthrough as well, but I'd be surprised if it happens soon. If it does, I'd be happy. While the architectures of LLMs continue to advance I don't see any evidence that significant progress is being made and I personally think the architectures are too primitive and inherently self-limiting. I am also a believer that bigger does not necessarily mean better. I think we've reached the limits or are near the point of reaching the limits of where size dictates how powerful the LLM is.

Therefore, I think, given the current architectural limitations, the external limits, namely those dictated by power availability, and the many resources/costs of building better LLMs, will slow AI development until a radical change comes along.

We've managed to survive without them and now that we have them, they are a great step forward and we'll continue using and improving what we have. There are many improvements that can be made around the LLM using NLP to improve what we expect from LLMs and that's where the focus will turn for the time being, such as xLLM. Better architectures are going to have to take into account the difference in statistical models of representations of the world and the way humans communicate through speech and writing.

replied to vincentg64's post 7 months ago
view reply

Vincent, thank you for your time, effort and especially for your willingness to share your expertise. I am really looking forward to this!

reacted to vincentg64's post with ❤️ 7 months ago
view post
Post
1455
Hyperfast Contextual Custom LLM with Agents, Multitokens, Explainable AI, and Distillation https://mltblog.com/4dNPSnB

New additions to this ground-breaking system include multi-token distillation when processing prompts, agents to meet user intent, more NLP, and a command prompt menu accepting both standard prompts and various actions.

I also added several illustrations, featuring xLLM in action with a full session and sample commands to fine-tune in real-time. All the code, input sources (anonymized corporate corpus from fortune 100 company), contextual backend tables including embeddings, are on GitHub. My system has zero weight, no transformer, and no neural network. It relies on explainable AI, does not require training, is fully reproducible, and fits in memory. Yet your prompts can retrieve relevant full text entities from the corpus with no latency — including URLs, categories, titles, email addresses, and so on — thanks to well-designed architecture.

Read more, get the code, paper and everything for free, at https://mltblog.com/4dNPSnB
  • 2 replies
·
reacted to ybelkada's post with 🔥 8 months ago