Nishith Jain's picture

Nishith Jain

KingNish

AI & ML interests

AI is fun actually. Busy till June 2025.

Recent Activity

reacted to burtenshaw's post with 🤗 about 19 hours ago
everybody and their dog is fine-tuning Gemma 3 today, so I thought I'd do a longer post on the tips and sharp edges I find. let's go! 1. has to be install everything form main and nightly. this is what I'm working with to get unsloth and TRL running ```txt git+https://github.com/huggingface/transformers@main git+https://github.com/huggingface/trl.git@main bitsandbytes peft ``` plus this with `--no-deps` ```txt git+https://github.com/unslothai/unsloth-zoo.git@nightly git+https://github.com/unslothai/unsloth.git@nightly ``` 2. will brown's code to turn GSM8k into a reasoning dataset is a nice toy experiment https://gist.github.com/willccbb/4676755236bb08cab5f4e54a0475d6fb 3. with a learning rate of 5e-6 rewards and loss stayed flat for the first 100 or so steps. 4. so far none of my runs have undermined the outputs after 1 epoch. therefore, I'm mainly experimenting with bigger LoRA adapters. ```python from trl import GRPOConfig training_args = GRPOConfig( learning_rate = 5e-6, adam_beta1 = 0.9, adam_beta2 = 0.99, weight_decay = 0.1, warmup_ratio = 0.1, lr_scheduler_type = "cosine", optim = "adamw_8bit", logging_steps = 1, per_device_train_batch_size = 2, gradient_accumulation_steps = 1, num_generations = 2, max_prompt_length = 256, max_completion_length = 1024 - 256, num_train_epochs = 1, max_steps = 250, save_steps = 250, max_grad_norm = 0.1, report_to = "none", ) ``` 5. vision fine-tuning isn't available in TRL's GRPOTrainer, so stick to text datasets. but no need to load the model differently in transformers or Unsloth ```python from transformers import AutoModelForImageTextToText model = AutoModelForImageTextToText.from_pretrained("google/gemma-3-4b-it) ``` if you want an introduction to GRPO, check out the reasoning course, it walks you through the algorithm, theory, and implementation in a smooth way. https://huggingface.co/reasoning-course
View all activity

Organizations

Wikimedia's profile picture OpenGVLab's profile picture Blog-explorers's profile picture Multi🤖Transformers's profile picture The Collectionists's profile picture HelpingAI's profile picture ZeroGPU Explorers's profile picture Project Fluently's profile picture Poscye's profile picture INNOVA AI's profile picture Narra's profile picture Social Post Explorers's profile picture Cognitive Computations's profile picture Dev Mode Explorers's profile picture Stable Diffusion Community (Unofficial, Non-profit)'s profile picture ONNX Community's profile picture Hugging Face Discord Community's profile picture Nerdy Face's profile picture grafite's profile picture None yet's profile picture Project R's profile picture Doge Face's profile picture

KingNish's activity

reacted to AdinaY's post with 🔥🤗 about 19 hours ago
reacted to burtenshaw's post with 🤗 about 19 hours ago
view post
Post
767
everybody and their dog is fine-tuning Gemma 3 today, so I thought I'd do a longer post on the tips and sharp edges I find. let's go!

1. has to be install everything form main and nightly. this is what I'm working with to get unsloth and TRL running

git+https://github.com/huggingface/transformers@main
git+https://github.com/huggingface/trl.git@main
bitsandbytes
peft


plus this with --no-deps

git+https://github.com/unslothai/unsloth-zoo.git@nightly
git+https://github.com/unslothai/unsloth.git@nightly


2. will brown's code to turn GSM8k into a reasoning dataset is a nice toy experiment https://gist.github.com/willccbb/4676755236bb08cab5f4e54a0475d6fb

3. with a learning rate of 5e-6 rewards and loss stayed flat for the first 100 or so steps.

4. so far none of my runs have undermined the outputs after 1 epoch. therefore, I'm mainly experimenting with bigger LoRA adapters.

from trl import GRPOConfig

training_args = GRPOConfig(
    learning_rate = 5e-6,
    adam_beta1 = 0.9,
    adam_beta2 = 0.99,
    weight_decay = 0.1,
    warmup_ratio = 0.1,
    lr_scheduler_type = "cosine",
    optim = "adamw_8bit",
    logging_steps = 1,
    per_device_train_batch_size = 2,
    gradient_accumulation_steps = 1,
    num_generations = 2,
    max_prompt_length = 256,
    max_completion_length = 1024 - 256,
    num_train_epochs = 1,
    max_steps = 250,
    save_steps = 250,
    max_grad_norm = 0.1,
    report_to = "none",
)


5. vision fine-tuning isn't available in TRL's GRPOTrainer, so stick to text datasets. but no need to load the model differently in transformers or Unsloth

from transformers import AutoModelForImageTextToText

model = AutoModelForImageTextToText.from_pretrained("google/gemma-3-4b-it)


if you want an introduction to GRPO, check out the reasoning course, it walks you through the algorithm, theory, and implementation in a smooth way.

https://huggingface.co/reasoning-course
  • 2 replies
·
reacted to thomwolf's post with 🔥 1 day ago
view post
Post
1560
We've kept pushing our Open-R1 project, an open initiative to replicate and extend the techniques behind DeepSeek-R1.

And even we were mind-blown by the results we got with this latest model we're releasing: ⚡️OlympicCoder ( open-r1/OlympicCoder-7B and open-r1/OlympicCoder-32B)

It's beating Claude 3.7 on (competitive) programming –a domain Anthropic has been historically really strong at– and it's getting close to o1-mini/R1 on olympiad level coding with just 7B parameters!

And the best part is that we're open-sourcing all about its training dataset, the new IOI benchmark, and more in our Open-R1 progress report #3: https://huggingface.co/blog/open-r1/update-3

Datasets are are releasing:
- open-r1/codeforces
- open-r1/codeforces-cots
- open-r1/ioi
- open-r1/ioi-test-cases
- open-r1/ioi-sample-solutions
- open-r1/ioi-cots
- open-r1/ioi-2024-model-solutions
reacted to BrigitteTousi's post with 🤗 1 day ago
view post
Post
3596
Regardless of X being down or not, so glad I can rely on HF Posts for AI news ❤️🤗
  • 1 reply
·
reacted to Smooke's post with 👍 2 days ago
view post
Post
1789
Hallucinations Blog Research Reading List:

Hallucinations Are A Feature of AI, Humans Are The Bug https://hackernoon.com/hallucinations-are-a-feature-of-ai-humans-are-the-bug

Overcome LLM Hallucinations Using Knowledge Bases https://hackernoon.com/overcome-llm-hallucinations-using-knowledge-bases

How to Detect and Minimise Hallucinations in AI Models https://hackernoon.com/how-to-detect-and-minimise-hallucinations-in-ai-models

Predictive Coding, AI: Modeling Placebos in RCTs for Psychedelics and Antidepressants https://hackernoon.com/predictive-coding-ai-modeling-placebos-in-rcts-for-psychedelics-and-antidepressants

A Simple Method to Improving the Accuracy of Your RAG System https://hackernoon.com/say-goodbye-to-ai-hallucinations-a-simple-method-to-improving-the-accuracy-of-your-rag-system

Gen AI Hallucinations: The Good, the Bad, and the Costly https://hackernoon.com/gen-ai-hallucinations-the-good-the-bad-and-the-costly

Why Do LLMs Hallucinate? https://hackernoon.com/why-do-llms-hallucinate

Truth Serum For The AI Age: Factiverse To Fight Fake News And Hallucinations https://hackernoon.com/truth-serum-for-the-ai-age-factiverse-to-fight-fake-news-and-hallucinations

A Secret Technique To Sidestepping LLM Hallucinations https://hackernoon.com/a-secret-technique-to-sidestepping-llm-hallucinations

The Importance of Explainability in AI (XAI) https://hackernoon.com/tackling-ai-hallucinations-the-importance-of-explainability-in-ai-xai

What You Need to Know About Amazon Bedrock’s RAG Evaluation and LLM-as-a-Judge for Advancing AI https://hackernoon.com/what-you-need-to-know-about-amazon-bedrocks-rag-evaluation-and-llm-as-a-judge-for-advancing-ai

I Over Relied on AI and Those Shortcuts Cost Me https://hackernoon.com/i-over-relied-on-ai-and-those-shortcuts-cost-me

AI’s Non-Determinism, Hallucinations, And... Cats? https://hackernoon.com/ais-non-determinism-hallucinations-and-cats

More to read --> https://hackernoon.com/search?query=hallucinations

reacted to JingzeShi's post with 🚀❤️ 4 days ago
reacted to BlinkDL's post with 🔥 4 days ago
view post
Post
4960
RWKV-7 "Goose" 0.4B trained w/ ctx4k automatically extrapolates to ctx32k+, and perfectly solves NIAH ctx16k 🤯 100% RNN and attention-free. Only trained on the Pile. No finetuning. Replicable training runs. tested by our community: https://github.com/Jellyfish042/LongMamba
reacted to fdaudens's post with 🤗 4 days ago
view post
Post
5650
Honored to be named among their 12 pioneers and power players in the news industry in the 2025 Tech Trends Report from Future Today Strategy Group.

Incredible group to be part of - each person is doing groundbreaking work at the intersection of AI and journalism. Worth following them all: they're consistently sharing practical insights on building the future of news.

Take the time to read this report, it's packed with insights as always. The news & information section's #1 insight hits hard: "The most substantive economic impact of AI to date has been licensing payouts for a handful of big publishers. The competition will start shifting in the year ahead to separate AI 'haves' that have positioned themselves to grow from the 'have-nots.'"

This AI-driven divide is something I've been really concerned about. Now is the time to build more than ever!

👉 Full report here: https://ftsg.com/wp-content/uploads/2025/03/FTSG_2025_TR_FINAL_LINKED.pdf
  • 2 replies
·
reacted to as-cle-bert's post with 👍 6 days ago
view post
Post
2627
I just released a fully automated evaluation framework for your RAG applications!📈

GitHub 👉 https://github.com/AstraBert/diRAGnosis
PyPi 👉 https://pypi.org/project/diragnosis/

It's called 𝐝𝐢𝐑𝐀𝐆𝐧𝐨𝐬𝐢𝐬 and is a lightweight framework that helps you 𝗱𝗶𝗮𝗴𝗻𝗼𝘀𝗲 𝘁𝗵𝗲 𝗽𝗲𝗿𝗳𝗼𝗿𝗺𝗮𝗻𝗰𝗲 𝗼𝗳 𝗟𝗟𝗠𝘀 𝗮𝗻𝗱 𝗿𝗲𝘁𝗿𝗶𝗲𝘃𝗮𝗹 𝗺𝗼𝗱𝗲𝗹𝘀 𝗶𝗻 𝗥𝗔𝗚 𝗮𝗽𝗽𝗹𝗶𝗰𝗮𝘁𝗶𝗼𝗻𝘀.

You can launch it as an application locally (it's Docker-ready!🐋) or, if you want more flexibility, you can integrate it in your code as a python package📦

The workflow is simple:
🧠 You choose your favorite LLM provider and model (supported, for now, are Mistral AI, Groq, Anthropic, OpenAI and Cohere)
🧠 You pick the embedding models provider and the embedding model you prefer (supported, for now, are Mistral AI, Hugging Face, Cohere and OpenAI)
📄 You prepare and provide your documents
⚙️ Documents are ingested into a Qdrant vector database and transformed into a synthetic question dataset with the help of LlamaIndex
📊 The LLM is evaluated for the faithfulness and relevancy of its retrieval-augmented answer to the questions
📊 The embedding model is evaluated for hit rate and mean reciprocal ranking (MRR) of the retrieved documents

And the cool thing is that all of this is 𝗶𝗻𝘁𝘂𝗶𝘁𝗶𝘃𝗲 𝗮𝗻𝗱 𝗰𝗼𝗺𝗽𝗹𝗲𝘁𝗲𝗹𝘆 𝗮𝘂𝘁𝗼𝗺𝗮𝘁𝗲𝗱: you plug it in, and it works!🔌⚡

Even cooler? This is all built on top of LlamaIndex and its integrations: no need for tons of dependencies or fancy workarounds🦙
And if you're a UI lover, Gradio and FastAPI are there to provide you a seamless backend-to-frontend experience🕶️

So now it's your turn: you can either get diRAGnosis from GitHub 👉 https://github.com/AstraBert/diRAGnosis
or just run a quick and painless:

uv pip install diragnosis


To get the package installed (lightning-fast) in your environment🏃‍♀️

Have fun and feel free to leave feedback and feature/integrations requests on GitHub issues✨
reacted to albertvillanova's post with 🔥 8 days ago
view post
Post
3768
🚀 Big news for AI agents! With the latest release of smolagents, you can now securely execute Python code in sandboxed Docker or E2B environments. 🦾🔒

Here's why this is a game-changer for agent-based systems: 🧵👇

1️⃣ Security First 🔐
Running AI agents in unrestricted Python environments is risky! With sandboxing, your agents are isolated, preventing unintended file access, network abuse, or system modifications.

2️⃣ Deterministic & Reproducible Runs 📦
By running agents in containerized environments, you ensure that every execution happens in a controlled and predictable setting—no more environment mismatches or dependency issues!

3️⃣ Resource Control & Limits 🚦
Docker and E2B allow you to enforce CPU, memory, and execution time limits, so rogue or inefficient agents don’t spiral out of control.

4️⃣ Safer Code Execution in Production 🏭
Deploy AI agents confidently, knowing that any generated code runs in an ephemeral, isolated environment, protecting your host machine and infrastructure.

5️⃣ Easy to Integrate 🛠️
With smolagents, you can simply configure your agent to use Docker or E2B as its execution backend—no need for complex security setups!

6️⃣ Perfect for Autonomous AI Agents 🤖
If your AI agents generate and execute code dynamically, this is a must-have to avoid security pitfalls while enabling advanced automation.

⚡ Get started now: https://github.com/huggingface/smolagents

What will you build with smolagents? Let us know! 🚀💡
reacted to clem's post with 🔥 8 days ago
view post
Post
5868
Super happy to welcome Nvidia as our latest enterprise hub customer. They have almost 2,000 team members using Hugging Face, and close to 20,000 followers of their org. Can't wait to see what they'll open-source for all of us in the coming months!

Nvidia's org: https://huggingface.co/nvidia
Enterprise hub: https://huggingface.co/enterprise
reacted to fdaudens's post with 🔥 14 days ago
view post
Post
3431
What if AI becomes as ubiquitous as the internet, but runs locally and transparently on our devices?

Fascinating TED talk by @thomwolf on open source AI and its future impact.

Imagine this for AI: instead of black box models running in distant data centers, we get transparent AI that runs locally on our phones and laptops, often without needing internet access. If the original team moves on? No problem - resilience is one of the beauties of open source. Anyone (companies, collectives, or individuals) can adapt and fix these models.

This is a compelling vision of AI's future that solves many of today's concerns around AI transparency and centralized control.

Watch the full talk here: https://www.ted.com/talks/thomas_wolf_what_if_ai_just_works
  • 1 reply
·
reacted to ehristoforu's post with 🔥 17 days ago
view post
Post
2742
Introducing our first standalone model – FluentlyLM Prinum

Introducing the first standalone model from Project Fluently LM! We worked on it for several months, used different approaches and eventually found the optimal one.

General characteristics:
- Model type: Causal language models (QwenForCausalLM, LM Transformer)
- Number of parameters: 32.5B
- Number of parameters (not embedded): 31.0B
- Number of layers: 64
- Context: 131,072 tokens
- Language(s) (NLP): English, French, Spanish, Russian, Chinese, Japanese, Persian (officially supported)
- License: MIT

Creation strategy:
The basis of the strategy is shown in Pic. 2.
We used Axolotl & Unsloth for SFT-finetuning with PEFT LoRA (rank=64, alpha=64) and Mergekit for SLERP and TIES mergers.

Evolution:
🏆 12th place in the Open LLM Leaderboard ( open-llm-leaderboard/open_llm_leaderboard) (21.02.2025)

Detailed results and comparisons are presented in Pic. 3.

Links:
- Model: fluently-lm/FluentlyLM-Prinum
- GGUF version: mradermacher/FluentlyLM-Prinum-GGUF
- Demo on ZeroGPU: ehristoforu/FluentlyLM-Prinum-demo
  • 7 replies
·
reacted to lysandre's post with ❤️ 20 days ago
view post
Post
5613
SmolVLM-2 and SigLIP-2 are now part of 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
·
reacted to mlabonne's post with 🤗 21 days ago
view post
Post
6031
🆕 LLM Course 2025 edition!

I updated the LLM Scientist roadmap and added a ton of new information and references. It covers training, datasets, evaluation, quantization, and new trends like test-time compute scaling.

The LLM Course has been incredibly popular (41.3k stars!) and I've been touched to receive many, many messages about how it helped people in their careers.

I know how difficult this stuff can be, so I'm super proud of the impact it had. I want to keep updating it in 2025, especially with the LLM Engineer roadmap.

Thanks everyone, hope you'll enjoy it!

💻 LLM Course: https://huggingface.co/blog/mlabonne/llm-course
reacted to m-ric's post with ❤️ 27 days ago
view post
Post
2906
𝗚𝗿𝗲𝗮𝘁 𝗳𝗲𝗮𝘁𝘂𝗿𝗲 𝗮𝗹𝗲𝗿𝘁: you can now share agents to the Hub! 🥳🥳

And any agent pushed to Hub get a cool Space interface to directly chat with it.

This was a real technical challenge: for instance, serializing tools to export them meant that you needed to get all the source code for a tool, verify that it was standalone (not relying on external variables), and gathering all the packages required to make it run.

Go try it out! 👉 https://github.com/huggingface/smolagents
  • 2 replies
·
reacted to Kseniase's post with 🔥 about 1 month ago
view post
Post
7785
8 New Types of RAG

RAG techniques continuously evolve to enhance LLM response accuracy by retrieving relevant external data during generation. To keep up with current AI trends, new RAG types incorporate deep step-by-step reasoning, tree search, citations, multimodality and other effective techniques.

Here's a list of 8 latest RAG advancements:

1. DeepRAG -> DeepRAG: Thinking to Retrieval Step by Step for Large Language Models (2502.01142)
Models retrieval-augmented reasoning as a Markov Decision Process, enabling strategic retrieval. It dynamically decides when to retrieve external knowledge and when rely on parametric reasoning.

2. RealRAG -> RealRAG: Retrieval-augmented Realistic Image Generation via Self-reflective Contrastive Learning (2502.00848)
Enhances  novel object generation by retrieving real-world images and using self-reflective contrastive learning to fill knowledge gap, improve realism and reduce distortions.

3. Chain-of-Retrieval Augmented Generation (CoRAG) -> Chain-of-Retrieval Augmented Generation (2501.14342)
Retrieves information step-by-step and adjusts it, also deciding how much compute power to use at test time. If needed it reformulates queries.

4. VideoRAG -> VideoRAG: Retrieval-Augmented Generation over Video Corpus (2501.05874)
Enables unlimited-length video processing, using dual-channel architecture that integrates graph-based textual grounding and multi-modal context encoding.

5. CFT-RAG ->  CFT-RAG: An Entity Tree Based Retrieval Augmented Generation Algorithm With Cuckoo Filter (2501.15098)
A tree-RAG acceleration method uses an improved Cuckoo Filter to optimize entity localization, enabling faster retrieval.

6. Contextualized Graph RAG (CG-RAG) -> CG-RAG: Research Question Answering by Citation Graph Retrieval-Augmented LLMs (2501.15067)
Uses Lexical-Semantic Graph Retrieval (LeSeGR) to integrate sparse and dense signals within graph structure and capture citation relationships

7. GFM-RAG -> GFM-RAG: Graph Foundation Model for Retrieval Augmented Generation (2502.01113)
A graph foundation model that uses a graph neural network to refine query-knowledge connections

8. URAG -> URAG: Implementing a Unified Hybrid RAG for Precise Answers in University Admission Chatbots -- A Case Study at HCMUT (2501.16276)
A hybrid system combining rule-based and RAG methods to improve lightweight LLMs for educational chatbots
  • 1 reply
·
reacted to retronic's post with 🔥 about 1 month ago
view post
Post
4427
Colox, a reasoning AI model. I am currently working on a model smarter than GPT o1 that thinks before it speaks. It is coming tomorrow in the afternoon.
·