15 69 347

alkinun

AtAndDev

AI & ML interests

LLMs, Alignment, Merging, Unsloth, DPO, SFT, ORPO, SPIN..

Recent Activity

updated a dataset about 14 hours ago

AtAndDev/ranky-dataset

published a dataset about 14 hours ago

AtAndDev/ranky-dataset

reacted to thomwolf's post with 🚀 about 16 hours ago

View all activity

Organizations

AtAndDev's activity

reacted to thomwolf's post with 🚀🔥 about 16 hours ago

Post

1661

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 clefourrier's post with 👍🚀 about 17 hours ago

Post

1438

Gemma3 family is out! Reading the tech report, and this section was really interesting to me from a methods/scientific fairness pov.

Instead of doing over-hyped comparisons, they clearly state that **results are reported in a setup which is advantageous to their models**.
(Which everybody does, but people usually don't say)

For a tech report, it makes a lot of sense to report model performance when used optimally!
On leaderboards on the other hand, comparison will be apples to apples, but in a potentially unoptimal way for a given model family (like some user interact sub-optimally with models)

Also contains a cool section (6) on training data memorization rate too! Important to see if your model will output the training data it has seen as such: always an issue for privacy/copyright/... but also very much for evaluation!

Because if your model knows its evals by heart, you're not testing for generalization.

reacted to eliebak's post with 🔥 1 day ago

Post

1190

Google just dropped an exciting technical report for the brand-new Gemma3 model! 🚀 Here are my personal notes highlighting the most intriguing architectural innovations, design choices, and insights from this release:

1) Architecture choices:
> No more softcaping, replace by QK-Norm
> Both Pre AND Post Norm
> Wider MLP than Qwen2.5, ~ same depth
> SWA with 5:1 and 1024 (very small and cool ablation on the paper!)
> No MLA to save KV cache, SWA do the job!

2) Long context
> Only increase the rope in the global layer (to 1M)
> Confirmation that it's harder to do long context for smol models, no 128k for the 1B
> Pretrained with 32k context? seems very high
> No yarn nor llama3 like rope extension

3) Distillation
> Only keep te first 256 logits for the teacher
> Ablation on the teacher gap (tl;dr you need some "patience" to see that using a small teacher is better)
> On policy distillation yeahh (by
@agarwl_
et al), not sure if the teacher gap behave the same here, curious if someone have more info?

4) Others
> Checkpoint with QAT, that's very cool
> RL using improve version of BOND, WARM/WARP good excuse to look at
@ramealexandre
papers
> Only use Zero3, no TP/PP if i understand correctly ?
> Training budget relatively similar than gemma2

1 reply

reacted to jasoncorkill's post with 🔥👀 1 day ago

Post

1942

Benchmarking Google's Veo2: How Does It Compare?

The results did not meet expectations. Veo2 struggled with style consistency and temporal coherence, falling behind competitors like Runway, Pika, Tencent, and even Alibaba. While the model shows promise, its alignment and quality are not yet there.

Google recently launched Veo2, its latest text-to-video model, through select partners like fal.ai. As part of our ongoing evaluation of state-of-the-art generative video models, we rigorously benchmarked Veo2 against industry leaders.

We generated a large set of Veo2 videos spending hundreds of dollars in the process and systematically evaluated them using our Python-based API for human and automated labeling.

Check out the ranking here: https://www.rapidata.ai/leaderboard/video-models

Rapidata/text-2-video-human-preferences-veo2

replied to burtenshaw's post 1 day ago

Bruh its been 8 hours since announcement. chill ya guys

reacted to burtenshaw's post with 🤗👍 1 day ago

Post

1188

Here’s a notebook to make Gemma reason with GRPO & TRL. I made this whilst prepping the next unit of the reasoning course:

In this notebooks I combine together google’s model with some community tooling

- First, I load the model from the Hugging Face hub with transformers’s latest release for Gemma 3
- I use PEFT and bitsandbytes to get it running on Colab
- Then, I took Will Browns processing and reward functions to make reasoning chains from GSM8k
- Finally, I used TRL’s GRPOTrainer to train the model

Next step is to bring Unsloth AI in, then ship it in the reasoning course. Links to notebook below.

https://colab.research.google.com/drive/1Vkl69ytCS3bvOtV9_stRETMthlQXR4wX?usp=sharing

3 replies

posted an update 1 day ago

Post

1024

Gemma 3 seems to be really good at human preference. Just waiting for ppl to see it.

reacted to hanzla's post with 👍 1 day ago

Post

984

Gemma 3 is a game changer for on device multimodal applications.

Try for yourself how a 4 billion parameter model can be so good.

hanzla/PlaygroundGemma3

1 reply

reacted to BrigitteTousi's post with 🤗 1 day ago

Post

3597

Regardless of X being down or not, so glad I can rely on HF Posts for AI news ❤️🤗

1 reply

reacted to mkurman's post with ❤️ 12 days ago

Post

3673

Introducing a new architecture, MedIT One – a single-token transformer with LSTM-like recurrence.

It is extremely fast in training and inference, but we lack funding for large-scale training. Enjoy 🍓

https://github.com/MedITSolutionsKurman/medit-one

reacted to Quazim0t0's post with 👍 12 days ago

Post

2234

Debugging Tags:
Imagine, Associated Thoughts, Dialectical Analysis, Backwards Induction, Metacognition, and Normal Thought Processes such as <think> or <begin_of_thought>

Edit: Uploaded new images w/ a Open WebUI function to organize the tags.
Open WebUI Function: https://openwebui.com/f/quaz93/imagine_phi

This Phi-4 model is part of a test project that I called Micro-Dose. My goal was to use a small dataset to activate reasoning and other cognitive processes without relying on a large dataset.

I found that this was possible with a tiny dataset of just 90 rows, specifically designed as math problems. In the initial iterations, the dataset only activated reasoning when a math-related question was asked. I then made a few changes to the dataset’s structure, including the order of information and the naming of tags. You can see the sample results in the pictures. Not really anything special, just thought I'd share.

Tweaked the dataset a bit:
Quazim0t0/Imagine-Phi-v0.2-GGUF
Quazim0t0/MicroDoseV0.2

First image shows the new tags, second shows the regular thought process and the third is the model in combination with web searches

2 replies

reacted to Jaward's post with 🔥🚀❤️👍🤗 12 days ago

Post

4936

made a few improvements on custom grpo trainer:
- added sequence similarity reward (seems to work)
- improved vllm support (5x inference speed)
- adjusted reward scores (this helped with format/accuracy)
- can now push to hf hub (already pushed mine lol: Jaward/smollm2_360m_grpo_gsm8k_reasoner)

Code: https://github.com/Jaykef/ai-algorithms/blob/main/smollm2_360M_135M_grpo_gsm8k.ipynb