Reasoning Cache: Continual Improvement Over Long Horizons via Short-Horizon RL
Paper
•
2602.03773
•
Published
•
5
AI & ML interests
None defined yet.
Recent Activity
View all activity
lewtun
submitted
a
paper
to Daily Papers
about 17 hours ago
Post
3324
👉 What happened in AI in 2025? 👈
We prepared the 2025 version of the HF AI Timeline Grid, highlighting open vs API-based model releases, and allowing you to browse and filter by access, modality, and release type!
Play with it here:
2025-ai-timeline/2025-ai-timeline
Here's my personal quarterly TL;DR:
1️⃣ Q1 — Learning to Reason
Deepseek not only releases a top-notch reasoning model, but shows how to train them and compete with closed frontier models. OpenAI debuts Deep Research.
Significant milestones: DeepSeek R1 & R1-Zero, Qwen 2.5 VL, OpenAI Deep Research, Gemini 2.5 Pro (experimental)
2️⃣ Q2 — Multimodality and Coding
More LLMs embrace multimodality by default, and there's a surge in coding agents. Strong vision, audio, and generative models emerge.
Significant milestones: Llama 4, Qwen 3, Imagen 4, OpenAI Codex, Google Jules, Claude 4
3️⃣ Q3 — "Gold" rush, OpenAI opens up, the community goes bananas
Flagship models get gold in Math olympiads and hard benchmarks. OpenAI releases strong open source models and Google releases the much anticipated nano-banana for image generation and editing. Agentic workflows become commonplace.
Significant milestones: Gemini and OpenAI IMO Gold, gpt-oss, Gemini 2.5 Flash Image, Grok 4, Claude Sonnet 4.5
4️⃣ Q4 — Mistral returns, leaderboard hill-climbing
Mistral is back with updated model families. All labs release impressive models to wrap up the year!
Significant milestones: Claude Opus 4.5, DeepSeek Math V2, FLUX 2, GPT 5.1, Kimi K2 Thinking, Nano Banana Pro, GLM 4.7, Gemini 3, Mistral 3, MiniMax M2.1 🤯
Credits
🙏 NHLOCAL for the source data https://github.com/NHLOCAL/AiTimeline
🫡 @reach-vb for the original idea, design and recipe
🙌 @ariG23498 and yours truly for compiling and verifying the 2025 edition
🥳 Here's to 2026, wishing it becomes the best year ever for open releases and on-device-first use-cases! 🥂
We prepared the 2025 version of the HF AI Timeline Grid, highlighting open vs API-based model releases, and allowing you to browse and filter by access, modality, and release type!
Play with it here:
2025-ai-timeline/2025-ai-timeline
Here's my personal quarterly TL;DR:
1️⃣ Q1 — Learning to Reason
Deepseek not only releases a top-notch reasoning model, but shows how to train them and compete with closed frontier models. OpenAI debuts Deep Research.
Significant milestones: DeepSeek R1 & R1-Zero, Qwen 2.5 VL, OpenAI Deep Research, Gemini 2.5 Pro (experimental)
2️⃣ Q2 — Multimodality and Coding
More LLMs embrace multimodality by default, and there's a surge in coding agents. Strong vision, audio, and generative models emerge.
Significant milestones: Llama 4, Qwen 3, Imagen 4, OpenAI Codex, Google Jules, Claude 4
3️⃣ Q3 — "Gold" rush, OpenAI opens up, the community goes bananas
Flagship models get gold in Math olympiads and hard benchmarks. OpenAI releases strong open source models and Google releases the much anticipated nano-banana for image generation and editing. Agentic workflows become commonplace.
Significant milestones: Gemini and OpenAI IMO Gold, gpt-oss, Gemini 2.5 Flash Image, Grok 4, Claude Sonnet 4.5
4️⃣ Q4 — Mistral returns, leaderboard hill-climbing
Mistral is back with updated model families. All labs release impressive models to wrap up the year!
Significant milestones: Claude Opus 4.5, DeepSeek Math V2, FLUX 2, GPT 5.1, Kimi K2 Thinking, Nano Banana Pro, GLM 4.7, Gemini 3, Mistral 3, MiniMax M2.1 🤯
Credits
🙏 NHLOCAL for the source data https://github.com/NHLOCAL/AiTimeline
🫡 @reach-vb for the original idea, design and recipe
🙌 @ariG23498 and yours truly for compiling and verifying the 2025 edition
🥳 Here's to 2026, wishing it becomes the best year ever for open releases and on-device-first use-cases! 🥂
- 2 replies
eliebak
submitted
a
paper
to Daily Papers
about 2 months ago
ariG23498
authored
a
paper
4 months ago
jfchi
authored
6
papers
4 months ago
FFB: A Fair Fairness Benchmark for In-Processing Group Fairness Methods
Paper
•
2306.09468
•
Published
•
1
The Llama 3 Herd of Models
Paper
•
2407.21783
•
Published
•
117
Persistent Pre-Training Poisoning of LLMs
Paper
•
2410.13722
•
Published
Towards Understanding the Fragility of Multilingual LLMs against Fine-Tuning Attacks
Paper
•
2410.18210
•
Published
Large Reasoning Models Learn Better Alignment from Flawed Thinking
Paper
•
2510.00938
•
Published
•
59
Shape it Up! Restoring LLM Safety during Finetuning
Paper
•
2505.17196
•
Published
•
1
Post
1623
New post is live!
This time we cover some major updates to
🤗
This time we cover some major updates to
transformers.🤗
-
- 2 replies
Post
4079
Super excited to announce that our research team at Hugging Face will be doing an AMA on reddit r/LocalLLaMA.
Come ask any questions to the team behind SmolLM, FineWeb and more! And who knows, maybe there’ll be a shiny new release to talk about?
Thursday 4th September, 8AM-11AM PST 🤗
science
Come ask any questions to the team behind SmolLM, FineWeb and more! And who knows, maybe there’ll be a shiny new release to talk about?
Thursday 4th September, 8AM-11AM PST 🤗
Post
735
Motif 2.6B tech report is pretty insane, first time i see a model with differential attention and polynorm trained at scale!
> It's trained on 2.5T of token, with a "data mixture schedule" to continuously adjust the mixture over training.
> They use WSD with a "Simple moving average" averaging the last 6 ckpt every 8B token.
> They trained on Finemath, Fineweb2, DCLM, TxT360.
> Lot of details in the finetuning data they used, for instance they used EvolKit and did some "dataset fusion" to have more compressed knowledge into the data.
> They mention they also tried Normalized GPT, QK-Norm and Cross Layer Attention.
Motif-Technologies/Motif-2.6B
> It's trained on 2.5T of token, with a "data mixture schedule" to continuously adjust the mixture over training.
> They use WSD with a "Simple moving average" averaging the last 6 ckpt every 8B token.
> They trained on Finemath, Fineweb2, DCLM, TxT360.
> Lot of details in the finetuning data they used, for instance they used EvolKit and did some "dataset fusion" to have more compressed knowledge into the data.
> They mention they also tried Normalized GPT, QK-Norm and Cross Layer Attention.
Motif-Technologies/Motif-2.6B
Post
17320
Okay this is insane... WebGPU-accelerated semantic video tracking, powered by DINOv3 and Transformers.js! 🤯
Demo (+ source code): webml-community/DINOv3-video-tracking
This will revolutionize AI-powered video editors... which can now run 100% locally in your browser, no server inference required (costs $0)! 😍
How does it work? 🤔
1️⃣ Generate and cache image features for each frame
2️⃣ Create a list of embeddings for selected patch(es)
3️⃣ Compute cosine similarity between each patch and the selected patch(es)
4️⃣ Highlight those whose score is above some threshold
... et voilà! 🥳
You can also make selections across frames to improve temporal consistency! This is super useful if the object changes its appearance slightly throughout the video.
Excited to see what the community builds with it!
Demo (+ source code): webml-community/DINOv3-video-tracking
This will revolutionize AI-powered video editors... which can now run 100% locally in your browser, no server inference required (costs $0)! 😍
How does it work? 🤔
1️⃣ Generate and cache image features for each frame
2️⃣ Create a list of embeddings for selected patch(es)
3️⃣ Compute cosine similarity between each patch and the selected patch(es)
4️⃣ Highlight those whose score is above some threshold
... et voilà! 🥳
You can also make selections across frames to improve temporal consistency! This is super useful if the object changes its appearance slightly throughout the video.
Excited to see what the community builds with it!
- 3 replies
Post
4750
The next generation of AI-powered websites is going to be WILD! 🤯
In-browser tool calling & MCP is finally here, allowing LLMs to interact with websites programmatically.
To show what's possible, I built a demo using Liquid AI's new LFM2 model, powered by 🤗 Transformers.js: LiquidAI/LFM2-WebGPU
As always, the demo is open source (which you can find under the "Files" tab), so I'm excited to see how the community builds upon this! 🚀
In-browser tool calling & MCP is finally here, allowing LLMs to interact with websites programmatically.
To show what's possible, I built a demo using Liquid AI's new LFM2 model, powered by 🤗 Transformers.js: LiquidAI/LFM2-WebGPU
As always, the demo is open source (which you can find under the "Files" tab), so I'm excited to see how the community builds upon this! 🚀
- 2 replies
Post
3571
Introducing Voxtral WebGPU: State-of-the-art audio transcription directly in your browser! 🤯
🗣️ Transcribe videos, meeting notes, songs and more
🔐 Runs on-device, meaning no data is sent to a server
🌎 Multilingual (8 languages)
🤗 Completely free (forever) & open source
That's right, we're running Mistral's new Voxtral-Mini-3B model 100% locally in-browser on WebGPU, powered by Transformers.js and ONNX Runtime Web! 🔥
Try it out yourself! 👇
webml-community/Voxtral-WebGPU
🗣️ Transcribe videos, meeting notes, songs and more
🔐 Runs on-device, meaning no data is sent to a server
🌎 Multilingual (8 languages)
🤗 Completely free (forever) & open source
That's right, we're running Mistral's new Voxtral-Mini-3B model 100% locally in-browser on WebGPU, powered by Transformers.js and ONNX Runtime Web! 🔥
Try it out yourself! 👇
webml-community/Voxtral-WebGPU
Post
4812
Kimi K2 tech report is full of gems as always. Here are my notes on it:
> MuonClip: Pretty crazy how after 70k the training stabilizes and the QK-clip is basically inactive. There is also no loss in perf with QK-clip which is not trivial at all (at small scale but with aggressive threshold). Also a cool explanation of why muon makes the logit explode in appendix E (tl;dr is that muon makes the singular value of the update matrix higher)
> Sparsity scaling laws to justify their ratio, they have a very solid training infra that allows the model to be trained at this sparsity level, they could have increased even more but as sparsity increases the training becomes less efficient.
> They diminish the number of attention heads to make it more efficient for long context since attention heads are a big bottleneck for long context. They also remove 2 of the 3 "first dense" layers in the dsv3 arch.
With the sparsity and attention heads (divided by 2) they achieve 83% increased flops compared to deepseek v3 arch at 128k.
> Data: Rephrasing is KEY. They do a lot more synthetic data generation and rephrase their corpus to have different styles, for longer documents they do it by chunk. I'm (half) surprised by the fact that ONLY 1 epoch (assuming same number of training tokens I think?) of data rephrased 10 times has better accuracy than 10 epochs of the same data rephrased once.
> They do rewriting for Math and Knowledge, for Math they apply the ShallowMath recipe and instruct the model to rephrase in a "learning note" style
> They talk about diversity and probably have some internal stuff/eval to test that, as always still a bit unclear for me how to properly measure that.
The infra is also very nice, quick summary:
> PP=16 (1F1B schedule, a bit custom), EP=16, zero1
> No FP8 computation but for storage of specific layers, selective recomputation for inexpensive block, activation offloading to CPU
> MuonClip: Pretty crazy how after 70k the training stabilizes and the QK-clip is basically inactive. There is also no loss in perf with QK-clip which is not trivial at all (at small scale but with aggressive threshold). Also a cool explanation of why muon makes the logit explode in appendix E (tl;dr is that muon makes the singular value of the update matrix higher)
> Sparsity scaling laws to justify their ratio, they have a very solid training infra that allows the model to be trained at this sparsity level, they could have increased even more but as sparsity increases the training becomes less efficient.
> They diminish the number of attention heads to make it more efficient for long context since attention heads are a big bottleneck for long context. They also remove 2 of the 3 "first dense" layers in the dsv3 arch.
With the sparsity and attention heads (divided by 2) they achieve 83% increased flops compared to deepseek v3 arch at 128k.
> Data: Rephrasing is KEY. They do a lot more synthetic data generation and rephrase their corpus to have different styles, for longer documents they do it by chunk. I'm (half) surprised by the fact that ONLY 1 epoch (assuming same number of training tokens I think?) of data rephrased 10 times has better accuracy than 10 epochs of the same data rephrased once.
> They do rewriting for Math and Knowledge, for Math they apply the ShallowMath recipe and instruct the model to rephrase in a "learning note" style
> They talk about diversity and probably have some internal stuff/eval to test that, as always still a bit unclear for me how to properly measure that.
The infra is also very nice, quick summary:
> PP=16 (1F1B schedule, a bit custom), EP=16, zero1
> No FP8 computation but for storage of specific layers, selective recomputation for inexpensive block, activation offloading to CPU
Post
895
I have always advocated for writing techinical stories without using LLMs.
The following one page editorial really drives the point home.
https://www.nature.com/articles/s44222-025-00323-4
The following one page editorial really drives the point home.
https://www.nature.com/articles/s44222-025-00323-4
eliebak
authored
a
paper
8 months ago
Post
7488
NEW: Real-time conversational AI models can now run 100% locally in your browser! 🤯
🔐 Privacy by design (no data leaves your device)
💰 Completely free... forever
📦 Zero installation required, just visit a website
⚡️ Blazingly-fast WebGPU-accelerated inference
Try it out: webml-community/conversational-webgpu
For those interested, here's how it works:
- Silero VAD for voice activity detection
- Whisper for speech recognition
- SmolLM2-1.7B for text generation
- Kokoro for text to speech
Powered by Transformers.js and ONNX Runtime Web! 🤗 I hope you like it!
🔐 Privacy by design (no data leaves your device)
💰 Completely free... forever
📦 Zero installation required, just visit a website
⚡️ Blazingly-fast WebGPU-accelerated inference
Try it out: webml-community/conversational-webgpu
For those interested, here's how it works:
- Silero VAD for voice activity detection
- Whisper for speech recognition
- SmolLM2-1.7B for text generation
- Kokoro for text to speech
Powered by Transformers.js and ONNX Runtime Web! 🤗 I hope you like it!
- 5 replies