We are reproducing the full DeepSeek R1 data and training pipeline so everybody can use their recipe. Instead of doing it in secret we can do it together in the open!
๐งช Step 1: replicate the R1-Distill models by distilling a high-quality reasoning corpus from DeepSeek-R1.
๐ง Step 2: replicate the pure RL pipeline that DeepSeek used to create R1-Zero. This will involve curating new, large-scale datasets for math, reasoning, and code.
๐ฅ Step 3: show we can go from base model -> SFT -> RL via multi-stage training.
๐ChemQwen-vL is a vision-language model fine-tuned based on the Qwen2VL-2B Instruct model. It has been trained using the International Chemical Identifier (InChI) format for chemical compounds and is optimized for chemical compound identification. The model excels at generating the InChI and providing descriptions of chemical compounds based on their images. Its architecture operates within a multi-modal framework, combining image-text-text capabilities. It has been fine-tuned using datasets from: https://iupac.org/projects/
โค๏ธโ๐ฅStranger Zone's MidJourney Mix Model Adapter is trending on the Very Model Page, with over 45,000+ downloads. Additionally, the Super Realism Model Adapter has over 52,000+ downloads, remains the top two adapter on Stranger Zone! strangerzonehf/Flux-Midjourney-Mix2-LoRA, strangerzonehf/Flux-Super-Realism-LoRA
๐ฏFine-tuning SmolLM2 on a lightweight synthetic reasoning dataset for reasoning-specific tasks. Future updates will focus on lightweight, blazing-fast reasoning models. Until then, check out the blog for fine-tuning details.
I was initially pretty sceptical about Meta's Coconut paper [1] because the largest perf gains were reported on toy linguistic problems. However, these results on machine translation are pretty impressive!
๐ฏTriangulum is a collection of pretrained and instruction-tuned generative models, designed for multilingual applications. These models are trained using synthetic datasets based on long chains of thought, enabling them to perform complex reasoning tasks effectively.
* Iteratively sample CoTs from the model, using a mix of different search strategies. This gives you something like Stream of Search via prompting. * Verify correctness of each CoT using GPT-4o (needed because exact match doesn't work well in medicine where there are lots of aliases) * Use GPT-4o to reformat the concatenated CoTs into a single stream that includes smooth transitions like "hmm, wait" etc that one sees in o1 * Use the resulting data for SFT & RL * Use sparse rewards from GPT-4o to guide RL training. They find RL gives an average ~3 point boost across medical benchmarks and SFT on this data already gives a strong improvement.
Applying this strategy to other domains could be quite promising, provided the training data can be formulated with verifiable problems!
๐ฏThe space handles documenting content from the input image along with standardized plain text. It includes adjustment tools with over 30 font styles, file formatting support for PDF and DOCX, textual alignments, font size adjustments, and line spacing modifications.
๐PDFs are rendered using the ReportLab software library toolkit.
We outperform Llama 70B with Llama 3B on hard math by scaling test-time compute ๐ฅ
How? By combining step-wise reward models with tree search algorithms :)
We show that smol models can match or exceed the performance of their much larger siblings when given enough "time to think"
We're open sourcing the full recipe and sharing a detailed blog post.
In our blog post we cover:
๐ Compute-optimal scaling: How we implemented DeepMind's recipe to boost the mathematical capabilities of open models at test-time.
๐ Diverse Verifier Tree Search (DVTS): An unpublished extension we developed to the verifier-guided tree search technique. This simple yet effective method improves diversity and delivers better performance, particularly at large test-time compute budgets.
๐งญ Search and Learn: A lightweight toolkit for implementing search strategies with LLMs and built for speed with vLLM
๐งชThe datasets were prepared for a 3:2 aspect ratio by processing images of any dimension (width ร height) in alignment with the adapter's concept. This involved using techniques such as magic expand, magic fill, or outpainting to adjust the remaining parts of the image to achieve the 3:2 ratio & posts training. This approach enhanced the desired image quality to up to 2 MB for detailed prompts and reduced artifacts in images sized at 1280 ร 832.
๐This approach was used instead of cropping down the 2x or 3x zoomed positions in the actual image. It generative filling to adjust the image's aspect ratio proportionally within the dataset.
๐งI used Canva's Magic Expand, Firefly's Generative Fill, and Flux's Outpaint for aspect ratio adjustments.
Fine-Textured [Polygon] Character 3D Design Renders ๐
Adapters capable of providing better lighting control (Bn+, Bn-) and richer textures compared to previous sets require more contextual prompts for optimal performance.
The ideal settings are achieved at inference steps around 30โ35, with the best dimensions being 1280 x 832 [ 3:2 ]. However, it also performs well with the default settings of 1024 x 1024 [ 1:1 ].
