Tag
Explore by tags
Made reinforcement learning from human feedback (RLHF) the standard alignment recipe: collect demonstrations and preference rankings, train a reward model, then optimize with PPO. A 1.3B aligned model was preferred over the 175B GPT-3 by human raters.
Benchmark dataset of ~8.5k grade-school math word problems with step-by-step solutions and calculator annotations for evaluating multi-step arithmetic reasoning in language models. Provided in two configs (main and socratic) and commonly used for chain-of-thought prompting, fine-tuning, and verifier training.
Provides human preference comparison pairs and red-team conversation transcripts collected by Anthropic for training preference/reward models and studying harmful model behaviors; intended for RLHF and safety research, not for supervised fine-tuning of dialogue agents.
Runs an agentic RAG loop over scientific papers: searches literature, gathers and re-ranks evidence chunks, then answers with in-text citations. Adds metadata-aware embeddings, retraction checks, and contradiction detection across full PDFs.
A multimodal model that accepts image and text inputs and returns text, scoring at human level on professional exams — including a bar exam in the top 10%. Its performance was forecast from models using 1/1000th the compute, showing predictable scaling.
Enables real-time (≥30 fps) 1080p novel-view synthesis by representing scenes as optimized anisotropic 3D Gaussians plus a visibility-aware splatting renderer; provides the paper's reference implementation, pretrained models and viewers — high-quality training requires CUDA GPU and significant VRAM.
Re-derives LLM scaling laws, tracing prior disagreements to how compute budget was modeled, then trains 7B and 67B models on 2T tokens. The 67B model beats LLaMA-2 70B on code, math, and reasoning; its chat variant tops GPT-3.5 on open-ended evals.
Reworks the MoE layer to push each expert toward a narrow specialty: split experts into many finer ones and activate more per token, plus reserve a few always-on shared experts for common knowledge. A 2B model matches GShard 2.9B; at 16B it rivals LLaMA2 7B on ~40% of the compute.
A family of open code models (1.3B-33B) trained from scratch on 2T tokens of project-level code, using a 16K-window fill-in-the-blank objective. Beats Codex and GPT-3.5 on code benchmarks and ships under a license permitting commercial use.
Reaches 51.7% on the competition-level MATH benchmark with a 7B model and no tools or voting, rivaling Gemini-Ultra and GPT-4. Built on a 120B-token math corpus mined from Common Crawl, and introduces GRPO, a memory-efficient PPO variant for reasoning.
Provides ~1.3 trillion tokens of web pages filtered for educational quality using an LLM-trained classifier; includes per-Crawl configs, smaller random samples (10B/100B/350B tokens), and the classifier code and model for reproducible filtering.
A 236B-parameter Mixture-of-Experts LLM that activates only 21B parameters per token, cutting training cost 42.5% versus a dense 67B model and shrinking the KV cache 93.3% via Multi-head Latent Attention, with 128K context.