Tag
Explore by tags
A mixed instruction dataset for SFT and RLHF research that combines chat, math, code and instruction-following samples from multiple public datasets under an Apache-2.0-compatible license; intended for instruction tuning and evaluation.
Open-source HybridFlow implementation for RL post-training of LLMs. Decouples control flow from compute so PPO, GRPO, GSPO and DAPO share one dataflow; pairs FSDP/Megatron with vLLM/SGLang rollout and reports 1.5-20x throughput over prior RLHF stacks.
A GitHub repository of learning notes and code dedicated to ML + SYS (machine learning systems). It collects tutorials, code walkthroughs and engineering notes on RLHF, distributed training (FSDP, Megatron), inference and scheduling (SGLang, vllm), quantization, CUDA/GPU optimization, system design, and practical engineering.
Generates video from text or images via a DiT-based latent diffusion model: text-to-video, image-to-video, frame extension, and multi-keyframe conditioning in one model. A distilled 2B variant runs near real-time on one H100; 13B for higher quality.
Custom ComfyUI nodes that run Lightricks' LTX-Video diffusion-transformer models for text-to-video and image-to-video, adding IC-LoRA control over depth, pose, edges, and motion plus distilled and low-VRAM variants for node-based workflows.
Generates high-quality, editable 3D assets from text or images and decodes to radiance fields, 3D Gaussians, or textured meshes. Ships pretrained models up to 2B parameters, a 500K asset dataset and training code; best used with image conditioning and a ≥16GB NVIDIA GPU.
A fine-grained, domain-labeled web corpus for NLP and LLM research that decomposes a deduplicated web crawl into 67 domains via iterative URL/content labeling and coarse→fine recall; includes per-domain statistics, validation subsets, and domain-benchmark analysis tools.
A 671B-parameter Mixture-of-Experts language model (37B activated) trained on 14.8T tokens with 128K context, FP8-first training, a Multi-Token Prediction module, and Hugging Face weights—focused on efficient MoE training and long-context use cases.
Provides an open platform of omnimodal world models, datasets, and tools to build Physical AI — joint perception, generation, and action reasoning for robots, autonomous vehicles, and smart infrastructure. Supports images, video, audio, and action-conditioned workflows.
Wires retrievers, rerankers, and generators as standalone MCP servers orchestrated in YAML, so iterative RAG logic fits in dozens of lines instead of glue code. Adds loops, conditional branches, one-command web UIs, and shared evaluation benchmarks.
Drives your computer from natural language: a vision-language model reads raw screenshots and works the mouse and keyboard like a person, controlling any GUI app without APIs or accessibility hooks. Local or remote operator modes on Windows and macOS.