AI Dataset2026
A curated collection of 49,772 teacher-generated chat traces from qwen3.8-max-preview for supervised fine-tuning and off-policy distillation. Preserves visible chain-of-thought blocks, emphasizes math/code/reasoning mixes, and includes provenance and licensing cautions tied to Alibaba Cloud Model Studio.
Performs full-parameter post-training of trillion-parameter MoE DeepSeek-V4 models on an Ascend NPU SuperPOD, using a hierarchical optimization of model parallelism, communication orchestration, and kernel execution to increase Model FLOPs Utilization. Also builds CPT/SFT pipelines with solver-verified synthetic data for Operations Research, reporting strong zero-shot Pass@1 results.
AI Model2026
A sparsely activated Mixture-of-Experts (MoE) causal language model with 16B total parameters and 2.8B active parameters per token, released with end-to-end checkpoints and training recipe; trained on AMD Instinct GPUs and licensed for research use.
AI Dataset2026
Provides a near-deduplicated, quality-filtered 15.9 TB training subset of GitHub source code grouped by repository, with inline UTF‑8 file contents and repo metadata for pre-training and analysis of code LLMs; cutoff Aug 7, 2025, ODC-By license.
AI Model2026
Delivers image and video understanding plus a built-in event‑gated streaming gate — a unified 4B multimodal foundation model that uses codec-aligned tokenization to cut visual tokens by >75% and yield up to 3.5× wall‑clock inference speedup for streaming and long‑horizon video tasks.
AI Dataset2026
A compact dataset of prompt templates and examples designed to teach LLMs to consistently report model identity fields (model ID, name, creator, family, architecture, parameter count, knowledge cutoff). Includes regex markers, usage guidance, example replacements, and a small personalization script for fine‑tuning or runtime substitution.
AI Dataset2026
Consolidated dataset of detection, visual grounding and pointing annotations with indexed WebDataset image shards and Megatron‑Energon training metadata. Covers diverse visual domains (COCO, RefCOCO, driving, GUI, documents) and uses a normalized spatial grid for cross‑domain vision–language grounding training.
Presents a 2.8T-parameter Mixture-of-Experts multimodal model with a 1-million-token context window and 104 billion activated parameters, targeting long-horizon agentic RL, coding, reasoning, and vision. Key innovations include Kimi Delta Attention, Attention Residuals, Stable LatentMoE (16 of 896 experts active per token), ~2.5× scaling efficiency over Kimi K2, and a public weight release.
AI Video Papers2026
Real-time streaming multimodal foundation model that uses a codec-native tokenizer (Mage-ViT) to encode motion- and residual-rich regions from video I/P frames, reducing visual token usage by over 75% and enabling up to ~3.5× wall-clock inference speedup after training on ~560M images and 100M video frames.
Embodied AI2026
Provides a portable, robot-free UMI capture pipeline and shows that policies post-trained only on this high-fidelity data deploy directly on real robots matching teleoperation baselines. Capture achieves ~3 mm end-effector accuracy, microsecond sync, ultra-wide FOV, and releases 2,000h HiFi-UMI-2K.
A pretrain-then-transfer method for streaming recommendation that decouples refreshable behavioral knowledge from task-specific geometry to enable continual model refresh without downstream interference; introduces Behavioral Multi-Token Prediction and Anchored Calibration Residual and shows 4–12% offline gains plus live Shopee A/B lifts.
AI Agent Papers2026
Analyzes how to build effective training environment distributions for multimodal agents and proposes Ability-aware Environment Selection (AES) and Hierarchical Difficulty Curriculum (HDC) to improve diversity and difficulty scheduling, yielding large relative gains in experiments.