Gate News message, April 23 — DeepSeek has open-sourced TileKernels under the MIT license, a GPU kernel library written in TileLang for large language model training and inference. TileLang is a domain-specific language developed by the tile-ai team for expressing high-performance GPU kernels in Python. DeepSeek stated that most kernels in the library have approached hardware performance limits in compute density and memory bandwidth, with portions already deployed in internal training and inference operations.
The library comprises six categories of kernels: MoE (mixture of experts) gating and routing, including Top-k expert selection, token-to-expert mapping, and fused expand/shrink with weight normalization; quantization supporting FP8, FP4, and E5M6 formats with per-token, per-block, and per-channel quantization, including fused SwiGLU+quantization operations; batch transpose; Engram gating with fused RMSNorm forward/backward propagation and weight gradient reduction; Manifold HyperConnection with Sinkhorn normalization and mixed split/apply; and high-level autograd interfaces that wrap low-level kernels into trainable layers.
Engram and Manifold HyperConnection are proprietary components of DeepSeek’s model architecture, with implementation details disclosed publicly for the first time. The library requires NVIDIA SM90 or SM100 architecture GPUs (H100/H200 or Blackwell series), CUDA Toolkit 13.1 or higher, and PyTorch 2.10 or higher.
Disclaimer: The information on this page may come from third parties and does not represent the views or opinions of Gate. The content displayed on this page is for reference only and does not constitute any financial, investment, or legal advice. Gate does not guarantee the accuracy or completeness of the information and shall not be liable for any losses arising from the use of this information. Virtual asset investments carry high risks and are subject to significant price volatility. You may lose all of your invested principal. Please fully understand the relevant risks and make prudent decisions based on your own financial situation and risk tolerance. For details, please refer to
Disclaimer.
Related Articles
Succinct Labs Launches ZCAM iPhone App Using Cryptography to Combat AI-Generated Media
Gate News message, April 24 — Succinct Labs, backed by Paradigm, unveiled ZCAM on Thursday, an iPhone app that uses cryptography to fingerprint photos and videos in order to combat AI-generated and altered media. The app signs photos and videos at the moment of capture, producing a tamper-proof
GateNews1h ago
Pi Network launches the PiRC1 token framework, banning projects that have no real-world applications from issuing tokens
Pi Network unveiled the Pi Token Design Framework PiRC1 on April 22 as part of the Protocol V22 upgrade. PiRC1’s core rule is: only projects that already have deployable applications within the Pi ecosystem and that have real user demand are eligible to issue tokens. Token proceeds do not flow to the project team; instead, they flow into a perpetual liquidity pool anchored by Pi Coin to prevent rug-pull behavior.
MarketWhisper1h ago
Frax Joins DeFi United Support Action for Aave, to Propose Governance Plan for rsETH Incident
Gate News message, April 24 — Frax Finance announced on X that as an Aave V4 partner, it is in direct communication with Aave to address the rsETH incident. While Frax has no direct risk exposure, it
GateNews2h ago
CoW DAO proposes compensation for victims of the cow.fi domain hijacking, with up to 100% reimbursement of losses
CoW DAO on April 23 published a compensation proposal (CIP) in the governance forum, proposing the establishment of a discretionary grant program to provide eligible victims of the April 14 cow.fi domain hijacking incident with up to 100% loss reimbursement. The incident is estimated to have caused user losses of about US$1.2 million in USDC. CoW DAO emphasized that the compensation is of a voluntary, special-discretionary nature and does not represent an admission of any legal liability.
MarketWhisper2h ago
Telegram Founder: TON fees drop to $0.0005, moving toward zero fees
Telegram founder Pavel Durov posted on the X platform on April 23, announcing that TON network transaction fees will be reduced by 6x within a week, down to 0.00039 TON per transaction (about 0.0005 USD). The fee rate is fixed and is not affected by network load. Durov also announced that after the fee reduction, most transactions will further move toward a fully free zero-fee model.
MarketWhisper2h ago
CoW DAO Proposes Discretionary Grant Program to Compensate Domain Hijacking Victims
Gate News message, April 24 — CoW DAO has proposed establishing a discretionary grant program to compensate users who suffered losses from the April 14 cow.fi domain hijacking incident. The program will provide up to 100% loss reimbursement through a one-time allocation from the legal defense
GateNews3h ago
