🆕 [GitHub Repos] orgs/deepseek-ai - DeepSpec


https://github.com/deepseek-ai/DeepSpec

🗓️ Weekly GitHub Activity


🦙 llama.cpp
└ Release: b9743 → b9828
└ 85 commits

- Added support for Step 3.5/3.7 flash MTP3 speculative decoding #24340,
Granite Speech Plus #24818,
LFM2.5-ColBERT-350M/Embedding-350M #24913,
Eagle3 Qwen3 draft models #24977,
Unlimited-OCR #24969
- Added SSE Replay Buffer to server and UI, allowing text generation to survive HTTP disconnects and resume seamlessly #23226
- Introduced real-time model loading progress tracking via SSE in both server and UI #24828 #24878
- Configured server to create checkpoints before every user message to improve session recovery #24176
- Redesigned the WebUI with a new logo, navigation cleanup, and significant mobile layout improvements #24897
- Enabled dual-GPU tensor parallelism on the SYCL backend via split-mode tensor #24152
- Overhauled Hexagon matrix multiplication kernels with tiled layouts, HVX/HMX microkernels, and graph caching #24954
- Upgraded OpenCL Flash Attention kernels for F16, F32, Q4_0, and Q8_0 #25069
- Moved server model downloading to a dedicated child process #24834
- Added CUDA fast path for strided 2D copies using cudaMemcpy2DAsync #25057
- Reduced synchronization overhead between CPU and CUDA async copies during split compute #20793
- Added 3D convolution support to Vulkan #24612
- Fixed CUDA integer overflows and transposed copy failures #24706 #25000
- Fixed incorrect vector dot computations on SVE-enabled ARM CPUs #24699

🔗 All changes | Latest release


🎨 stable-diffusion.cpp
└ Release: master-709-92a3b73 → master-721-8caa3f9
└ 12 commits

- Added support for Boogu image generation #1688
- Added support for Krea2 models #1705
- Introduced guidance_schedule support for generation control #1684
- Added logit-normal scheduler #1669
- Added --eager-load flag to pre-load parameters during model initialization #1687
- Added --prompt-file and --negative-prompt-file flags for file-based inputs #1693
- Fixed memory mapping by avoiding writable mmap for read-only weights #1698

🔗 All changes | Latest release


🤗 Fresh models trending on HuggingFace:

empero-ai/Qwythos-9B-Claude-Mythos-5-1M ♡488
krea/Krea-2-Turbo ♡310
krea/Krea-2-Raw ♡214
deepreinforce-ai/Ornith-1.0-9B ♡167
deepreinforce-ai/Ornith-1.0-35B ♡161
deepreinforce-ai/Ornith-1.0-397B ♡121
Chunjiang-Intelligence/DeepSeek-v4-Fable ♡112
hustvl/Moebius ♡51
AutoArk-AI/ARK-ASR-3B ♡37
paom/texture2albedo-v2 ♡32
SupraLabs/Supra-A2A-Nano-Exp ♡30
Gryphe/Gemma-4-26B-A4B-StyleTune-V2 ♡24
0xSero/GLM-5.2-504B ♡19
g-astruc/UniverSat ♡18
allenai/tmax-27b ♡18
ValiantLabs/Qwen3.6-27B-Esper4 ♡14
wikeeyang/Flux2-Klein-9B-True-V3 ♡14
vrgamedevgirl84/Krea2_Enhancer ♡12

🆕 [HF Models] deepseek-ai - eagle3_gemma4_12b_ttt7

https://huggingface.co/deepseek-ai/eagle3_gemma4_12b_ttt7


🆕 [HF Models] deepseek-ai - eagle3_qwen3_14b_ttt7

https://huggingface.co/deepseek-ai/eagle3_qwen3_14b_ttt7


🆕 [HF Models] deepseek-ai - eagle3_qwen3_8b_ttt7

https://huggingface.co/deepseek-ai/eagle3_qwen3_8b_ttt7


🆕 [HF Models] deepseek-ai - eagle3_qwen3_4b_ttt7

https://huggingface.co/deepseek-ai/eagle3_qwen3_4b_ttt7


🆕 [HF Models] deepseek-ai - dflash_gemma4_12b_block7

https://huggingface.co/deepseek-ai/dflash_gemma4_12b_block7


🆕 [HF Models] deepseek-ai - dflash_qwen3_14b_block7

https://huggingface.co/deepseek-ai/dflash_qwen3_14b_block7


🆕 [HF Models] deepseek-ai - dflash_qwen3_8b_block7

https://huggingface.co/deepseek-ai/dflash_qwen3_8b_block7


🆕 [HF Models] deepseek-ai - dflash_qwen3_4b_block7

https://huggingface.co/deepseek-ai/dflash_qwen3_4b_block7


🆕 [HF Models] deepseek-ai - dspark_gemma4_12b_block7

https://huggingface.co/deepseek-ai/dspark_gemma4_12b_block7


🆕 [HF Models] deepseek-ai - dspark_qwen3_14b_block7

https://huggingface.co/deepseek-ai/dspark_qwen3_14b_block7


🆕 [HF Models] deepseek-ai - dspark_qwen3_8b_block7

https://huggingface.co/deepseek-ai/dspark_qwen3_8b_block7


🆕 [HF Models] deepseek-ai - dspark_qwen3_4b_block7

https://huggingface.co/deepseek-ai/dspark_qwen3_4b_block7

📰 PyTorch - Introducing Cross-Repository CI Relay: Scalable CI for PyTorch’s Out-of-Tree Backends
TL;DR PyTorch now has a Cross-Repository CI Relay (CRCR) that automatically triggers and tracks CI in downstream repositories whenever a PR is opened or a commit is pushed against pytorch/pytorch....

https://pytorch.org/blog/introducing-cross-repository-ci-relay-scalable-ci-for-pytorchs-out-of-tree-backends/

📰 HuggingFace - DiScoFormer: One transformer for density and score, across distributions


https://huggingface.co/blog/allenai/discoformer

🔄 [GitHub Releases] vllm-project/vllm - v0.24.0


https://github.com/vllm-project/vllm/releases/tag/v0.24.0

🔄 [GitHub Releases] open-webui/open-webui - v0.10.1


https://github.com/open-webui/open-webui/releases/tag/v0.10.1

📰 OpenAI - Mapping Europe’s AI Workforce Opportunity

https://openai.com/index/mapping-ai-jobs-transition-eu


📰 OpenAI - HP Inc. launches Frontier strategic partnership with OpenAI

https://openai.com/index/hp-frontier-partnership

📰 NVIDIA - How to Govern Autonomous Agents in Enterprise AI Factories 
AI agents are quickly moving beyond chat. They inspect code, run tests, read documents, search knowledge bases, query internal systems, and operate for hours on…

https://developer.nvidia.com/blog/how-to-govern-autonomous-agents-in-enterprise-ai-factories/

🆕 [HF Models] microsoft - vermeer-XL-CA

https://huggingface.co/microsoft/vermeer-XL-CA


🔓 [HF Models] microsoft - Dayhoff-170M-GRS-SS-74000

https://huggingface.co/microsoft/Dayhoff-170M-GRS-SS-74000


🔓 [HF Models] microsoft - Dayhoff-170M-GRS-SS-86000

https://huggingface.co/microsoft/Dayhoff-170M-GRS-SS-86000

🔄 [GitHub Releases] invoke-ai/InvokeAI - InvokeAI v6.13.5 (release candidate 1)


https://github.com/invoke-ai/InvokeAI/releases/tag/v6.13.5.rc1

🆕 [HF Models] meituan-longcat - LongCat-2.0


https://huggingface.co/meituan-longcat/LongCat-2.0

🆕 [GitHub Repos] orgs/NVlabs - falcor2-data


https://github.com/NVlabs/falcor2-data

📰 HuggingFace - Featuring Every Eval Ever Results on Hugging Face Model Pages


https://huggingface.co/blog/eee-community-evals

🆕 [HF Models] microsoft - GELab-Zero-4B-preview-Sico-Evolution


https://huggingface.co/microsoft/GELab-Zero-4B-preview-Sico-Evolution

📰 HuggingFace - Why Specialization Is Inevitable


https://huggingface.co/blog/Dharma-AI/why-specialization-is-inevitable

🆕 [HF Models] google - tabfm-1.0.0-jax

https://huggingface.co/google/tabfm-1.0.0-jax


🆕 [HF Models] google - tabfm-1.0.0-pytorch

https://huggingface.co/google/tabfm-1.0.0-pytorch

📰 Google Model Cards - Gemini 3.1 Flash-Lite Image


https://deepmind.google/models/model-cards/gemini-3-1-flash-lite-image/

📰 Google AI Blog - Driving the Agent Quality Flywheel from Your Coding Agent
Building AI agents often leaves developers uncertain if prompt tweaks to fix single errors will accidentally cause widespread regressions in production. To bridge this gap, Google has introduced a new developer skill for coding agents that automates a five-stage evaluation flywheel: preparing data, running inference, grading with adaptive AutoRaters, analyzing failure clusters, and executing targeted optimizations. Running continuously against production traffic or on-demand via synthetic scenarios, this tool allows developers to describe testing goals in plain language while an independent evaluation service safely validates and counts actual performance improvements.

https://developers.googleblog.com/en/driving-the-agent-quality-flywheel-from-your-coding-agent/


📰 Google AI Blog - Build reliable multi-agent applications with ADK Go 2.0. Discover our new graph-based workflow engine, built-in human-in-the-loop, and dynamic orchestration
The Agent Development Kit (ADK) for Go 2.0 has been released, introducing a first-class, graph-based workflow engine to help developers compose complex, multi-agent applications. This update adds built-in primitives for human-in-the-loop (HITL) orchestration, dynamic execution using plain Go code, and automated resilience features like exponential backoff retries. By unifying the execution model, both single-agent applications and intricate graphs now run on the same runtime, simplifying telemetry and state persistence.

https://developers.googleblog.com/en/announcing-adk-go-20/

📰 HuggingFace - ScarfBench: Benchmarking AI Agents for Enterprise Java Framework Migration


https://huggingface.co/blog/ibm-research/scarfbench

📰 PyTorch - Miles: A PyTorch-Native Stack for Large-Scale LLM RL Post-Training
TL;DR Miles is RadixArk’s open source framework for large-scale LLM RL post-training. It composes SGLang for rollout, NVIDIA Megatron-LM for training, Ray orchestration, and PyTorch-native extensibility behind a small, pluggable...

https://pytorch.org/blog/miles-a-pytorch-native-stack-for-large-scale-llm-rl-post-training/