#Hugging Face

OmniDR-Bench: Video-centric deep report generation benchmark evaluation.

Researchers have introduced the OmniDR-Bench benchmark and evaluation framework for in-depth study of video-centric deep agents, which can synthesize external evidence to generate interactive reports correlated with video moments. This task has direct practical value, helping users locate relevant video segments, verify claims, and reproduce demonstration steps. OmniDR-Bench includes 50 tasks in Chinese and English, covering 19 real-world subcategories, with a total of 6,414 standards, and the best system score is 63.3.

Cola2: Adaptive Hierarchical Language Model improves hierarchical language generation through continuous semantic compression and balanced discrete generation.

Cola2, through dynamic span splitting, masked diffusion decoding, and stream matching priors jointly optimizes continuous semantic compression, reconstruction-generation balance, and encoder-decoder adaptation, verifying its core design on five downstream tasks. Using 2100 billion training tokens, it demonstrates competitive non-saturated scalability, with a 3460 billion source token checkpoint showing an average of 60.10 points on eight tasks and surpassing Llama-3-8B on HellaSwag and StoryCloze.

RouteWeaver: Unifying LLM routing model selection and execution

RouteWeaver unifies heterogeneous LLM routing models through collaborative training and progressive exploration, improving accuracy in various reasoning tasks; its average accuracy reaches 61.4%, a relative improvement of 11.03% compared to the strongest baseline, and surpasses all comparison baselines in five domains.

Different tasks, different time steps: temporal task-aware quantization for diffusion language models T²Quant

T²Quant framework predicts accuracy sensitivity through minimal unlabeled calibration prompts, allocating activation precision within a fixed bit-width budget, improving existing quantization methods by 6.7 percentage points across 3 diffusion language models and 6 benchmark tests without increasing average accuracy budget.

Microsoft and Hugging Face use ThinkingBox database validation for RL reward training agents

Microsoft and Hugging Face release ThinkingBox-Training, directly using ThinkingBox's deterministic verification of database final states as reinforcement learning rewards, training Qwen3.8-27B for 50 steps with GRPO

Quant-Arena: A large language model agent for benchmark quantitative trading research

Quant-Arena evaluates whether language models can construct profitable trading strategies based on historical A-share data and verifies their performance on closed future year data, as a benchmark quantitative research. The platform allows models to build feature and stock rating code, based solely on feedback from 2022-2023, freeze strategies, and evaluate in 2024. GPT-6 Astra scored the highest among nine cutting-edge models, but no model surpassed the 2024 equal-weight market performance. Quant-Arena is open to all models or agents.

LASAGNA evaluated the composition of retrieval augmented generation (RAG) pipelines

LASAGNA evaluated the performance of 450 RAG pipeline configurations on 8 benchmarks and 5K-200K document corpora, finding that individual technology advantages at different stages do not necessarily combine effectively, retrieval quality and answer faithfulness optimization directions are different, the best combination varies with corpus size, indicating that isolated test results are unreliable, new RAG technology needs to be evaluated in a complete environment.

AKV introduces caching project management for LLM intelligent entity key-value caching

AKV proposes a caching engineering solution, managing LLM agent key-value caches through drop and repos operations, enhancing efficiency and accuracy. Experiments show that tool response rolling drop caching strategy improves the average benchmark score by 1.7 points, reduces costs by 27.7%, maximizes output throughput by 1.2 times, and accelerates P95 request latency by 1.5 times.

Future Chunk Prediction (FCP) enhances the physical plausibility of autoregressive video generation.

Future Chunk Prediction (FCP) improves autoregressive video diffusion models by introducing future chunk supervision, generating videos that better adhere to physical laws and reducing training costs. This method, under the teacher forcing framework, predicts temporally separated video chunks without relying on intermediate visual context, directly utilizing training videos for supervision, requiring no additional pre-trained visual encoder, and not altering autoregressive inference.

StreamEval: Benchmark for evaluating interactive streaming video generation models

StreamEval benchmarked the performance of interactive streaming video generation models on five tasks, finding reliable request execution and sustained state consistency to be major challenges. The benchmark includes 300 scripted test cases, covering character dialogue, character performance, world generation, instruction editing, and reference editing tasks, scored across four evaluation dimensions (task execution, state consistency, visual fidelity, and audiovisual coordination).

NVIDIA and partners launch RFD4-Proteina open-source model to accelerate drug development and cancer treatment

NVIDIA and the Protein Design Institute at the University of Washington Medical School jointly release the open-source RFD4-Proteina model, combining programmable protein design and search capabilities, surpassing baseline models in four benchmark tests, helping scientists explore drug design candidates.

Hugging Face releases IBM Granite time series model integration and prediction solution

Hugging Face announced IBM Granite, a time series model integration and forecasting solution, combining four Granite models (PatchTST-FM-r1, PatchTST-FM-r2, FlowState-r1.1, TTM-r3) to provide out-of-the-box integrated forecasting.

How to reduce inference cost by 77% in real-world applications using a specialized SLM vocabulary list.

Qwen3-4B's specialized vocabulary list VocabX reduces inference cost by 77% by focusing on the cost of each token rather than frequency, without changing the base model. This method for the Portuguese essay scorer reduces the output layer vocabulary from about 152,000 lines to about 65,000 lines, significantly reducing the computational overhead of token splitting and unused vocabulary items.

FairStress: A benchmark for a more in-depth evaluation of LLM decision fairness

FairStress benchmark evaluated LLM performance in demographic fairness across 300 decision scenarios across 29 fields (such as recruitment, healthcare), including 13 million test cases. The study found that the same persuasive prefixes (adversarial prefixes) have varying impacts on model decisions across different demographic groups, and there are differences in how models explain decisions containing demographic information.

Transformers releases release v5.20.2 patch update.

Transformers releases patch update v5.20.2, fixing an issue where input types were lost in the generate function, affecting multimodal models; it also addressed several other issues, including input dictionary type conversion, bfloat16 expert parallel gradients being zero, VLM input modal declaration for only reporting text, and host-to-device copy in eager_mask.

How to run n8n AI workflows locally on NVIDIA DGX Spark without cloud API

n8n workflows can run locally on NVIDIA DGX Spark without calling cloud APIs. By deploying the 35B hybrid expert model Qwen3.6-35B-A3B (NVFP4 format) and 8 other models on DGX Spark, tasks such as transcription, OCR, and image generation are supported. All models are accessed by n8n through OpenAI-compatible endpoints. Deployment requires caching models and using version 0.30.0 of vLLM to ensure offline operation and license compliance.

Tencent releases Hy-Memory-30B-A3B-GGUF model quantization file

The Hy-Memory-30B-A3B-GGUF model quantization file released by Tencent's Huayan uses a hybrid precision quantization scheme with imatrix calibration and MoE perception, achieving efficient local inference on llama.cpp, where the ~8 GB expert-mixed build file size is reduced to about 24.71 GB, with performance only 1.53% lower than Q6_K.

LangChain Deep Agents 0.1.86 Released

LangChain Deep Agents released version 0.1.86, adding NVIDIA OpenShell as a sandbox provider, fixed multiple bugs including improved alias matching, accurate reporting of workspace rejection, replacing duplicate thread recovery errors, replacing chat input on recall prompts, displaying expiration prompts in the footer, providing actionable errors when Groq tools are limited, and comparing effective interpreter permissions, etc.

Mistral AI releases Timestral-123M time series base model

Mistral AI releases Timestral-123M time series base model, with 123M parameters, using a patch-based Transformer architecture, context length of 32,768 time steps, single prediction covers 1,024 time steps, outputs 21 quantiles. The model achieves optimal accuracy in zero-shot prediction on GIFT-Eval, TIME, and fev-bench evaluations, and is licensed under the Modified MIT License.

Mistral AI releases Timestral-28M time series base model