GitHub Repo

GitHub 熱門專案、開源工具、框架、runtime 與重要 release。

Qwen3.8 Makes 2.4T-Parameter Max-Class Weights Available for the First Time, Activating 95B per Token

Alibaba’s Qwen team has released Qwen3.8-2.4T-A95B, featuring 512 experts and a hybrid of linear and standard attention, with support for contexts of up to approximately 1.01 million tokens. The model offers adjustable reasoning depth, but accepts text only and cannot disable thinking mode, making its practical deployment requirements far higher than those of typical open-weight models.

2026-08-13

代理工具

Bench2Robust Injects Nine Types of Tool Failures; Performance Regresses in 69 of 70 Agent Evaluations

An Amazon research team transformed existing tool-use benchmarks into fault environments in which agents can be instructed to retry, switch tools, or stop. After structured recovery information was added to Qwen3-4B, its pass rate on disrupted retail tasks increased by as much as 16.8 percentage points, but the system has yet to truly validate correct abstention on unsolvable tasks.

2026-08-13

生成式影音與 3D

StateFlow Replaces Shot-by-Shot Regeneration With Persistent 3D State, Earns 4.5/5 User Rating for Video Previews

StateFlow stores object geometry, poses, semantic states, and camera settings as an editable 3D world, leaving the video model to handle visual refinement. In a 30-person study, its video previews received an overall rating of 4.5/5, though the system relies on several closed models and its code has not yet been released.

2026-08-13

代理工具

Strong Models Compile Task Rules Into a Harness, Raising GPT‑5.4‑mini Accuracy From 0.488 to 0.912

In a strong-to-weak test-time transfer experiment, stronger models wrote routing, format-checking, and deterministic-solving programs for a fixed smaller model without updating its weights. The best harness corrected about 83% of baseline errors, but performance depended heavily on theory-of-mind benchmarks whose patterns could be compiled into rules.

2026-08-13

LFM2.5‑VL‑3B Runs a Vision Model in About 3GB of Memory, With Native Support for vLLM, llama.cpp, and MLX

Liquid AI has released LFM2.5‑VL‑3B, a 3.1B-parameter model that adds screen grounding, layout OCR, multi-image understanding, and function calling. The company reports 228 tokens/s on an Apple M5 Max and roughly 11,000 tokens/s in high-concurrency throughput on a single H100, though the figures have yet to be independently reproduced.

2026-08-13

代理工具

Long-term memory summaries retain only 3.05% of temporal expressions; a one-sentence prompt improves temporal QA accuracy by 31.4 points

A LoCoMo experiment found that general-purpose summaries preserve people and events but selectively discard dates, times, and durations. After an explicit instruction to preserve temporal anchors verbatim, judged accuracy on temporal questions rose from 15.6% to 47.0%.

2026-08-13

AI 數學與形式驗證

OEIS Open Uses Lean to Verify 492 Unsolved Conjectures, but Cross-Checking Revises the Pass Count From 147 to 144

OEIS Open tasks general-purpose language models with proving or disproving 492 integer-sequence conjectures under a fixed budget. The initial SafeVerify evaluation recorded 147 passes, but a second Lean verifier confirmed 144, highlighting how formal benchmarks remain sensitive to differences in specifications and checkers.

2026-08-13

代理工具

VAKRA Replays More Than 8,000 Local APIs, Revealing Sharp Accuracy Drops Under Multi-Hop and Policy Constraints

IBM’s VAKRA puts APIs, document retrieval, and natural-language policies into a single executable trajectory. Even the strongest model achieves only 50% to 70.4% full correctness across different API interfaces. When policies make questions unanswerable, some models’ success rates fall as low as 2.4%.

2026-08-13

Ling-3.0-tiny Activates Just 1.3B Parameters per Token, but Local Execution Still Relies on Dedicated Runtime Branches

InclusionAI has released the 7.9B-parameter Ling-3.0-tiny, which combines linear attention with sparse MoE and offers BF16, FP8, and INT4 weights. The company says FP8 reaches 86 to 90 tokens per second on an M4 Pro, but vLLM and Ollama support has yet to fully land in stable upstream releases.

2026-08-13

AI 安全

IO Factory Replays AI Influence Operations with 100,000 Simulated Personas, but Cannot Estimate Real-World Persuasion Effects

IO Factory connects narrative planning, agent posting, platform exposure, audience-state updates, and baseline comparisons into a traceable simulation lifecycle. The system has completed runs with 100,000 simulated civilians and 10,000 operator agents, but all influence measurements derive from manually configured rules and LLM judges.

2026-08-13

代理工具

ReTree Rolls Back the Search Tree When Evidence Conflicts, Raising Qwen3-8B’s Overall Answer Accuracy From 30.1% to 44.0%

ReTree stores a search agent’s summaries, evidence sources, and revision history as a dependency tree. When new information overturns an earlier premise, the system returns to the node that introduced the error and deletes the affected branches. Results across four search benchmarks show that it both shortens per-step context and improves answers, though hard pruning may discard information that is not actually affected by the conflict.

2026-08-12

代理工具

SKILLER Treats SKILL.md as an Optimizable Policy, Helping Qwen3.5-9B Reach 82.8% on Software Engineering Tests

Rather than updating the small model’s weights, SKILLER has a stronger model repeatedly revise text-based skills using execution traces and verifier feedback. Qwen3.5-9B’s three-run average on SWE-Skills-Bench rose from 26.2% without skills to 82.8%, although skill generation still relies on GPT-5.4 and successful reference trajectories.

2026-08-12