The week in brief
Behind it, Whisper (+77) and Unsloth (+59) followed. This first comparison spans 1.5 days; from next week each report compares a full seven days.
On Hacker News, “Kolibri: A Sovereign Open-Weight Model” drew the most attention in this space (702 points, 336 comments). Among listed tools, Mistral AI's site was linked from 14 Hacker News stories over the past 90 days, more than any other tool here.
In research, “Rethinking Cross-Tokenizer On-Policy Distillation: From Alignment Coverage to Supervision Reliability” collected the most upvotes on Hugging Face Daily Papers this week (147). Among new GitHub repositories created in the past 28 days, youngyangyang04/llm-master leads with 1,106 stars.
Biggest GitHub star gainers
GitHub star counts for the 44 open-source tools we track, compared between Oct 5, 2026 and Oct 7, 2026 (37 hours apart).
| Tool | Stars before | Stars after | Gain | Change |
|---|---|---|---|---|
| llama.cpp | 130,405 | 130,572 | +167 | 0.13% |
| Whisper | 110,008 | 110,085 | +77 | 0.07% |
| Unsloth | 77,241 | 77,300 | +59 | 0.08% |
| PyTorch | 103,779 | 103,831 | +52 | 0.05% |
| Transformers | 166,982 | 167,019 | +37 | 0.02% |
| nanoGPT | 63,569 | 63,596 | +27 | 0.04% |
| MLX | 28,661 | 28,679 | +18 | 0.06% |
| TRL | 19,450 | 19,468 | +18 | 0.09% |
Most discussed on Hacker News
The highest-scoring Hacker News stories in this space during the week.
| Story | Points | Comments | Discussion |
|---|---|---|---|
| Kolibri: A Sovereign Open-Weight Model | 702 | 336 | thread |
| Clef: Open-weight decision models, and new RL fine-tuning platform | 641 | 217 | thread |
| Beam: Reflection's 501B open-weight model | 546 | 170 | thread |
| Aleph Alpha Kolibri: How the sovereign German LLM works | 422 | 12 | thread |
| From the creator of Redis; run LLM locally with ds4 | 363 | 105 | thread |
| Greg Kroah-Hartman – Security in the LLM Age [video] | 341 | 127 | thread |
Listed tools with the most Hacker News links
Hacker News stories linking to each tool's own website over the past 90 days.
Notable new papers
The most-upvoted papers on Hugging Face Daily Papers and the newest relevant arXiv preprints from this week.
- Rethinking Cross-Tokenizer On-Policy Distillation: From Alignment Coverage to Supervision Reliability
- DuoMatching: Joint-Marginal Distribution Matching for Few-Step Video Generation
- TRACE: Rollout-Guided Quantization-Aware Training for FP4 Reinforcement Learning of MoE Language Models
- EVISKILL: Grounding Skill Evolution in Replayable Evidence
- From Evidence to Action: How Tool-Using Agents Fail
- IdeaAnchor: Teaching LLMs to Turn Literature into Research Ideas
- Sherpa: Teaching LLMs to Teach Adaptively
- The Missing Minimal Pair: Stereotype Evaluation in LLMs
- Denoising Hierarchical Representations: Joint Continuous Diffusion for Language Modeling
- Agreement Is Not Validity: Cross-Model LLM Consensus in Diagnosing Student Failure Modes in K-12 Math Tutoring Dialogue
New GitHub repos in LLM research
Repositories created in the 28 days before Oct 07, 2026, ranked by stars.
| Repository | Stars | Language | Description |
|---|---|---|---|
| youngyangyang04/llm-master | 1,106 | 大模型（LLM）全栈学习路线与中文教程🔥：覆盖 Prompt Engineering、RAG、AI Agent、MCP、微调、模型部署、Transformer、AI 编程与大厂面试，从入门到生产实践。 | |
| amitshekhariitbhu/ai-engineering-course | 514 | Markdown | AI Engineering Course - A free and complete AI Engineering Course to learn AI Engineering step by step - from Machine Learning, Neural Networks, and Transformers to LLMs, Fine-Tuning, RAG, AI Agents, |
| liyupi/ai-model-world | 218 | TypeScript | AI 大模型世界，把 556 个大模型拟人化成像素小人的可视化站点。进来就能看到此刻谁最聪明、谁最会写代码、谁最便宜、谁刚发布，往下是国内与国外分区的厂商广场、完整的发布时间线和多维排行榜。搜索认模型名、厂商和能力，输入「多模态」会直接列出全部多模态模型。数据取自 Epoch AI、models.dev、LiveBench 与 Hugging Face，每小时自动同步，所有文案由真实数据生成，不调 |
| Sidiora-Labs/centra-gideon-agent | 152 | Python | The companion AI agent that learns, adapts and gets the work done no matter the task |
| Dreamer-Toby/STEPQuant | 83 | Python | STEPQuant: When and Where Errors Matter in Delta-Rule Recurrent State Quantization |
| NiuTrans/AML-Notes | 47 | TeX | 《高级机器学习》：面向生成式人工智能时代的机器学习教材，涵盖强化学习、大模型训练与对齐、生成建模和多模态学习等内容。 |
| NiuTrans/RL-without-Tears | 44 | HTML | An Introduction to reinforcement learning in the era of large language models |
How we compile this report
Every figure comes from a public source: star counts from the GitHub API, story scores and links from Hacker News (via Algolia), papers from arXiv and Hugging Face Daily Papers, and new repositories from GitHub search. Hacker News link counts include stories whose link points to the tool's own domain. Listing dates are the day a tool joined Axiomi. Numbers are shown as recorded; if a figure is unavailable, it is left out rather than estimated.