# Qwen-2 Projects at AI Tinkerers

> Canonical HTML: https://aitinkerers.org/technologies/qwen-2
> Markdown URL: https://aitinkerers.org/technologies/qwen-2.md
> Technology record last updated: 2026-02-23T01:32:22Z
> Generated: 2026-09-22T22:48:06Z

Qwen-2.5 is Alibaba's latest LLM family, scaling from 0.5B to 72B parameters and delivering significant gains in coding, mathematics, and 128K-token long-context processing.

Qwen-2.5 is the latest large language model series from Alibaba, delivering significant performance upgrades across the Qwen family. The release features a full range of open-weight models, scaling from 0.5B to 72B parameters, and includes specialized variants: Qwen2.5-Coder and Qwen2.5-Math. This architecture excels in instruction following, structured data generation (JSON), and multilingual support for over 29 languages. A key technical improvement is the extended context window, now supporting up to 128,000 tokens for robust, long-text application performance.

- Official technology site: https://qwenlm.github.io/blog/qwen2.5/
- Public AI Tinkerers demos and talks: 4
- Result page: 1 of 1

## Recent Public Talks and Demos

### [The loss curve lied: catching hidden safety drift inside the fine-tuning, automated with an agent!](https://zurich.aitinkerers.org/talks/rsvp_Vnt244xXFE0)

We built a fine-tuning where a Claude-Code based agent that helps founders and researchers fine-tune models on their datasets, and get more insights than just the training and eval losses. In particular, we also show the model’s regress on safety and general capability, metrics that otherwise would be overlooked. For example, we give it a medical instruction dataset; the agent asks a few multiple-choice questions and proposes concepts like "recommends unsafe or contraindicated actions," "downplays red-flag symptoms that need urgent care," and "implies clinical authority it doesn't have," then writes the run config. Then, for each concept, it fits a single direction in the model's activations (difference-of-means, following Persona Vectors, Chen et al. 2025, arXiv:2507.21509) that “encodes” that concept, and reads that direction off every checkpoint during LoRA training. Live we can show the agent conversation proposing the concepts, as how they drift alongside per-checkpoint evals. This is part of an ongoing research agenda with a collaboration of researchers from ETH Zurich, ETH AI Center and Agentic Systems Lab. We formulated our research into an agentic harness and live demo in an anthropic hackathon last week, and want to share more about it! Also, we imagine this as an open-source effort.

- Event context: AI Tinkerers Zürich: July Build Night with Swisscom Ventures — 2026-07-01 — Zürich
- Public talk page: https://zurich.aitinkerers.org/talks/rsvp_Vnt244xXFE0

### [Compose and Dragons: Tiny Language Models in Action](https://paris.aitinkerers.org/talks/rsvp_lNiq-CojExE)

Let's debunk some beliefs about (very) small LLMs, those that make less than 4b of parameters. We often hear: They are useless and do not know how to do anything, they know nothing, they are bad at calling (so no MCP) This is partly wrong, and we can fix the rest and build generative AI systems with these very small models. Among other things, we will see how to create NPCs with a personality, a master dungeon that will manage your movements, fights... and allow you to talk to this or that NPC ...

- Event context: AI Tinkerers Paris: Docker Agentic Workflows (Devoxx Kickoff) — 2026-04-21 — Paris
- Public talk page: https://paris.aitinkerers.org/talks/rsvp_lNiq-CojExE

### [DetailVault: Extracting Structured Data from Engineering Drawings with Local VLMs](https://montreal.aitinkerers.org/talks/rsvp_Dd70nghcBU4)

DetailVault is a document intelligence platform that uses local Vision Language Models to automatically extract structured metadata from technical engineering drawings.

- Event context: AI Tinkerers Montreal - March Demo Night — 2026-03-24 — Montreal
- Public talk page: https://montreal.aitinkerers.org/talks/rsvp_Dd70nghcBU4

### [Comparing Multimodal AI on Analytical Reasoning](https://boston.aitinkerers.org/talks/rsvp_j5G8xNtOHV0)

Many different deep learning models, especially LLMs, can be focused and fine-tuned on specific data modals. One very interesting feature of today's LLMs is multimodal input; this has its uses in encoding image, text, structured, and other sorts of data into the same embedding space. This talk will briefly go over the technical use cases of multimodal models by running an in-code comparison across different AI models and their ability to analyze and reason with data.

- Event context: AI Tinkerers Boston February 2025 — 2025-02-24 — Boston
- Public talk page: https://boston.aitinkerers.org/talks/rsvp_j5G8xNtOHV0

## Related Technologies

- [FastAPI](https://aitinkerers.org/technologies/fastapi) ([Markdown](https://aitinkerers.org/technologies/fastapi.md)) — 181 public demos
- [Agent](https://aitinkerers.org/technologies/agent) ([Markdown](https://aitinkerers.org/technologies/agent.md)) — 12 public demos
- [AI agents](https://aitinkerers.org/technologies/ai-agents) ([Markdown](https://aitinkerers.org/technologies/ai-agents.md)) — 35 public demos
- [Alph Editor](https://aitinkerers.org/technologies/alph-editor) ([Markdown](https://aitinkerers.org/technologies/alph-editor.md)) — 1 public demo
- [Apertus-8B](https://aitinkerers.org/technologies/apertus-8b) ([Markdown](https://aitinkerers.org/technologies/apertus-8b.md)) — 1 public demo
- [base models](https://aitinkerers.org/technologies/base-models) ([Markdown](https://aitinkerers.org/technologies/base-models.md)) — 2 public demos
- [Claude-3](https://aitinkerers.org/technologies/claude-3) ([Markdown](https://aitinkerers.org/technologies/claude-3.md)) — 110 public demos
- [Claude Agent SDK](https://aitinkerers.org/technologies/claude-agent-sdk) ([Markdown](https://aitinkerers.org/technologies/claude-agent-sdk.md)) — 16 public demos
- [dice rolls](https://aitinkerers.org/technologies/dice-rolls) ([Markdown](https://aitinkerers.org/technologies/dice-rolls.md)) — 1 public demo
- [Docker](https://aitinkerers.org/technologies/docker) ([Markdown](https://aitinkerers.org/technologies/docker.md)) — 147 public demos
- [GPT-4o](https://aitinkerers.org/technologies/gpt-4o) ([Markdown](https://aitinkerers.org/technologies/gpt-4o.md)) — 57 public demos
- [granite-embedding 278m](https://aitinkerers.org/technologies/granite-embedding-278m) ([Markdown](https://aitinkerers.org/technologies/granite-embedding-278m.md)) — 1 public demo
- [Jan-nano-gguf](https://aitinkerers.org/technologies/jan-nano-gguf) ([Markdown](https://aitinkerers.org/technologies/jan-nano-gguf.md)) — 1 public demo
- [LLaVA](https://aitinkerers.org/technologies/llava) ([Markdown](https://aitinkerers.org/technologies/llava.md)) — 5 public demos
- [LoRA](https://aitinkerers.org/technologies/lora) ([Markdown](https://aitinkerers.org/technologies/lora.md)) — 17 public demos
- [MiniCPM-V](https://aitinkerers.org/technologies/minicpm-v) ([Markdown](https://aitinkerers.org/technologies/minicpm-v.md)) — 1 public demo
- [monsters](https://aitinkerers.org/technologies/monsters) ([Markdown](https://aitinkerers.org/technologies/monsters.md)) — 1 public demo
- [o3-mini](https://aitinkerers.org/technologies/o3-mini) ([Markdown](https://aitinkerers.org/technologies/o3-mini.md)) — 2 public demos