kaggle-tpu-lab
A recipe for serving the 27B-parameter Qwen3.8 model on Kaggle free TPUs with a public OpenAI-compatible endpoint and a 262k-token context window, no paid GPU.
View arahim3/kaggle-tpu-lab on GitHub- GitHub stars
- 796
- Forks
- —
- Language
- Python
Project details
- Repository
- arahim3/kaggle-tpu-lab
- Licence
- MIT
- Primary language
- Python
- Catalog status
- Available
- Catalog check
- 2026-11-13
Repository description
Frontier-class open models on a free Kaggle TPU v5e-8: GLM-5.3-Flash 320B MoE (~165 tok/s, our own JAX engine) and Qwen3.8-27B bf16 (~130 tok/s), 262k context, prefix caching. Works with Claude Code, Codex, opencode and pi.
Explore by topic
More tools in Models & training
- slotstream
carloslfu · MIT
Runs a 125-billion-parameter language model on a Mac with far less RAM than it needs by streaming parts of the model from the SSD.
- project-maya
mw00 · MIT
Lets anyone run a very large open-weight AI model privately on their own GPUs instead of relying on a cloud provider.
- vllm
vllm-project · Apache-2.0
vLLM is an open-source engine that runs large language models fast and cheaply in production, used widely to self-host models like Llama, Qwen and DeepSeek.
- llama-modchip
synchopate · Apache-2.0
Lets you run a huge 284-billion-parameter AI model on an ordinary 64GB computer using only the CPU, no GPU required.