Running
thinletter demo
🔎
Search scientific papers locally with a browser‑based query encoder
thinletter
community
AI & ML interests
None defined yet.
Recent Activity
Organization Card
thinletter is a small research group. We make existing models lighter to run and measure them against the originals. thinletter.io · info@thinletter.io
ThinCtx
Qwen3.5 models converted so that the GPU memory for the context stays small: it grows 20 to 30 times more slowly than the original's cache, with accuracy on par with the original model. The full context is kept in system memory, so long documents and long agent sessions fit on one consumer GPU.
- ThinCtx-Qwen3.5-2B
- ThinCtx-Qwen3.5-0.8B
- ThinCtx-Qwen3.5-4B: coming soon
- Runtime:
pip install thinctx· results: thinletter.io/thinctx
ThinEmbed
The query side of an embedding model, compressed into a small file that runs in the browser on WebGPU, against your unchanged document index. Open source, Apache-2.0.
- harrier-0.6b (English): WebGPU clients · GGUF clients
- Qwen3-Embedding-0.6B (Czech): WebGPU client · GGUF client
- bge-m3 (Czech, multilingual): GGUF clients
- Demo · code · results: thinletter.io/query-encoders
models 7
thinletter/ThinCtx-Qwen3.5-0.8B
Text Generation • 0.8B • Updated • 1
thinletter/ThinCtx-Qwen3.5-2B
Text Generation • 2B • Updated • 1
thinletter/qwen3-embedding-0.6b-vq-clients
Updated • 8
thinletter/harrier-0.6b-vq-clients
Feature Extraction • Updated • 107
thinletter/bge-m3-query-clients
Feature Extraction • 0.6B • Updated • 111
thinletter/qwen3-embedding-0.6b-query-clients
Feature Extraction • 0.6B • Updated • 200
thinletter/harrier-0.6b-query-clients
Feature Extraction • 0.6B • Updated • 110