---
title: README
emoji: 🃏
colorFrom: red
colorTo: gray
sdk: static
pinned: false
---

# PruhaNLP

In iqmaxxx we trust

### Robotics

- [**RoboSim at Home**](https://github.com/PruhaNLP/robosim-at-home) — browser studio for the SO-100 arm: collect with a leader arm, train, RL, eval in MuJoCo.
- [**TurboVLA-base**](https://huggingface.co/PruhaNLP/TurboVLA-base) — vision-language-action model without an LLM. 0.2B params, ~40 ms per action chunk, ~1 GB VRAM.
- Demo policies on one pick-and-place task: [TurboVLA](https://huggingface.co/PruhaNLP/TurboVLA-pickplace-blue-krill-demo) · [SmolVLA](https://huggingface.co/PruhaNLP/SmolVLA-pickplace-blue-krill-demo) · [ACT](https://huggingface.co/PruhaNLP/ACT-pickplace-blue-krill-demo) · [dataset](https://huggingface.co/datasets/PruhaNLP/pickplace-blue-krill-demo)

### On-device

- [**needle-3-esp32-s3**](https://huggingface.co/PruhaNLP/needle-3-esp32-s3) — tool-calling model running offline on an ESP32-S3. Text in, schema-valid JSON out.

### 1C code search

- [**USER2-1C-code**](https://huggingface.co/PruhaNLP/USER2-1C-code) — first open embedding model for 1C:Enterprise / BSL code.
- [**1C-Ebench**](https://huggingface.co/datasets/PruhaNLP/1C-Ebench) — code retrieval benchmark built from real developer questions.
- [**1C-Code-Train**](https://huggingface.co/datasets/PruhaNLP/1C-Code-Train) — 784K query / code / hard-negative triplets.

### Russian NLP

- [**ModernMT-en-ru-EXP**](https://huggingface.co/PruhaNLP/ModernMT-en-ru-EXP) — 66M-param EN→RU translator, on par with NLLB-1.3B on FLORES-200. [Demo](https://huggingface.co/spaces/PruhaNLP/ModernMT-en-ru).
- [**RuSlangX**](https://huggingface.co/datasets/PruhaNLP/RuSlangX) — do LLMs understand Russian internet slang? Scores range from 24% to 97%.