small-llm

raix = R + AI + eXperiment — AI coding assistant for RStudio. Chat, explain, debug, generate code with any model (OpenAI, Ollama, Claude, Groq, Mistral, DeepSeek + more).

UCL COMP0087 Statistical Natural Language Processing - NanoGlot Team

A curated showcase of small, on-premise models with great utility, enabling privacy-centered deployment and easy adaptation to bio-inspired use cases.

A 36M-parameter goldfish language model with a 10-second memory + pixel-art PWA desk pet. Runs in your browser, fully offline. Adopt it at den-sec.github.io/glublm/desk-pet/

Agent-enhanced LLM: 23 tools, ReAct/Plan-Execute, multi-turn RL, from 64M to 6B

Create a Small LLM using EleutherAI/gpt-neo-2.7B - Fine Tune It for a Specalized Purpouse and Leverage as a Co-Pilot

Train & Benchmarks LLM Sudoku solving on Sakana AI's Sudoku-Bench with multi-step & single-shot evaluation, animated solve visualizations, model comparisons, reports, and LoRA/GRPO post-training experiments to measure how far small local models can close the gap to frontier models.

Small-model-friendly OpenClaw skills — simple, local-only tools optimized for reliable single-tool-call execution by 4B-parameter LLMs (e.g. Qwen3-4B-AWQ)

quickthink is a local-first CLI and Python library that wraps Ollama-backed LLM calls with a compressed plan-then-answer scaffold and latency-aware routing. It adds a short validated planning step for multi-step prompts and routes simple ones straight through. Local inference control for small models.

Experiments with In Context Learning using small functions