Modelpublic

Offlin33er/SmolLM2-1.7B-telegram-chats-GGUF

sourceHugging Faceapache-2.0updated 2d agoView on Hugging Face

0likes134downloads

Model Card

SmolLM2-1.7B-telegram-chats — GGUF

GGUF quantizations of Offlin33er/SmolLM2-1.7B-telegram-chats — a full SFT of SmolLM2-1.7B on 3,000 synthetic Telegram Desktop export conversations, trained with TRL. The model writes chat histories in the Telegram Desktop export (result.json) format.

Live demo: Offlin33er/telegram-chat-generator runs the Q4KM file on a free CPU Space.

Files

File	Quant	Size
`smollm2-1.7b-telegram-chats-f16.gguf`	f16 (full precision)	3.4 GB
`smollm2-1.7b-telegram-chats-q4_k_m.gguf`	Q4KM	1.1 GB

Usage

Drop a file into LM Studio or any llama.cpp-based runner and prompt it with a Telegram-export style conversation. Typical use: generating synthetic chat data for fine-tuning or testing chat-rendering pipelines.

Training

SFT with TRL on Offlin33er/telegram-chat-export-synthetic — see the full-precision model card for the training procedure and framework versions.