← sources

microsoft/UserRepGen-4B

source: reddit (reddit:LocalLLaMA)published 2026-10-14 23:16ingested 2026-10-15 02:42

retained text

https://preview.redd.it/gwb4of5vmivh1.png?width=2400&format=png&auto=webp&s=853f4cf25b4140ccf8b151b4124547788a544b33

UserRepGen-4B is a fine-tuned Qwen3.5-4B language model that generates a structured, natural-language user representation from a multi-turn user-assistant conversation. Each representation contains a **task description** and a **behavior description** for a user, and is intended to condition a separate LLM-based user simulator.

Unlike [UserLM-8b](https://huggingface.co/microsoft/UserLM-8b) and [UserLM-2.0](https://huggingface.co/microsoft/UserLM-2.0), which generate user utterances, **UserRepGen-4B generates representations, not simulated user turns**. It does not itself execute the user's task or serve as the simulator.

The model was trained with online supervised fine-tuning through hindsight distillation, using feedback obtained by comparing simulated and original user behavior. At inference time, it generates a representation directly from the source conversation without requiring simulation rollouts or hindsight feedback.

[https://huggingface.co/microsoft/UserRepGen-4B](https://huggingface.co/microsoft/UserRepGen-4B)

document fetched 2026-10-15 00:45 · canonical https://reddit.com/r/LocalLLaMA/comments/1x664yt/microsoftuserrepgen4b