---
base_model:
- microsoft/UserRepGen-4B
license: mit
language:
- en
pipeline_tag: text-generation
library_name: transformers
tags:
- text-generation-inference
- llama-cpp
- user-representation
- user-simulation
- simulation
- multi-turn
---

# **microsoft-UserRepGen-4B-GGUF**

> [UserRepGen-4B](https://huggingface.co/microsoft/UserRepGen-4B) is a 4-billion parameter language model developed by Microsoft, fine-tuned from Qwen3.5-4B via online supervised fine-tuning (OSFT) and hindsight distillation. Designed to advance research in human-AI interaction, it processes multi-turn user-assistant conversations to generate structured, natural-language representations comprising a "task" description and a "user_behavior" profile. Rather than generating simulated user turns itself, the model produces these representations to condition separate LLM-based user simulators, significantly improving conversation-reconstruction fidelity and preserving behavioral diversity in simulated populations. While it effectively captures nuanced communication styles, incremental constraint disclosures, and engagement patterns, users must carefully review its outputs for unsupported inferences and task-specific leakage, as the generated behavior descriptions are not anonymized and can increase the risk of cross-conversation linkability.

## Model Files

| File Name | Quant Type | File Size | File Link | Description |
|-----------|------------|-----------|-----------|-------------|
| UserRepGen-4B.BF16.gguf | BF16 | 8.42 GB | [Link](https://huggingface.co/prithivMLmods/microsoft-UserRepGen-4B-GGUF/blob/main/UserRepGen-4B.BF16.gguf) | Full BF16 weights. Highest quality, largest file size. |
| UserRepGen-4B.Q3_K_L.gguf | Q3_K_L | 2.42 GB | [Link](https://huggingface.co/prithivMLmods/microsoft-UserRepGen-4B-GGUF/blob/main/UserRepGen-4B.Q3_K_L.gguf) | Lower quality but usable, good for low RAM availability. |
| UserRepGen-4B.Q3_K_M.gguf | Q3_K_M | 2.26 GB | [Link](https://huggingface.co/prithivMLmods/microsoft-UserRepGen-4B-GGUF/blob/main/UserRepGen-4B.Q3_K_M.gguf) | Low quality. |
| UserRepGen-4B.Q4_K_M.gguf | Q4_K_M | 2.71 GB | [Link](https://huggingface.co/prithivMLmods/microsoft-UserRepGen-4B-GGUF/blob/main/UserRepGen-4B.Q4_K_M.gguf) | Good quality, default size for most use cases, *recommended*. |
| UserRepGen-4B.Q4_K_S.gguf | Q4_K_S | 2.56 GB | [Link](https://huggingface.co/prithivMLmods/microsoft-UserRepGen-4B-GGUF/blob/main/UserRepGen-4B.Q4_K_S.gguf) | Slightly lower quality with more space savings, *recommended*. |
| UserRepGen-4B.Q5_K_M.gguf | Q5_K_M | 3.07 GB | [Link](https://huggingface.co/prithivMLmods/microsoft-UserRepGen-4B-GGUF/blob/main/UserRepGen-4B.Q5_K_M.gguf) | High quality, *recommended*. |
| UserRepGen-4B.Q5_K_S.gguf | Q5_K_S | 2.99 GB | [Link](https://huggingface.co/prithivMLmods/microsoft-UserRepGen-4B-GGUF/blob/main/UserRepGen-4B.Q5_K_S.gguf) | High quality, *recommended*. |
| UserRepGen-4B.Q6_K.gguf | Q6_K | 3.46 GB | [Link](https://huggingface.co/prithivMLmods/microsoft-UserRepGen-4B-GGUF/blob/main/UserRepGen-4B.Q6_K.gguf) | Very high quality, near perfect, *recommended*. |
| UserRepGen-4B.Q8_0.gguf | Q8_0 | 4.48 GB | [Link](https://huggingface.co/prithivMLmods/microsoft-UserRepGen-4B-GGUF/blob/main/UserRepGen-4B.Q8_0.gguf) | Extremely high quality, generally unneeded but max available quant. |

## llama.cpp

LLM inference in C/C++ — https://github.com/ggml-org/llama.cpp