---
tags:
- trl
- grpo
- gguf
base_model:
- openbmb/MiniCPM5-1B
---

# YLM-1B
YLM-1B is ylm-ai's first generation of MiniCPM5-based 1 billion parameter reasoning models. This model excels at reasoning, and features an internal scratchpad to solve problems.
This repository hosts the raw, original fp16 weights of YLM-1B, alongside llama.cpp-compatible GGUF files.

## Training Details
YLM-1B was trained with reinforcement learning to solve problems. The base model is MiniCPM5.

## Downloads
**q4_0 GGUF**: 600MB (for the desperate), click [here](https://huggingface.co/ylm-ai/ylm-1b/resolve/main/ylm-1b-q4_0.gguf?download=true)

**q8_0 GGUF**: 1.2 GB (good quality), click [here](https://huggingface.co/ylm-ai/ylm-1b/resolve/main/ylm-1b-q8_0.gguf?download=true)

**f16 GGUF**: 2.2 GB (exceptional quality), click [here](https://huggingface.co/ylm-ai/ylm-1b/resolve/main/ylm-1b-f16.gguf?download=true)

**HF Transformers**: 2.2 GB, [safetensors](https://huggingface.co/ylm-ai/ylm-1b/resolve/main/model.safetensors?download=true) [tokenizer.json](https://huggingface.co/ylm-ai/ylm-1b/resolve/main/tokenizer.json?download=true) [tokenizer_config.json](https://huggingface.co/ylm-ai/ylm-1b/resolve/main/tokenizer_config.json?download=true) [generation_config.json](https://huggingface.co/ylm-ai/ylm-1b/resolve/main/generation_config.json?download=true) [config.json](https://huggingface.co/ylm-ai/ylm-1b/resolve/main/config.json?download=true) [chat_template.jinja](https://huggingface.co/ylm-ai/ylm-1b/resolve/main/chat_template.jinja?download=true)