---

base_model:

• lhpku20010120/Omni-Edu-4B

license: other

language:

• en

library_name: transformers

tags:

• text-generation-inference
• llama-cpp
• llama-factory
• full
• generated_from_trainer

pipeline_tag: image-text-to-text

datasets:

• lhpku20010120/Omni-Edu

---

Omni-Edu-4B-GGUF

> Omni-Edu-4B is a fine-tuned version of Qwen3.5-4B-Base from OpenDCAI, trained on the Omni-Edu-70K dataset as part of the "OmniEdu: Open Foundation Models for Learning and Teaching" project (arXiv:2609.23088), positioning it as an open foundation model targeting educational use cases. Training was conducted via LLaMA-Factory over 3 epochs on 8 GPUs with a learning rate of 5e-6 (cosine schedule, 10% warmup), a total effective batch size of 64, and the fused AdamW optimizer, using Transformers 5.2.0 and PyTorch 2.10.0. The model card itself is auto-generated and sparse — model description, intended uses/limitations, and detailed training/evaluation results are not yet documented in the repository, so specifics on capabilities and benchmark performance should be sought in the accompanying paper; it is released under a custom "other" license rather than a standard open license.

Model Files

| File Name | Quant Type | File Size | File Link | Description |

|-----------|------------|-----------|-----------|-------------|

| Omni-Edu-4B.BF16.gguf | BF16 | 8.42 GB | Link | Full BF16 weights. Highest quality, largest file size. |

| Omni-Edu-4B.Q3_K_L.gguf | Q3_K_L | 2.42 GB | Link | Lower quality but usable, good for low RAM availability. |

| Omni-Edu-4B.Q3_K_M.gguf | Q3_K_M | 2.26 GB | Link | Low quality. |

| Omni-Edu-4B.Q4_K_M.gguf | Q4_K_M | 2.71 GB | Link | Good quality, default size for most use cases, recommended. |

| Omni-Edu-4B.Q4_K_S.gguf | Q4_K_S | 2.56 GB | Link | Slightly lower quality with more space savings, recommended. |

| Omni-Edu-4B.Q5_K_M.gguf | Q5_K_M | 3.07 GB | Link | High quality, recommended. |

| Omni-Edu-4B.Q5_K_S.gguf | Q5_K_S | 2.99 GB | Link | High quality, recommended. |

| Omni-Edu-4B.Q6_K.gguf | Q6_K | 3.46 GB | Link | Very high quality, near perfect, recommended. |

| Omni-Edu-4B.mmproj-bf16.gguf | mmproj-bf16 | 676 MB | Link | Multimodal projection file in BF16 format. Used for vision/language models. |

llama.cpp

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