GLM-5.2 is live. #1 throughput on OpenRouter, pay-per-token on FriendliAI. Try it today ➜
tuan3110
Available on FriendliAI
Run this model inference on single tenant GPU with unmatched speed and reliability at scale.
Model Details
Model Provider
tuan3110
Model Tree
phxember/Uni-MuMER-Qwen3.5-2B
Input Modalities
Output Modalities
Supported Functionality
The adapter fine-tunes phxember/Uni-MuMER-Qwen3.5-2B on a real classroom handwritten calculus dataset.
Training used LoRA with Unsloth.
This is the main robust model used for the project demo.
It is designed for:
Dataset: University-HMER-RealClassroom
Real classroom split:
text
train: 1103 samplesvalidation: 259 samplesblind test: 274 samplestotal: 1636 images
Uploaded adapter directory:
text
outputs/unimumer_lora_unsloth_real/best_adapter/
The repository contains the LoRA adapter files and tokenizer/processor assets needed for inference with the base Uni-MuMER model.
| Model | ExpRate | TER | Latency |
|---|---|---|---|
| Uni-MuMER zero-shot | 14.67% | 13.50% | ~2.09 s/img |
| Uni-MuMER LoRA | 64.48% | 4.62% | ~2.60 s/img |
| Model | ExpRate | TER | Latency |
|---|---|---|---|
| Uni-MuMER zero-shot | 23.72% | 7.18% | 1.92 s/img |
| Uni-MuMER LoRA | 74.82% | 3.38% | 2.55 s/img |
Load the base model first:
text
phxember/Uni-MuMER-Qwen3.5-2B
Then load this repository as a LoRA adapter.
The adapter is trained on a relatively small real classroom dataset. It is more robust than the compact TAMER specialist model for live inputs, but it can still fail on unusual symbols, ambiguous handwriting, severe blur, or expressions outside the training distribution.
Thanks to: