Text Generation
Transformers
English
custom
tokenizer
symbolic-ai
mathematics
llm
reasoning
ast
compiler
nlp
deep-learning
machine-learning
mathematical-reasoning
symbolic-reasoning
tokenization
parser
artificial-intelligence
Eval Results (legacy)
Instructions to use SurweeshSP/mathtok with libraries, inference providers, notebooks, and local apps. Follow these links to get started.
- Libraries
- Transformers
How to use SurweeshSP/mathtok with Transformers:
# Use a pipeline as a high-level helper from transformers import pipeline pipe = pipeline("text-generation", model="SurweeshSP/mathtok")# Load model directly from transformers import AutoModel model = AutoModel.from_pretrained("SurweeshSP/mathtok", dtype="auto") - Notebooks
- Google Colab
- Kaggle
- Local Apps
- vLLM
How to use SurweeshSP/mathtok with vLLM:
Install from pip and serve model
# Install vLLM from pip: pip install vllm # Start the vLLM server: vllm serve "SurweeshSP/mathtok" # Call the server using curl (OpenAI-compatible API): curl -X POST "http://localhost:8000/v1/completions" \ -H "Content-Type: application/json" \ --data '{ "model": "SurweeshSP/mathtok", "prompt": "Once upon a time,", "max_tokens": 512, "temperature": 0.5 }'Use Docker
docker model run hf.co/SurweeshSP/mathtok
- SGLang
How to use SurweeshSP/mathtok with SGLang:
Install from pip and serve model
# Install SGLang from pip: pip install sglang # Start the SGLang server: python3 -m sglang.launch_server \ --model-path "SurweeshSP/mathtok" \ --host 0.0.0.0 \ --port 30000 # Call the server using curl (OpenAI-compatible API): curl -X POST "http://localhost:30000/v1/completions" \ -H "Content-Type: application/json" \ --data '{ "model": "SurweeshSP/mathtok", "prompt": "Once upon a time,", "max_tokens": 512, "temperature": 0.5 }'Use Docker images
docker run --gpus all \ --shm-size 32g \ -p 30000:30000 \ -v ~/.cache/huggingface:/root/.cache/huggingface \ --env "HF_TOKEN=<secret>" \ --ipc=host \ lmsysorg/sglang:latest \ python3 -m sglang.launch_server \ --model-path "SurweeshSP/mathtok" \ --host 0.0.0.0 \ --port 30000 # Call the server using curl (OpenAI-compatible API): curl -X POST "http://localhost:30000/v1/completions" \ -H "Content-Type: application/json" \ --data '{ "model": "SurweeshSP/mathtok", "prompt": "Once upon a time,", "max_tokens": 512, "temperature": 0.5 }' - Docker Model Runner
How to use SurweeshSP/mathtok with Docker Model Runner:
docker model run hf.co/SurweeshSP/mathtok
SurweeshSP commited on
Update README.md
Browse files
README.md
CHANGED
|
@@ -68,6 +68,28 @@ thumbnail: assets/mathtok_architecture_improvements.svg
|
|
| 68 |
# MathTok
|
| 69 |
|
| 70 |
**A Hybrid Canonicalized AST-Based Tokenization Framework for Mathematical Language Modeling**
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 71 |
|
| 72 |
## Overview
|
| 73 |
|
|
@@ -94,6 +116,13 @@ Compressed Token Stream
|
|
| 94 |
```
|
| 95 |
|
| 96 |
---
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 97 |
## Installation
|
| 98 |
|
| 99 |
Clone the repository and install the package in editable mode:
|
|
@@ -329,6 +358,18 @@ math_token/
|
|
| 329 |
|
| 330 |
---
|
| 331 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 332 |
## Citation
|
| 333 |
|
| 334 |
```bibtex
|
|
@@ -342,3 +383,10 @@ math_token/
|
|
| 342 |
url = {https://huggingface.co/Surweesh/MathTok}
|
| 343 |
}
|
| 344 |
```
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 68 |
# MathTok
|
| 69 |
|
| 70 |
**A Hybrid Canonicalized AST-Based Tokenization Framework for Mathematical Language Modeling**
|
| 71 |
+

|
| 72 |
+

|
| 73 |
+

|
| 74 |
+

|
| 75 |
+

|
| 76 |
+
|
| 77 |
+
---
|
| 78 |
+
|
| 79 |
+
## Why MathTok?
|
| 80 |
+
|
| 81 |
+
Traditional tokenizers such as BPE and SentencePiece treat mathematical
|
| 82 |
+
expressions as plain text sequences, fragmenting semantic structure and
|
| 83 |
+
discarding operator hierarchy.
|
| 84 |
+
|
| 85 |
+
MathTok introduces a structure-aware tokenization pipeline that:
|
| 86 |
+
- canonicalizes equivalent mathematical expressions,
|
| 87 |
+
- preserves AST hierarchy,
|
| 88 |
+
- encodes operator semantics explicitly,
|
| 89 |
+
- improves symbolic compression efficiency,
|
| 90 |
+
- and enables future tree-aware transformer architectures.
|
| 91 |
+
|
| 92 |
+
---
|
| 93 |
|
| 94 |
## Overview
|
| 95 |
|
|
|
|
| 116 |
```
|
| 117 |
|
| 118 |
---
|
| 119 |
+
|
| 120 |
+
## Architecture
|
| 121 |
+
|
| 122 |
+

|
| 123 |
+
|
| 124 |
+
---
|
| 125 |
+
|
| 126 |
## Installation
|
| 127 |
|
| 128 |
Clone the repository and install the package in editable mode:
|
|
|
|
| 358 |
|
| 359 |
---
|
| 360 |
|
| 361 |
+
## Future Work
|
| 362 |
+
|
| 363 |
+
- Tree-aware transformer attention integration
|
| 364 |
+
- Native mathematical pretraining corpus
|
| 365 |
+
- Symbolic reasoning benchmarks
|
| 366 |
+
- Neural theorem proving interfaces
|
| 367 |
+
- Equation graph embeddings
|
| 368 |
+
- Mathematical multimodal tokenization
|
| 369 |
+
- Integration with Lean/Coq theorem systems
|
| 370 |
+
|
| 371 |
+
---
|
| 372 |
+
|
| 373 |
## Citation
|
| 374 |
|
| 375 |
```bibtex
|
|
|
|
| 383 |
url = {https://huggingface.co/Surweesh/MathTok}
|
| 384 |
}
|
| 385 |
```
|
| 386 |
+
|
| 387 |
+
---
|
| 388 |
+
|
| 389 |
+
## Links
|
| 390 |
+
|
| 391 |
+
- GitHub: https://github.com/SurweeshSP/mathtok
|
| 392 |
+
- Hugging Face: https://huggingface.co/Surweesh/MathTok
|