Text Generation
Transformers
Safetensors
English
math
arithmetic
reasoning
grpo
small-language-model
from-scratch
Instructions to use arpjw/Microreason-43M with libraries, inference providers, notebooks, and local apps. Follow these links to get started.
- Libraries
- Transformers
How to use arpjw/Microreason-43M with Transformers:
# Use a pipeline as a high-level helper from transformers import pipeline pipe = pipeline("text-generation", model="arpjw/Microreason-43M")# pip install -U transformers accelerate # Load model directly from transformers import AutoModel model = AutoModel.from_pretrained("arpjw/Microreason-43M", device_map="auto") - Notebooks
- Google Colab
- Kaggle
- Local Apps Settings
- vLLM
How to use arpjw/Microreason-43M with vLLM:
Install from pip and serve model
# Install vLLM from pip: pip install vllm # Start the vLLM server: vllm serve "arpjw/Microreason-43M" # Call the server using curl (OpenAI-compatible API): curl -X POST "http://localhost:8000/v1/completions" \ -H "Content-Type: application/json" \ --data '{ "model": "arpjw/Microreason-43M", "prompt": "Once upon a time,", "max_tokens": 512, "temperature": 0.5 }'Use Docker
docker model run hf.co/arpjw/Microreason-43M
- SGLang
How to use arpjw/Microreason-43M 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 "arpjw/Microreason-43M" \ --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": "arpjw/Microreason-43M", "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 "arpjw/Microreason-43M" \ --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": "arpjw/Microreason-43M", "prompt": "Once upon a time,", "max_tokens": 512, "temperature": 0.5 }' - Docker Model Runner
How to use arpjw/Microreason-43M with Docker Model Runner:
docker model run hf.co/arpjw/Microreason-43M
Link the MicroReason Math dataset
Browse files
README.md
CHANGED
|
@@ -11,6 +11,7 @@ tags:
|
|
| 11 |
- small-language-model
|
| 12 |
- from-scratch
|
| 13 |
datasets:
|
|
|
|
| 14 |
- HuggingFaceTB/cosmopedia
|
| 15 |
---
|
| 16 |
|
|
@@ -112,7 +113,9 @@ Splitting digits, and generating problems from random expression trees instead o
|
|
| 112 |
templates, raised held-out accuracy from 0.2% to 70%.
|
| 113 |
|
| 114 |
**Synthetic math:** generated by a seeded expression-tree grammar. Every training problem is
|
| 115 |
-
checked against the frozen evaluation suites to exclude overlap.
|
|
|
|
|
|
|
| 116 |
|
| 117 |
## Findings
|
| 118 |
|
|
|
|
| 11 |
- small-language-model
|
| 12 |
- from-scratch
|
| 13 |
datasets:
|
| 14 |
+
- arpjw/microreason-math
|
| 15 |
- HuggingFaceTB/cosmopedia
|
| 16 |
---
|
| 17 |
|
|
|
|
| 113 |
templates, raised held-out accuracy from 0.2% to 70%.
|
| 114 |
|
| 115 |
**Synthetic math:** generated by a seeded expression-tree grammar. Every training problem is
|
| 116 |
+
checked against the frozen evaluation suites to exclude overlap. The evaluation suites and the
|
| 117 |
+
100k SFT set are published as
|
| 118 |
+
[arpjw/microreason-math](https://huggingface.co/datasets/arpjw/microreason-math).
|
| 119 |
|
| 120 |
## Findings
|
| 121 |
|