Practical overview
When it fits
Use SmallDoge when you want a compact-model codebase that combines model examples with training stages and evaluation resources, or when you want to investigate a local WebUI for Doge models.
Who it is for
- Developers who want to prototype text generation with small language models
- Researchers and learners exploring language-model training and evaluation
- Teams investigating instruction or reasoning fine-tuning for small models
Documented uses
- Load a Doge model and generate text from a prompt
- Compare model sizes using the project’s reported speed and evaluation figures
- Follow a pre-training, instruction fine-tuning, or reasoning fine-tuning workflow
- Start the documented local chat interface and API
Requirements & limits
The repository is public and identifies its license as Apache-2.0.
- Available model sizes range from 20M to 320M parameters.
- The README reports speed figures for an i7-11 CPU and training times for an RTX 4090; these are project-reported examples.
- Instruction-tuned variants use the “-Instruct” suffix, and continued-training checkpoints use “-checkpoint”.
- The README’s three-hour claim for training a 20M model differs from its listed 14-hour RTX 4090 training time for Doge-20M; the conditions are not explained.
Check current requirements in the official documentation. Source-based guidance does not mean OpenViST has run or independently tested this project.
Install SmallDoge and generate text with Doge-60M-Instruct
Set up the repository and use its documented Python example to generate text from a prompt.
Before you start
- Python environment with pip
- Access to the SmallDoge repository and Hugging Face model resources
- Transformers and PyTorch imports used by the README example
Steps
Clone https://github.com/SmallDoges/small-doge.git and change into the small-doge directory.
Install the project in editable mode with pip install -e .
Load SmallDoge/Doge-60M-Instruct with AutoTokenizer and AutoModelForCausalLM, using trust_remote_code=True as shown in the README.
Tokenize a prompt, call model.generate, and decode the generated tokens.
Expected outcome
A Python workflow that produces decoded text from a prompt using Doge-60M-Instruct.
Guide limitations
- The README example does not specify hardware or supported execution environments.
- Model loading and generation are documented examples, not execution tests.
Install and start the SmallDoge WebUI
Install the documented WebUI extra and start its local interface.
Before you start
- Cloned SmallDoge repository
- Python environment with pip
Steps
From the repository directory, install the WebUI extra with pip install -e '.[webui]'.
Run the small-doge-webui command.
Use the documented local addresses: frontend at http://localhost:7860 and API at http://localhost:8000.
Expected outcome
The documented startup path for the local frontend and API.
Guide limitations
- The README does not detail authentication, deployment beyond localhost, or supported environments.
- The startup instructions are documentation, not a verified execution result.
Sources & review
Source record details
- Source availability
- Active
- Sources checked
- 14 October 2026
- Original website title
- SmallDoges/small-doge · Official homepage title · retained in its source language