🤗 Hugging Face | 🤖 ModelScope | 📑 Technical Report
SmallThinker is a family of on-device native Mixture-of-Experts (MoE) language models specially designed for local deployment, co-developed by the IPADS and School of AI at Shanghai Jiao Tong University and Zenergize AI. Designed from the ground up for resource-constrained environments, SmallThinker brings powerful, private, and low-latency AI directly to your personal devices, without relying on the cloud.
- SmallThinker-21B-A3B: The foundation model, a strong start for researchers and builders who want full control for fine-tuning and custom solutions.
- SmallThinker-4B-A0.6B: The post-trained model best for drop-in, general-purpose chat and agentic experiences. It is a reflex-grade model without long thinking.
| Architecture | Large MoE Model | Small MoE Model |
|---|---|---|
| Total Parameters | 21B | 4B |
| Activated Parameters | 3B | 0.6B |
| Number of Layers | 52 | 32 |
| Attention Hidden Dimension | 2560 | 1536 |
| MoE Hidden Dimension (per Expert) | 768 | 768 |
| Number of Attention Heads | 28 | 12 |
| Number of KV Heads | 4 | 2 |
| Number of Experts | 64 | 32 |
| Selected Experts per Token | 6 | 4 |
| Vocabulary Size | 151,936 | 151,936 |
| Context Length | 16K | 32K |
| Attention Mechanism | GQA | GQA |
| Activation Function | ReGLU | ReGLU |
transformers==4.53.3 is required, we are actively working to support the latest version.
The following contains a code snippet illustrating how to use the model generate content based on given inputs.
from transformers import AutoModelForCausalLM, AutoTokenizer
import torch
path = "PowerInfer/SmallThinker-21BA3B-Instruct"
device = "cuda"
tokenizer = AutoTokenizer.from_pretrained(path, trust_remote_code=True)
model = AutoModelForCausalLM.from_pretrained(path, torch_dtype=torch.bfloat16, device_map=device, trust_remote_code=True)
messages = [
{"role": "user", "content": "Give me a short introduction to large language model."},
]
model_inputs = tokenizer.apply_chat_template(messages, return_tensors="pt", add_generation_prompt=True).to(device)
model_outputs = model.generate(
model_inputs,
do_sample=True,
max_new_tokens=1024
)
output_token_ids = [
model_outputs[i][len(model_inputs[i]):] for i in range(len(model_inputs))
]
responses = tokenizer.batch_decode(output_token_ids, skip_special_tokens=True)[0]
print(responses)ModelScope adopts Python API similar to (though not entirely identical to) Transformers. For basic usage, simply modify the first line of the above code as follows:
from modelscope import AutoModelForCausalLM, AutoTokenizer- Due to the constraints of its model size and the limitations of its training data, its responses may contain factual inaccuracies, biases, or outdated information.
- Users bear full responsibility for independently evaluating and verifying the accuracy and appropriateness of all generated content.
- SmallThinker does not possess genuine comprehension or consciousness and cannot express personal opinions or value judgments.
Both the code and the model weights are released under the apache 2.0.