Autotrain compatible Base model:finetune:qwen/qwen2... Base model:qwen/qwen2.5-3b-ins... Conversational Dataset:powerinfer/longcot-ref... Dataset:powerinfer/qwq-longcot... En Endpoints compatible Instruct Qwen2 Region:us Safetensors Sharded Tensorflow
Model Card on HF 🤗: https://huggingface.co/PowerInfer/SmallThinker-3B-Preview
SmallThinker 3B Preview Benchmarks
MMLU Pro: 22.42
GPQA: 1.45
MUSR: 3.59
BBH: 22.06
IFEval: 62 vs 88 (so35)-29.5%
MATH Lvl 5: 27.79
LLME Score: 0.40349
nn.n% — How the model compares to the reference models: Anthropic Sonnet 3.5 ("so35"), GPT-4o ("gpt4o") or GPT-4 ("gpt4").
What is the LLM Explorer Rank (Score)
SmallThinker 3B Preview Parameters and Internals
| LLM Name | SmallThinker 3B Preview |
| Repository 🤗 | https://huggingface.co/PowerInfer/SmallThinker-3B-Preview |
| Base Model(s) | |
| Model Size | 3b |
| Required VRAM | 6.8 GB |
| Updated | 2025-05-18 |
| Maintainer | PowerInfer |
| Model Type | qwen2 |
| Instruction-Based | Yes |
| Model Files | |
| Supported Languages | en |
| Model Architecture | Qwen2ForCausalLM |
| Context Length | 32768 |
| Model Max Length | 32768 |
| Transformers Version | 4.43.4 |
| Tokenizer Class | Qwen2Tokenizer |
| Padding Token | <|endoftext|> |
| Vocabulary Size | 151936 |
| Torch Data Type | bfloat16 |
| Errors | replace |
Best Alternatives to SmallThinker 3B Preview
Best Alternatives |
Context / RAM |
Downloads |
Likes |
|---|---|---|---|
| Saba2 3B | 128K / 6.2 GB | 5 | 0 |
| Tessa T1 3B | 117K / 6.2 GB | 20 | 3 |
| UIGEN T1.5 3B | 117K / 6.2 GB | 5 | 1 |
| Qwen2.5 3B Instruct | 32K / 6.2 GB | 2119964 | 257 |
| Chirp 01 | 32K / 6.2 GB | 72 | 14 |
| Calme 3.2 Baguette 3B | 32K / 11 GB | 2993 | 1 |
| Calme 3.1 Baguette 3B | 32K / 6.2 GB | 3013 | 1 |
| Calme 3.2 Instruct 3B | 32K / 11 GB | 3014 | 3 |
| Menda 3B 750 | 32K / 6.2 GB | 24 | 1 |
| Qwen2.5 3B Model Stock V3.1 | 32K / 6.8 GB | 29 | 3 |
Note: green Score (e.g. "73.2") means that the model is better than PowerInfer/SmallThinker-3B-Preview .
Rank the SmallThinker 3B Preview Capabilities
🆘 Have you tried this model? Rate its performance. This feedback would greatly assist ML community in identifying the most suitable model for their needs. Your contribution really does make a difference! 🌟
What open-source LLMs or SLMs are you in search of? 47410 in total.
Original data from HuggingFace, OpenCompass and various public git repos.
Release v20241227