Upload README.md
Browse files
README.md
ADDED
|
@@ -0,0 +1,46 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
---
|
| 2 |
+
license: other
|
| 3 |
+
language:
|
| 4 |
+
- en
|
| 5 |
+
pipeline_tag: text-generation
|
| 6 |
+
inference: false
|
| 7 |
+
tags:
|
| 8 |
+
- transformers
|
| 9 |
+
- gguf
|
| 10 |
+
- imatrix
|
| 11 |
+
- SmallThinker-3B-Preview
|
| 12 |
+
---
|
| 13 |
+
Quantizations of https://huggingface.co/PowerInfer/SmallThinker-3B-Preview
|
| 14 |
+
|
| 15 |
+
### Inference Clients/UIs
|
| 16 |
+
* [llama.cpp](https://github.com/ggerganov/llama.cpp)
|
| 17 |
+
* [KoboldCPP](https://github.com/LostRuins/koboldcpp)
|
| 18 |
+
* [ollama](https://github.com/ollama/ollama)
|
| 19 |
+
* [text-generation-webui](https://github.com/oobabooga/text-generation-webui)
|
| 20 |
+
* [jan](https://github.com/janhq/jan)
|
| 21 |
+
* [GPT4All](https://github.com/nomic-ai/gpt4all)
|
| 22 |
+
---
|
| 23 |
+
|
| 24 |
+
# From original readme
|
| 25 |
+
|
| 26 |
+
We introduce **SmallThinker-3B-preview**, a new model fine-tuned from the [Qwen2.5-3b-Instruct](https://huggingface.co/Qwen/Qwen2.5-3B-Instruct) model.
|
| 27 |
+
|
| 28 |
+
Now you can directly deploy SmallThinker On your phones with [PowerServe](https://github.com/powerserve-project/PowerServe).
|
| 29 |
+
|
| 30 |
+
## Benchmark Performance
|
| 31 |
+
|
| 32 |
+
| Model | AIME24 | AMC23 | GAOKAO2024_I | GAOKAO2024_II | MMLU_STEM | AMPS_Hard | math_comp |
|
| 33 |
+
|---------|--------|-------|--------------|---------------|-----------|-----------|-----------|
|
| 34 |
+
| Qwen2.5-3B-Instruct | 6.67 | 45 | 50 | 35.8 | 59.8 | - | - |
|
| 35 |
+
| SmallThinker | 16.667 | 57.5 | 64.2 | 57.1 | 68.2 | 70 | 46.8 |
|
| 36 |
+
| GPT-4o | 9.3 | - | - | - | 64.2 | 57 | 50 |
|
| 37 |
+
|
| 38 |
+
Limitation: Due to SmallThinker's current limitations in instruction following, for math_comp we adopt a more lenient evaluation method where only correct answers are required, without constraining responses to follow the specified AAAAA format.
|
| 39 |
+
|
| 40 |
+
Colab Link: [Colab](https://colab.research.google.com/drive/182q600at0sVw7uX0SXFp6bQI7pyjWXQ2?usp=sharing)
|
| 41 |
+
## Intended Use Cases
|
| 42 |
+
|
| 43 |
+
SmallThinker is designed for the following use cases:
|
| 44 |
+
|
| 45 |
+
1. **Edge Deployment:** Its small size makes it ideal for deployment on resource-constrained devices.
|
| 46 |
+
2. **Draft Model for QwQ-32B-Preview:** SmallThinker can serve as a fast and efficient draft model for the larger QwQ-32B-Preview model. From my test, in llama.cpp we can get 70% speedup (from 40 tokens/s to 70 tokens/s).
|