Annoy: This should be a paper Title
This is the resource page of the our resources collection on Huggingface, we highlight your currect position with a blue block.
Dataset
| Dataset | Link |
|---|
| Annoy-PythonEdu-Rs |
🤗
|
Please also check the raw data after our processing if you are interested: [mm-tool/Annoy-PyEdu-Rs-Raw](https://huggingface.co/datasets/mm-tool/Annoy-PyEdu-Rs-Raw).
Models
| Base Model / Training | Annoy | Annoy++ |
|---|
| Stage 1 | Stage 2 | Stage 1 | Stage 2 |
|---|
| Qwen 2.5 7B Coder | 🤗 | 🤗 | 🤗 | 🤗 |
| LLaMA 3.1 8B | 🤗 | 🤗 | 🤗 | 🤗 |
| DeepSeek v2 Lite Coder | 🤗 | 🤗 | 🤗 | 🤗 |
Introduction
While having full executable code theoretically allows us to generate reliable execution trajectories as responses, two challenges arise: 1) Obtaining a deterministic reverse function for input prediction is impractical; 2) Automatically constructed trajectories are constrained by pre-designed templates and lack the expressiveness and generalizability of free-form natural language reasoning. Thus, we adopt a fully LLM-based approach for synthesizing all the desired responses using DeepSeek-V2.5, as it has top-tier performance but extremely low cost compared to other advanced LLMs.
*Due to our collaborators' compliance requirements, we only release the PythonEdu-Rs subset (this page) of full dataset.
License
The license for this dataset is CC-BY-4.0.
License
The license for this dataset is The Stack v2 License.
License
The license for this dataset is Apache 2.0.
