Agentic-Coding-Tessa

# Agentic Coding Dataset for Tessa

A comprehensive dataset for training coding agents with tool-use, reasoning, and software engineering capabilities.

## Dataset Composition

This dataset combines multiple high-quality sources:

- **hermes_reasoning** (20.0%): Tool-use and reasoning dataset - [interstellarninja/hermes_reasoning_tool_use](https://huggingface.co/datasets/interstellarninja/hermes_reasoning_tool_use)
- **search_arena** (15.0%): Search and retrieval tasks - [lmarena-ai/search-arena-24k](https://huggingface.co/datasets/lmarena-ai/search-arena-24k)
- **arena_human_pref** (15.0%): Human preference data for alignment - [lmarena-ai/arena-human-preference-140k](https://huggingface.co/datasets/lmarena-ai/arena-human-preference-140k)
- **rstar_coder** (25.0%): Advanced coding problems with reasoning - [microsoft/rStar-Coder](https://huggingface.co/datasets/microsoft/rStar-Coder)
- **swe_bench** (25.0%): Software engineering trajectories - [SWE-bench/SWE-smith-trajectories](https://huggingface.co/datasets/SWE-bench/SWE-smith-trajectories)

## Dataset Statistics

- **Total samples**: 44,100
- **Format**: Axolotl-compatible conversation format
- **Fields**: `conversations` (list of turns with `from` and `value` keys)

## Usage with Axolotl

```yaml
datasets:
- path: smirki/Agentic-Coding-Tessa
type: chat_template
field_messages: conversations
message_property_mappings:
role: from
content: value
split: train
```

## Training Configuration for UIGEN-X

Recommended configuration for UIGEN-X-4B with this dataset:

```yaml
# Model
base_model: Tesslate/UIGEN-X-4B-0729
chat_template: chatml # For Qwen3-based models

# LoRA Configuration
adapter: lora
lora_r: 256
lora_alpha: 512
lora_dropout: 0.05
lora_target_modules:
- q_proj
- k_proj
- v_proj
- o_proj
- gate_proj
- up_proj
- down_proj

# Training
sequence_len: 8192 # Extended for code
micro_batch_size: 4
gradient_accumulation_steps: 4
num_epochs: 2
learning_rate: 5e-4
```

## Example Structure

```json
{
"conversations": [
{
"from": "system",
"value": "You are an expert programming assistant..."
},
{
"from": "human",
"value": "Help me implement a binary search algorithm"
},
{
"from": "gpt",
"value": "I'll help you implement binary search..."
}
],
"source": "dataset_name"
}
```

## License

Apache 2.0 (inherited from constituent datasets)

## Citation

```bibtex
@dataset{agentic_coding_tessa_2024,
title={Agentic Coding Dataset for Tessa},
author={Smirki},
year={2024},
publisher={HuggingFace}
}
```

# 面向Tessa的智能体编程数据集（Agentic Coding Dataset for Tessa）

本数据集为综合性训练数据集，用于赋能具备工具调用、逻辑推理与软件工程能力的编程智能体（AI Agent）。

## 数据集构成
本数据集整合了多个高质量数据源：
- **hermes_reasoning**（占比20.0%）：工具调用与逻辑推理数据集，数据源链接：[interstellarninja/hermes_reasoning_tool_use](https://huggingface.co/datasets/interstellarninja/hermes_reasoning_tool_use)
- **search_arena**（占比15.0%）：搜索与检索任务数据集，数据源链接：[lmarena-ai/search-arena-24k](https://huggingface.co/datasets/lmarena-ai/search-arena-24k)
- **arena_human_pref**（占比15.0%）：用于对齐训练的人类偏好数据，数据源链接：[lmarena-ai/arena-human-preference-140k](https://huggingface.co/datasets/lmarena-ai/arena-human-preference-140k)
- **rstar_coder**（占比25.0%）：带逻辑推理的高阶编程问题数据集，数据源链接：[microsoft/rStar-Coder](https://huggingface.co/datasets/microsoft/rStar-Coder)
- **swe_bench**（占比25.0%）：软件工程轨迹数据集，数据源链接：[SWE-bench/SWE-smith-trajectories](https://huggingface.co/datasets/SWE-bench/SWE-smith-trajectories)

## 数据集统计信息
- **总样本量**：44100
- **数据格式**：兼容Axolotl的对话格式
- **数据字段**：`conversations`（包含多轮对话的列表，带有`from`与`value`两个键）

## Axolotl适配使用方法
yaml
datasets:
- path: smirki/Agentic-Coding-Tessa
type: chat_template
field_messages: conversations
message_property_mappings:
role: from
content: value
split: train

## UIGEN-X训练配置
针对UIGEN-X-4B模型使用本数据集的推荐配置如下：
yaml
# Model
base_model: Tesslate/UIGEN-X-4B-0729
chat_template: chatml # For Qwen3-based models

# LoRA Configuration
adapter: lora
lora_r: 256
lora_alpha: 512
lora_dropout: 0.05
lora_target_modules:
- q_proj
- k_proj
- v_proj
- o_proj
- gate_proj
- up_proj
- down_proj

# Training
sequence_len: 8192 # Extended for code
micro_batch_size: 4
gradient_accumulation_steps: 4
num_epochs: 2
learning_rate: 5e-4

## 示例数据结构
json
{
"conversations": [
{
"from": "system",
"value": "You are an expert programming assistant..."
},
{
"from": "human",
"value": "Help me implement a binary search algorithm"
},
{
"from": "gpt",
"value": "I'll help you implement binary search..."
}
],
"source": "dataset_name"
}

## 授权协议
Apache 2.0（继承自各组成数据集的授权协议）

## 引用格式
bibtex
@dataset{agentic_coding_tessa_2024,
title={Agentic Coding Dataset for Tessa},
author={Smirki},
year={2024},
publisher={HuggingFace}
}