sammshen/swebench-minimax-traces-2

---
license: mit
task_categories:
- text-generation
tags:
- agent-traces
- tool-use
- multi-turn
- benchmark
- swebench
---

# swebench-minimax-traces-2

Complete HTTP-level agentic traces from running **swebench** benchmark tasks through an instrumented reverse proxy.

Each trace captures full request/response pairs including system prompts, user messages, assistant responses, tool calls and results, and token usage metadata.

## Stats
- **Total sessions**: 250
- **Multi-turn sessions** (2+ LLM calls): 210
- **Total records**: 15198
- **Total LLM requests**: 7623

## Format

Raw JSONL traces from the instrumented proxy. Each line is a request or response record:

### Request record
```json
{"type": "request", "request_id": "...", "timestamp_rel_s": 0.0, "timestamp_utc": "...", "method": "POST", "path": "/v1/chat/completions", "headers": {...}, "body": {...}, "thread_id": 0, "task_metadata": {"source_dataset": "swebench", "task_id": "...", "session_name": "..."}}
```

### Response record
```json
{"type": "response", "request_id": "...", "timestamp_rel_s": 1.23, "timestamp_utc": "...", "status_code": 200, "headers": {...}, "body": {...}, "thread_id": 0, "task_metadata": {"source_dataset": "swebench", "task_id": "...", "session_name": "..."}}
```

## Files
- `swebench_all_traces.jsonl` — All sessions merged into one file
- `sessions/` — Individual per-session trace files

## Collection Method
Traces collected via instrumented reverse proxy recording all LLM API calls during agent benchmark execution.

---
license: mit
task_categories:
- 文本生成
tags:
- 智能体轨迹（agent-traces）
- 工具使用（tool-use）
- 多轮对话（multi-turn）
- 基准测试（benchmark）
- SWEbench（swebench）
---

# SWEbench-Minimax轨迹集2

本数据集收录了通过插装式反向代理运行SWEbench基准任务时产生的完整HTTP级智能体交互轨迹。

每条轨迹均完整记录了请求/响应对，涵盖系统提示词、用户消息、助手回复、工具调用及结果，以及词元（Token）使用量元数据。

## 统计信息
- **总会话数**：250
- **多轮会话**（LLM调用次数≥2）：210
- **总记录数**：15198
- **LLM请求总次数**：7623

## 数据格式

本数据集采用插装式反向代理生成的原始JSONL轨迹格式，每行对应一条请求或响应记录：

### 请求记录
json
{"type": "request", "request_id": "...", "timestamp_rel_s": 0.0, "timestamp_utc": "...", "method": "POST", "path": "/v1/chat/completions", "headers": {...}, "body": {...}, "thread_id": 0, "task_metadata": {"source_dataset": "swebench", "task_id": "...", "session_name": "..."}}

### 响应记录
json
{"type": "response", "request_id": "...", "timestamp_rel_s": 1.23, "timestamp_utc": "...", "status_code": 200, "headers": {...}, "body": {...}, "thread_id": 0, "task_metadata": {"source_dataset": "swebench", "task_id": "...", "session_name": "..."}}

## 文件说明
- `swebench_all_traces.jsonl`：包含所有会话的合并轨迹文件
- `sessions/`：按会话拆分的独立轨迹文件目录

## 采集方式
本数据集通过插装式反向代理采集智能体基准测试运行期间的所有大语言模型（LLM）API调用轨迹。