AscendGYM / eval /bare_harness /README.md
SLAI-AITP's picture
Fix evaluation guides and connect the Hugging Face leaderboard link
961d5d4 verified
|
Raw History Blame Contribute Delete
3.74 kB
# bare_harness
编译执行环境可选择[官方镜像](../../docs/SANDBOX_IMAGE.md)或[兼容自建服务](../../docs/SELF_HOSTED_ENVIRONMENT.md)。连接地址为服务基础 URL,`op_manager.py run` 的 `--sandbox-url` 优先于 `SANDBOX_URL`,均未设置时默认 `http://127.0.0.1:9000`。源码和冻结数据通过 HTTP 上传;客户端无需服务源码。请求与返回格式见[服务协议](../../docs/EXECUTION_BACKEND.md)。
## 1. 定位
`bare` harness 评测裸模型本身的算子优化能力。模型只能使用 Backend 提供的工具池(读取文件、回填代码、提交编译/精度/profiling、结束任务),不依赖 coding agent CLI,也不使用 Pi。
实现基于 k-lens-v2:
- `client_v2/bare_client/`:单会话客户端;
- `client_v2/config.yaml` + `config_loader.py`:客户端配置;
- `Backend/`:k-lens-v2 Backend,包含 `finish_task` 工具。
## 2. 前置条件
- Python 3.10+,依赖见根目录 `requirements.txt`。
- 可访问的远端沙盒。
- 一个 OpenAI 兼容的模型端点。
不需要 Node 或 claude/codex/opencode CLI。
## 3. 模型配置
模型登记在 `bridge/configs/op_catalog.json → models`,运行时用 `--model <registry-key>` 选择。
k-lens-v2 支持的 provider:
```text
vllm / openrouter / openai / glm-api / glm-client
```
catalog 中的 `local` 会在 bare 路径映射为 `vllm`。本地 vLLM 示例:
```json
"my-local-model": {
"provider": "local",
"name": "DeepSeek-V4-Flash-BF16",
"openai_base": "http://127.0.0.1:8008/v1",
"wire_api": "chat",
"context": 1000000,
"api_key": "any-value",
"tool_format": "json"
}
```
## 4. 单算子运行
```bash
cd $ASCEND_GYM_WORKSPACE/bridge
python3 op_manager.py run \
--op swi_glu \
--harness bare \
--agent bare \
--model <MODEL_KEY> \
--candidate-evaluations-max 8 \
--final-reprofile-count 3 \
--sandbox-url http://<sandbox-host>:<port> \
--tag BARE-$(date +%Y%m%d-%H%M%S)
```
## 5. 批量运行
```bash
python3 start_all_runs.py \
--harness bare --agent bare \
--ops all \
--model <MODEL_KEY> \
--candidate-evaluations-max 8 --final-reprofile-count 3 \
--parallel 2 \
--sandbox-url http://<sandbox-host>:<port> \
--tag BARE-BATCH-$(date +%Y%m%d-%H%M%S)
```
## 6. 预算与评测协议
- `--candidate-evaluations-max`:候选评测硬上限。
- `--final-reprofile-count`:最终复测次数,不占候选预算。
- `--candidate-solutions-target`:可选软目标,默认等于候选上限。
- profiling 协议固定为 8 总轮 = 3 预热 + 5 测量,统计方式为 mean;与 bridge 的通用 `--profiling-runs` 口径不同。
## 7. 产物
输出目录:
```text
$ASCEND_GYM_WORKSPACE/runs/<TAG>/<op>-<agent>-<TAG>/
```
主要文件:
| 文件 | 说明 |
|---|---|
| `result.json` | `klens-bare-client-result-v1`,包含 success、speedup、baseline/best、终止原因 |
| `evaluations.jsonl` | 每次候选评测 |
| `trajectory.jsonl` | 模型与工具交互轨迹 |
| `usage.json` | token 使用 |
| `_client/` | bare_client 原始产物 |
## 8. 注意事项
- bare 使用 k-lens-v2 Backend(含 `finish_task`),不要指向其他版本的 Backend。
- bare 的 profiling 协议固定为 8 总轮 / 3 预热 / mean;与 native 的聚合口径不同,不应直接比较绝对 `duration_ms`。
- 模型偶尔只返回文本而不发起工具调用时,可通过 catalog 模型条目的 `protocol_correction_max` 调整纠偏次数。
相关文档:
- [README:安装与环境](../../README_PROJECT.md#installation)
- [README:快速开始](../../README_PROJECT.md#quick-start)
- [docs/BRIDGE_USAGE.md](../../docs/BRIDGE_USAGE.md)
- [docs/BRIDGE_CONFIG.md](../../docs/BRIDGE_CONFIG.md)