| update logs path
| 1 个月前 |
| refactor: rename project from aikg to akg_agents
Complete project renaming:
- Renamed main directory: aikg/ -> akg_agents/
- Renamed Python package: ai_kernel_generator -> akg_agents
- Renamed benchmark: aikgbench -> akg_kernels_bench
- Updated client API: AIKGClient -> AKGAgentsClient
- Updated all import paths and module references
- Updated environment variables: AIKG_* -> AKG_AGENTS_*
- Updated output directories and log paths
- Updated all documentation and configuration files
- Fixed Git submodules paths
Breaking changes:
- All Python imports need to update to 'from akg_agents'
- Environment variable names changed: AIKG_* -> AKG_AGENTS_*
- Package name changed from ai_kernel_generator to akg_agents
- Client class renamed: AIKGClient -> AKGAgentsClient
Co-authored-by: Cursor <cursoragent@cursor.com>
| 3 个月前 |
| LLM 输出去 JSON 化改造 + stream 模式支持
- KernelGen/Coder: 移除 JSON parser 依赖，改用代码提取 + py_compile 语法校验
- 删除 robust_parse 节点和相关 parser 配置
- 新增 stream 模式支持
- 新增 KernelGen 基准测试工具 (bench_kernel_gen.py)
- 更新依赖文件 requirements
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
| 2 个月前 |
| Refactor: remove legacy code, standardize test structure, and enhance core module test coverage
| 2 个月前 |
| feat: integrate autoresearch framework with preflight validation
Core integration (autoresearch as callable workflow):
- AutoresearchWorkflow: LangGraph single-node workflow wrapping AgentLoop
with KernelVerifier eval, supporting all DSLs and backends.
- Fail-closed preflight: acquire worker → validate reference (static +
runtime via check_task_desc_static/check_task_desc_runtime) → resolve
seed (user-supplied or KernelGen with CodeChecker + Verifier loop) →
release worker. AgentLoop only starts with verified ref + kernel.
- eval_fn: single call per round (verify + profile). Profiler's run_times
controls internal repetition. eval_repeats removed entirely.
- LangGraphTask: previous_code parameter threaded to initial state.
CLI (scripts/run_autoresearch.py):
- Input modes: --desc, --desc --kernel, --ref, --ref --kernel.
- AST-level reference validation. generate_task_desc retries on failure.
Eval robustness:
- run_one_round: contract boundary — always returns RoundRecord, never
raises. Inner exceptions caught, rollback + FAIL record returned.
- Core loop: turn execution crashes caught, files rolled back, continue.
- Profile templates: try/except/finally guarantees JSON result file is
always written. inf→None sanitization for json.dump compatibility.
Tests, docs, config, bugfixes from prior session carried forward.
| 1 个月前 |
| feat: integrate autoresearch framework with preflight validation
Core integration (autoresearch as callable workflow):
- AutoresearchWorkflow: LangGraph single-node workflow wrapping AgentLoop
with KernelVerifier eval, supporting all DSLs and backends.
- Fail-closed preflight: acquire worker → validate reference (static +
runtime via check_task_desc_static/check_task_desc_runtime) → resolve
seed (user-supplied or KernelGen with CodeChecker + Verifier loop) →
release worker. AgentLoop only starts with verified ref + kernel.
- eval_fn: single call per round (verify + profile). Profiler's run_times
controls internal repetition. eval_repeats removed entirely.
- LangGraphTask: previous_code parameter threaded to initial state.
CLI (scripts/run_autoresearch.py):
- Input modes: --desc, --desc --kernel, --ref, --ref --kernel.
- AST-level reference validation. generate_task_desc retries on failure.
Eval robustness:
- run_one_round: contract boundary — always returns RoundRecord, never
raises. Inner exceptions caught, rollback + FAIL record returned.
- Core loop: turn execution crashes caught, files rolled back, continue.
- Profile templates: try/except/finally guarantees JSON result file is
always written. inf→None sanitization for json.dump compatibility.
Tests, docs, config, bugfixes from prior session carried forward.
| 1 个月前 |
| test(cache): complete test specification rectification, directory migration and code normalization
- Remove obsolete cache test directory in core_v2 and clean up redundant files
- Migrate test files as per SPEC: UT to tests/ut/cache_tests, ST to tests/op/st
- Add Apache 2.0 license headers and fix file ending newlines for all test files
- Clean hardcoded paths and optimize robust project root path locating
- Remove unused project_cache_dir fixture and eliminate invalid code
- Add level2 marker in pytest.ini to resolve test warnings
- Define strict replay miss behavior: raise RuntimeError without live LLM fallback
- Pass all cache unit tests and system tests successfully
- Run autopep8 to format code and comply with PEP8 standards
- Delete redundant test_bench_cpp_cpu_cache_real_samples.py file
- Verify file compliance with SPEC requirements
- style(cache): format code per specification and pass all validations and tests
| 1 个月前 |
| test(cache): complete test specification rectification, directory migration and code normalization
- Remove obsolete cache test directory in core_v2 and clean up redundant files
- Migrate test files as per SPEC: UT to tests/ut/cache_tests, ST to tests/op/st
- Add Apache 2.0 license headers and fix file ending newlines for all test files
- Clean hardcoded paths and optimize robust project root path locating
- Remove unused project_cache_dir fixture and eliminate invalid code
- Add level2 marker in pytest.ini to resolve test warnings
- Define strict replay miss behavior: raise RuntimeError without live LLM fallback
- Pass all cache unit tests and system tests successfully
- Run autopep8 to format code and comply with PEP8 standards
- Delete redundant test_bench_cpp_cpu_cache_real_samples.py file
- Verify file compliance with SPEC requirements
- style(cache): format code per specification and pass all validations and tests
| 1 个月前 |
| LLM 输出去 JSON 化改造 + stream 模式支持
- KernelGen/Coder: 移除 JSON parser 依赖，改用代码提取 + py_compile 语法校验
- 删除 robust_parse 节点和相关 parser 配置
- 新增 stream 模式支持
- 新增 KernelGen 基准测试工具 (bench_kernel_gen.py)
- 更新依赖文件 requirements
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
| 2 个月前 |
| Refactor: remove legacy code, standardize test structure, and enhance core module test coverage
| 2 个月前 |
| LLM 输出去 JSON 化改造 + stream 模式支持
- KernelGen/Coder: 移除 JSON parser 依赖，改用代码提取 + py_compile 语法校验
- 删除 robust_parse 节点和相关 parser 配置
- 新增 stream 模式支持
- 新增 KernelGen 基准测试工具 (bench_kernel_gen.py)
- 更新依赖文件 requirements
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
| 2 个月前 |
| init fix codegen
| 1 个月前 |
| refactor tests dir
| 3 个月前 |
| refactor tests dir
| 3 个月前 |
| LLM 输出去 JSON 化改造 + stream 模式支持
- KernelGen/Coder: 移除 JSON parser 依赖，改用代码提取 + py_compile 语法校验
- 删除 robust_parse 节点和相关 parser 配置
- 新增 stream 模式支持
- 新增 KernelGen 基准测试工具 (bench_kernel_gen.py)
- 更新依赖文件 requirements
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
| 2 个月前 |
| feat: 优化 baseline profile 性能，避免重复测量
在 evolve 和 adaptive_search 流程中预先测量一次 baseline，后续所有任务
复用该结果，避免重复测量，显著提升执行效率。
核心改进：
- 新增 baseline_profiler.py 模块，提供统一的 baseline profiling 接口
- AdaptiveSearchController 和 evolve.py 在开始前预先测量 baseline
- Worker 层支持 override_base_time_us，直接使用缓存的 baseline 值
- 简化 KernelVerifier 逻辑，删除冗余的 speedup 重复计算
- 保存 baseline profile 脚本到 log 目录，方便复现
- 新增系统测试验证优化逻辑
性能收益：
- adaptive_search (100 任务): 节省 ~8 分钟
- evolve (50 任务): 节省 ~4 分钟
向后兼容，如果 baseline profile 失败，回退到原有逻辑。
支持 LocalWorker 和 RemoteWorker，无需额外配置。
Made-with: Cursor
| 2 个月前 |
| feat(pypto): support pypto
| 3 个月前 |
| remove legacy code
| 2 个月前 |
| refactor: rename project from aikg to akg_agents
Complete project renaming:
- Renamed main directory: aikg/ -> akg_agents/
- Renamed Python package: ai_kernel_generator -> akg_agents
- Renamed benchmark: aikgbench -> akg_kernels_bench
- Updated client API: AIKGClient -> AKGAgentsClient
- Updated all import paths and module references
- Updated environment variables: AIKG_* -> AKG_AGENTS_*
- Updated output directories and log paths
- Updated all documentation and configuration files
- Fixed Git submodules paths
Breaking changes:
- All Python imports need to update to 'from akg_agents'
- Environment variable names changed: AIKG_* -> AKG_AGENTS_*
- Package name changed from ai_kernel_generator to akg_agents
- Client class renamed: AIKGClient -> AKGAgentsClient
Co-authored-by: Cursor <cursoragent@cursor.com>
| 3 个月前 |
| add cache pt for triton-cuda-to-ascend
bugfix: cuda cache
bugfix: baseline profiler get proper device id
update .gitignore
support batch triton-cuda-to-ascend search
bugfix: print ms to us
update scripts
| 1 个月前 |
| Remove deprecated files
| 3 个月前 |
| add triton-ascend triton-ascend-case-matmul-large-k SKILL.md
| 1 个月前 |
| adapt sol benchmark
| 1 个月前 |
| Enhance hf_id preprocessing in SOL templates
| 1 个月前 |
| feat(pypto): support pypto
| 3 个月前 |
| fix(op): enhance verification timeout mechanisms to prevent infinite hangs
- Add fail-fast mechanism in to abort early after consecutive timeouts
- Implement process group killing () in to ensure child processes are terminated on timeout
- Add global (default 1800s) in to prevent infinite loops
- Fix compatibility in verification templates for Windows
- Pass explicit timeout to and profilers
Made-with: Cursor
| 2 个月前 |
| docs: split triton api references and add aggregator
| 2 个月前 |
| docs: split triton api references and add aggregator
| 2 个月前 |
