master

cann-robotrefactor(modeling): Restructure modeling/ to match TileGym layout

d38b8a3a创建于 4月29日历史提交

文件	最后提交记录	最后更新时间
input_prompt_256K.txt	refactor(modeling): Restructure modeling/ to match TileGym layout Co-authored-by: u010986451<suntao37@huawei.com> # message auto-generated for no-merge-commit merge: !12 merge refactor/modeling-restructure into master refactor(modeling): Restructure modeling/ to match TileGym layout Created-by: u010986451 Commit-by: u010986451 Merged-by: cann-robot Description: Reorganize the modeling/ directory to follow the TileGym pattern with a unified transformers/ entry point containing infer.py, benchmark scripts, sample inputs, Dockerfile, and README. Changes: - Create src/pypto_gym/transformers/ with Qwen3-1.7B model definitions and PyPTO kernel adapters (moved from modeling/qwen3_1_7b/core/ and modeling/qwen3_1_7b/qwen3_pto_kernels/) - Create modeling/transformers/ with unified infer.py (supports --use_pypto, benchmark mode, per-token step timing) - Add bench_qwen3_1_7b.sh for PyTorch baseline vs PyPTO comparison - Add sample_inputs/, Dockerfile, build_docker.sh, README.md - Remove old modeling/qwen3_1_7b/ directory See merge request: cann/pypto-gym!12	2 个月前
input_prompt_32K.txt	refactor(modeling): Restructure modeling/ to match TileGym layout Co-authored-by: u010986451<suntao37@huawei.com> # message auto-generated for no-merge-commit merge: !12 merge refactor/modeling-restructure into master refactor(modeling): Restructure modeling/ to match TileGym layout Created-by: u010986451 Commit-by: u010986451 Merged-by: cann-robot Description: Reorganize the modeling/ directory to follow the TileGym pattern with a unified transformers/ entry point containing infer.py, benchmark scripts, sample inputs, Dockerfile, and README. Changes: - Create src/pypto_gym/transformers/ with Qwen3-1.7B model definitions and PyPTO kernel adapters (moved from modeling/qwen3_1_7b/core/ and modeling/qwen3_1_7b/qwen3_pto_kernels/) - Create modeling/transformers/ with unified infer.py (supports --use_pypto, benchmark mode, per-token step timing) - Add bench_qwen3_1_7b.sh for PyTorch baseline vs PyPTO comparison - Add sample_inputs/, Dockerfile, build_docker.sh, README.md - Remove old modeling/qwen3_1_7b/ directory See merge request: cann/pypto-gym!12	2 个月前
input_prompt_small.txt	refactor(modeling): Restructure modeling/ to match TileGym layout Co-authored-by: u010986451<suntao37@huawei.com> # message auto-generated for no-merge-commit merge: !12 merge refactor/modeling-restructure into master refactor(modeling): Restructure modeling/ to match TileGym layout Created-by: u010986451 Commit-by: u010986451 Merged-by: cann-robot Description: Reorganize the modeling/ directory to follow the TileGym pattern with a unified transformers/ entry point containing infer.py, benchmark scripts, sample inputs, Dockerfile, and README. Changes: - Create src/pypto_gym/transformers/ with Qwen3-1.7B model definitions and PyPTO kernel adapters (moved from modeling/qwen3_1_7b/core/ and modeling/qwen3_1_7b/qwen3_pto_kernels/) - Create modeling/transformers/ with unified infer.py (supports --use_pypto, benchmark mode, per-token step timing) - Add bench_qwen3_1_7b.sh for PyTorch baseline vs PyPTO comparison - Add sample_inputs/, Dockerfile, build_docker.sh, README.md - Remove old modeling/qwen3_1_7b/ directory See merge request: cann/pypto-gym!12	2 个月前