lecture 06, AttributeError: 'JITFunction' object has no attribute 'cache' for triton-3.5.0

[marvin-0042]

Hi Master,

Thanks for the great lecture! it's very solid and deep for developers who need systematic training in AI infra.

requirements.txt doesn't identify torch triton version, when I install latest torch (torch-2.9.0) which needs triton>=3.5.0,
python execute.py -m lecture_06
hit below issue like due to triton JITFunction interface changed

File "/workspace/spring2025-lectures/lecture_06.py", line 607, in print_ptx
return list(kernel.cache[0].values())[0].asm["ptx"]
^^^^^^^^^^^^
AttributeError: 'JITFunction' object has no attribute 'cache'

pip uninstall torch triton -y
pip install torch==2.6.0 triton==3.2.0

reverted triton to 3.2.0 (and torch back to 2.6.0), issue was fixed.

shall we limit triton / torch version into requirements.txt? or revise the kernel.cache according to latest triton API?
I can submit PR for requirements.txt if that helps.

[8147 lecture_06.py:601] """Print out the PTX code generated by Triton for the given kernel."""
[8148 lecture_06.py:602] ptx_path = f"var/{name}-ptx.txt"
[8149 lecture_06.py:603] text("Let's go poke around at the PTX code.")
[8150 lecture_06.py:604] link(get_local_url(ptx_path))
[8151 lecture_06.py:606] with open(ptx_path, "w") as f:
[8152 lecture_06.py:607] return list(kernel.cache[0].values())[0].asm["ptx"]
[8153 lecture_06.py:607] return list(kernel.cache[0].values())[0].asm["ptx"]
[8153 lecture_06.py:606] with open(ptx_path, "w") as f:
[8154 lecture_06.py:586] ptx = print_ptx("triton_gelu", triton_gelu_kernel)
[8155 lecture_06.py:586] ptx = print_ptx("triton_gelu", triton_gelu_kernel)
[8155 lecture_06.py:508] print_ptx_main() # Look at the generated instructions
[8156 lecture_06.py:508] print_ptx_main() # Look at the generated instructions
[8156 lecture_06.py:477] triton_gelu_main()
[8157 lecture_06.py:477] triton_gelu_main()
[8157 lecture_06.py:30] triton_kernels() # Write kernels in Python
[8158 lecture_06.py:30] triton_kernels() # Write kernels in Python
Traceback (most recent call last):
File "/workspace/spring2025-lectures/execute.py", line 264, in
trace = execute(module_name=module, inspect_all_variables=args.inspect_all_variables)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "/workspace/spring2025-lectures/execute.py", line 241, in execute
module.main()
File "/workspace/spring2025-lectures/lecture_06.py", line 30, in main
triton_kernels() # Write kernels in Python
^^^^^^^^^^^^^^^^
File "/workspace/spring2025-lectures/lecture_06.py", line 477, in triton_kernels
triton_gelu_main()
File "/workspace/spring2025-lectures/lecture_06.py", line 508, in triton_gelu_main
print_ptx_main() # Look at the generated instructions
^^^^^^^^^^^^^^^^
File "/workspace/spring2025-lectures/lecture_06.py", line 586, in print_ptx_main
ptx = print_ptx("triton_gelu", triton_gelu_kernel)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "/workspace/spring2025-lectures/lecture_06.py", line 607, in print_ptx
return list(kernel.cache[0].values())[0].asm["ptx"]
^^^^^^^^^^^^
AttributeError: 'JITFunction' object has no attribute 'cache'