# The files in profile folder are not json
**URL:** https://discuss.vllm.ai/t/the-files-in-profile-folder-are-not-json/1985
**Category:** General
**Created:** [November 20, 2025, 12:31am UTC](https://discuss.vllm.ai/t/the-files-in-profile-folder-are-not-json/1985 "2025-11-20T00:31:32Z")
**Posts on this page:** 2
**Page:** 1
### Author:  [@blackdamn2](https://discuss.vllm.ai/u/blackdamn2)
#### Post date: [November 20, 2025, 12:31am UTC](https://discuss.vllm.ai/t/the-files-in-profile-folder-are-not-json/1985/1 "2025-11-20T00:31:32Z")
I’m using vllm’s openai server and enabled the mode via post /start\_profile (having configured PROFILE\_DIR beforehand). However, in the profile\_dir, there are first torch.op\_mark and torch.op\_range files, followed by some slice.01234 files and a done file. I wonder if this is normal, and how to combine these files together.
---
### Author:  [@RunLLM](https://discuss.vllm.ai/u/RunLLM)
#### Post date: [November 20, 2025, 12:31am UTC](https://discuss.vllm.ai/t/the-files-in-profile-folder-are-not-json/1985/2 "2025-11-20T00:31:50Z")
Yes, this is normal. When profiling is enabled in vLLM’s OpenAI server (with `VLLM_TORCH_PROFILER_DIR` set and `/start_profile` called), the profiler writes intermediate files like `torch.op_mark`, `torch.op_range`, `slice.*`, and a `done` file as part of its trace output. These files are generated by the PyTorch profiler and are intended to be read by visualization tools such as Perfetto or TensorBoard, not manually combined by the user. You do not need to merge them yourself—just point Perfetto ([https://ui.perfetto.dev/](https://ui.perfetto.dev/)) or TensorBoard at the profile directory to view the results. For more, see [Profiling vLLM](https://docs.vllm.ai/en/latest/contributing/profiling/profiling_index.html) and [example usage](https://docs.vllm.ai/en/latest/getting_started/examples/simple_profiling.html).
Would you like more detail on how to visualize or interpret these files?
Sources:
- [Profiling vLLM](https://docs.vllm.ai/en/latest/contributing/profiling/profiling_index.html)
- [example usage](https://docs.vllm.ai/en/latest/getting_started/examples/simple_profiling.html)
Hint: Mention @RunLLM in the post for followups.