Skip to content

[Examples] add Python runner for BuddyDeepSeekR1 E2E inference #699

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Open
GuoningHuang wants to merge 2 commits into
base: main
Choose a base branch
from
Open

[Examples] add Python runner for BuddyDeepSeekR1 E2E inference #699

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
98e5589
[Examples] add Python AOT runner for BuddyDeepSeekR1 subgraph end-to-…
GuoningHuang Feb 25, 2026
1d07437
[Examples] add Python AOT runner for BuddyDeepSeekR1 subgraph end-to-…
GuoningHuang Mar 1, 2026
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
29 changes: 28 additions & 1 deletion examples/BuddyDeepSeekR1/README.md
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -116,7 +116,34 @@ $ ./bin/buddy-deepseek-r1-cli --interactive --no-stats
- Use options like `--max-tokens` and `--eos-id` to constrain the generation length/termination. Add `--no-stats` when you want pure text output without the performance summary.
- `--interactive` starts a REPL similar to `buddy-deepseek-r1-main.cpp`, handling one prompt and one response at a time. `--prompt` can act as a system prefix prepended to every user entry.

6. Enjoy it!
6. Python end-to-end inference by executing exported subgraphs

This path reuses artifacts already generated under `build/examples/BuddyDeepSeekR1`, then uses Python to run prefill/decode MLIR subgraphs end to end.

```bash
# from buddy-mlir/examples/BuddyDeepSeekR1
python3 run-subgraphs-python.py \
--prompt "Hello, who are you?" \
--artifact-dir path_to_buddyDeepSeekR1_build \
--llvm-build-dir path_to_llvm_build
```

Example:

```bash
# from buddy-mlir/examples/BuddyDeepSeekR1
python3 run-subgraphs-python.py --prompt "hello" --artifact-dir ../../build/examples/BuddyDeepSeekR1 --llvm-build-dir ../../llvm/build --omp-num-threads 48 --omp-proc-bind close
```

Useful options:

- `--artifact-dir`: directory containing `forward_prefill.mlir`, `forward_decode.mlir`, `arg0.data` (default: `build/examples/BuddyDeepSeekR1`).
- `--llvm-build-dir`: LLVM build directory that provides `libmlir_runner_utils.so` and related runtime libs (default: `llvm/build`).
- `--model-path`: HF model id or local model path for tokenizer.
- `--max-new-tokens`: generated token upper bound.
- `--omp-num-threads`, `--omp-proc-bind`, `--omp-places`, `--kmp-affinity`: optional OpenMP tuning knobs.

7. Enjoy it!

## How to run on RISC-V machine

Expand Down
Loading