-
Notifications
You must be signed in to change notification settings - Fork 167
asm_fmoe_codegen #702
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.
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
asm_fmoe_codegen #702
Conversation
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Pull Request Overview
This PR introduces code generation for ASM FMOE (Fused Mixture of Experts) kernel configurations by adding CSV configuration files and Python codegen scripts. The changes support various data types (fp16, bf16), quantization schemes (Int8, Fp8, blockscale), and activation functions (gelu, silu) for the gfx942 GPU architecture.
- Addition of 19 CSV configuration files containing kernel specifications for different FMOE variants
- Python code generator script to convert CSV configurations into C++ header file with kernel metadata
- Updates to function signatures to support activation parameter passing
- Integration with the build system to generate ASM FMOE configurations
Reviewed Changes
Copilot reviewed 24 out of 417 changed files in this pull request and generated 1 comment.
Show a summary per file
| File | Description |
|---|---|
| hsa/gfx942/fmoe/*.csv | Configuration files defining kernel parameters for various FMOE data type and activation combinations |
| hsa/gfx942/fmoe/codegen.py | Python script that generates C++ header file from CSV configuration data |
| csrc/include/moe_op.h | Updated function signatures to include activation parameter and improve formatting |
| aiter/ops/moe_op.py | Added activation parameter to fmoe function signature |
| aiter/jit/optCompilerConfig.json | Updated build configuration to include FMOE codegen in blob generation |
| aiter/fused_moe_bf16_asm.py | Added activation parameter to fmoe function call |
No description provided.