Liger Auto-Patch

Adds Liger Kernel optimization support for a new HuggingFace model, or modifies existing monkey-patching, through a staged pipeline with human review between stages. Supports creating new model patches and modifying existing ones.

Mode Detection

• Create mode: User asks to add/patch/support a new model → full pipeline (Analyze → Generate → Validate)
• Modify mode: User asks to update/fix/change/extend an existing monkey-patch → lighter pipeline (Change Impact Analysis → Apply Changes → Validate)

Keywords that suggest modify mode: update, fix, change, add [kernel] to [existing model], extend, modify, new activation, new norm, bug in patch, upstream changed

Pipeline (Create Mode)

Stage 1: Analyze

Follow the Model Analyzer workflow in model-analyzer.md. If the host runtime supports parallel subagents, this stage may be delegated to one; otherwise execute the workflow directly.

This stage reads the HF modeling_*.py source and produces a model profile answering 12 architectural questions from decision-matrix.md.

Human checkpoint: Present the profile. Confirm before proceeding.

Stage 2: Generate

Follow the Code Generator workflow in code-generator.md.

Generates/modifies up to 13 files:

1. src/liger_kernel/transformers/model/{model}.py — NEW lce_forward
2. src/liger_kernel/transformers/monkey_patch.py — MODIFY
3. src/liger_kernel/transformers/__init__.py — MODIFY
4. src/liger_kernel/transformers/model/output_classes.py — MODIFY if needed
5. test/transformers/test_monkey_patch.py — MODIFY
6. test/convergence/bf16/test_mini_models.py — MODIFY (FLCE path)
7. test/convergence/bf16/test_mini_models_with_logits.py — MODIFY (non-FLCE path)
8. test/convergence/fp32/test_mini_models.py — MODIFY (FLCE path)
9. test/convergence/fp32/test_mini_models_with_logits.py — MODIFY (non-FLCE path)
10. test/convergence/bf16/test_mini_models_multimodal.py — MODIFY if VL model
11. test/convergence/fp32/test_mini_models_multimodal.py — MODIFY if VL model
12. test/utils.py — MODIFY
13. README.md — MODIFY

Human checkpoint: Present changes for review.

Stage 3: Validate

Follow the Validator workflow in validator.md.

Runs instance patching test, convergence test, and lint check. Retries up to 3 times on failure.

Human checkpoint: Report final test results.

Pipeline (Modify Mode)

Stage 1: Change Impact Analysis

Read the existing apply_liger_kernel_to_{model_type} function in monkey_patch.py and the relevant section of the upstream HF modeling_{model_type}.py. Produce a short change plan:

• What is being added/changed/fixed
• Which Liger kernel(s) are involved
• Which files need modification (subset of the 13 files from create mode)
• What the expected behavior should be after the change

Human checkpoint: Present the change plan. Confirm before proceeding.

Stage 2: Apply Changes

Follow the Code Generator workflow in code-generator.md in modify mode.

Human checkpoint: Present changes for review.

Stage 3: Validate

Follow the Validator workflow in validator.md. This stage is mandatory — do not skip it. At minimum, run:

1. Instance patching test: pytest test/transformers/test_monkey_patch.py -k "{model_type}" -xvs
2. All convergence tests for the model:
   • pytest test/convergence/bf16/test_mini_models.py -k "{model_type}" -xvs (FLCE, bf16)
   • pytest test/convergence/bf16/test_mini_models_with_logits.py -k "{model_type}" -xvs (non-FLCE, bf16)
   • pytest test/convergence/fp32/test_mini_models.py -k "{model_type}" -xvs (FLCE, fp32)
   • pytest test/convergence/fp32/test_mini_models_with_logits.py -k "{model_type}" -xvs (non-FLCE, fp32)
   • If VL (multimodal) model, also run:
     • pytest test/convergence/bf16/test_mini_models_multimodal.py -k "{model_type}" -xvs
     • pytest test/convergence/fp32/test_mini_models_multimodal.py -k "{model_type}" -xvs
3. Checkstyle: make checkstyle

Human checkpoint: Report final test results.

Reference Files

• decision-matrix.md — 12 architectural decisions to resolve per model
• examples/llama-profile.md — Reference profile for standard dense model
• examples/gemma-profile.md — Reference profile showing GeGLU + offset variant
• Templates in templates/ — Code generation patterns for each file type