Initial test suite setup

[colesmcintosh]

This PR introduces a test suite.

Testing Infrastructure

1. Kernel Testing Suite:

   • Implemented comprehensive test_kernels.py for unit testing kernel modules
   • Added tests for critical components: cross_entropy_loss, fast_lora, flex_attention, geglu, layernorm, rms_layernorm, rope_embedding, and swiglu
   • Included smoke tests and forward pass checks
   • Added graceful handling for CUDA unavailability

2. Model Testing Suite:

   • Created test_models.py for validating model module imports
   • Implemented parameterized tests for thorough model import validation
   • Added smoke tests to verify unsloth.models package integrity
   • Implemented graceful handling of NVIDIA GPU availability in tests

Repository Management

• Added .gitignore file for better version control

These changes significantly improve the testing coverage and development workflow while ensuring robust handling of dependencies and hardware configurations.

[shimmyshimmer]

Thank you for this. We'll take a look

[darkacorn]

i think the idea is correct

however .. - i think we have to test in isolation with different versions as even if the core unit it self passes .. dependencies may fail

a ci/cd pipeline test with different cuda versions would probably be the higher value - and we could run that with github runners

happy to hear other input on that - but i agree .. more automated unit / integration testing is a good idea going forward