tahamajs/RR/Codes/tests - Storage Bucket

Test Suite Subsystem (Codes/framework/tests)

The tests subsystem provides a comprehensive pytest test harness verifying the mathematical correctness, numerical stability, zero-shot fidelity, gradient flow, and hardware performance of all framework components.

---

🏗 Test Organization

• Attention Mechanism Tests (test_attention_modules.py, test_attentions.py): Validates shape correctness $, numerical finiteness (no NaN/Inf), gradient flow, padding mask compliance, multi-head reshaping, and zero-shot weight transfer across all registered attention variants.
• Attention Injection Engine (test_attention_injection.py): Verifies universal layer replacement on real HuggingFace transformer architectures with both copy_weights=True and copy_weights=False.
• Fused Kernel Verification (test_fused_svd_kernel.py): Analytical gradient checks (gradcheck), forward/backward numerical parity against un-fused references, and RMSNorm integration.
• Compression & SVD Specialist Suites (test_compressors.py, test_all_compression_engines.py, test_svd_audit_suite.py): Verifies SVD truncation, rank selection algorithms, and weight reconstruction fidelity.
• Dataset & Trainer Verification (test_datasets.py, test_results_saver.py, test_multi_mode.py): Checks data loaders, tokenizer collators, trainer factory resolution, and artifact serialization.

---

📦 Key Test Files

Test File	Target Subsystem / Verification Scope
test_attention_injection.py	Universal attention injection into DistilBERT, BERT, GPT-2 models.
test_attention_modules.py	Exhaustive 7-point verification suite across all attention modules.
test_fused_svd_kernel.py	Fused SVD linear forward/backward numerical stability and gradchecks.
test_all_compression_engines.py	Verification of spectral, pruning, and low-rank compression algorithms.
test_results_saver.py	Multi-backend serialization (JSON, Parquet, checkpoints).
test_datasets.py	GLUE dataset loading, tokenization pipelines, and synthetic generator.
conftest.py	Shared pytest fixtures (model configs, hidden states, attention masks).

---

🚀 Running the Test Suite

1. Run Fast Core Tests

pytest Codes/framework/tests/test_attention_injection.py Codes/framework/tests/test_results_saver.py Codes/framework/tests/test_fused_svd_kernel.py -v

2. Run Attention Mechanism Tests

pytest Codes/framework/tests/test_attention_modules.py -v -k "svd or linear or full"

3. Run All Integration Tests

pytest Codes/framework/tests/ -m "integration" -v