Add Medium FP16 Dot Product Challenge (#58)

[Copilot]

Plan: Add Medium FP16 Dot Product Challenge

Based on repository analysis, I will create challenge #58 - FP16 Dot Product, which will be a 16-bit floating point version of the existing dot product challenge.

• Explore repository structure and understand existing challenges
• Identify challenge number (58) and understand file structure
• Create challenge directory: challenges/medium/58_fp16_dot_product/
• Create challenge.py with reference implementation and tests
• Create challenge.html with problem description
• Generate starter code files (created manually due to network restrictions)
• Verify all files are created correctly
• Fix CUDA starter to use 'half' instead of '__half' for consistency
• Remove CUDA-specific references from challenge description
• Update performance test size to 100M elements
• Test the challenge structure matches existing patterns

Implementation Details

Created a complete FP16 dot product challenge that mirrors the existing FP32 dot product challenge but uses 16-bit floating point numbers:

Challenge Files:

• challenge.py: Reference implementation using torch.float16, converts to FP32 for accumulation
• challenge.html: Problem description with examples and constraints (framework-agnostic)
• starter/starter.cu: CUDA starter with half type (consistent with other fp16 challenges)
• starter/starter.mojo: Mojo starter with Float16 type
• starter/starter.pytorch.py: PyTorch starter
• starter/starter.triton.py: Triton starter
• starter/starter.cute.py: CuTe starter

Key Differences from FP32 version:

• Uses torch.float16 / half / Float16 types
• Higher tolerance values (atol=5e-2, rtol=5e-2) for FP16 precision
• Reference implementation converts to FP32 for accurate accumulation
• Framework-agnostic problem description (mentions "GPU program" instead of "CUDA program")
• Performance test uses 100M elements (400MB for 2 vectors) - fits comfortably in T4 memory

Validation:

• ✅ All Python files pass syntax validation
• ✅ Directory structure matches existing challenges
• ✅ File naming conventions followed
• ✅ Starter code format consistent with repository patterns
• ✅ Challenge numbering is unique (58)
• ✅ Framework-agnostic wording following FP16 batched matmul pattern
• ✅ Performance test size appropriate for GPU benchmarking

Original prompt

Add medium fp16 dot product

---

💡 You can make Copilot smarter by setting up custom instructions, customizing its development environment and configuring Model Context Protocol (MCP) servers. Learn more Copilot coding agent tips in the docs.