MODEL_GPU_CONFIGURATION.md

Model & GPU Configuration

Current Model Configuration

Parameter	Value	Purpose
Model	glm-5 (opus)	Primary reasoning model
Temperature	0.9	Creativity & brainstorming
Context Window	128K tokens	DeepSeek-Reasoner limit

GPU Resources

Resource	Limit	Notes
GPU	NVIDIA RTX 4050 (6GB VRAM)	CuPy 14.0.1, CUDA 13.1.1
Max Utilization	80%	Leave 20% for system
Usable VRAM	~4GB	Leave 2GB for OS
Batch Size Guideline	matrix_dim < 2000	For 6GB VRAM

Simulation Guidelines

1. Use GPU for:

   • Matrix operations (CuPy)
   • Monte Carlo simulations
   • Large-scale validation tests
   • Performance benchmarks

2. Back ideation with rigor:

   • Run simulations to validate claims
   • Generate empirical evidence
   • Document all assumptions
   • Provide statistical confidence intervals

3. GPU Resource Management:

   # Safe GPU usage pattern
   import cupy as cp
   
   # Check available memory
   mempool = cp.get_default_memory_pool()
   used_bytes = mempool.used_bytes()
   total_bytes = mempool.total_bytes()
   
   # Stay under 80%
   max_usage = 0.8 * total_bytes
   assert used_bytes < max_usage, "GPU usage exceeds 80%"

Temperature 0.9 Ideation Protocol

1. Brainstorming Phase: Generate multiple hypotheses
2. Simulation Phase: Test hypotheses with GPU compute
3. Validation Phase: Back creative ideas with rigorous proofs
4. Documentation Phase: Record all results with statistical measures

---

Configuration Active: 2026-03-13 Model: glm-5 (opus) Temperature: 0.9 GPU Limit: 80% utilization