| # Full pipeline for Mistral-7B | |
| set -e | |
| echo "=== Per-Head KV Cache Compression — Mistral-7B ===" | |
| echo "Step 1: Download model" | |
| hf download mistralai/Mistral-7B-Instruct-v0.3 --local-dir ./mistral-model | |
| echo "Step 2: Baseline" | |
| python3 scripts/baseline.py mistral-7b | |
| echo "Step 3: Calibrate (20 min)" | |
| python3 scripts/calibrate.py mistral-7b | |
| echo "Step 4: Run quantized inference" | |
| python3 scripts/integrate.py mistral-7b | |
| echo "Step 5: Full benchmark" | |
| python3 scripts/benchmark.py mistral-7b | |
| echo "Step 6: Generate graphs" | |
| python3 scripts/visualize_results.py | |
| python3 scripts/visualize_long_context.py | |
| echo "=== Done! Check results/ and figures/ ===" | |