# BurstGPT AIPerf Window CSVs

These CSVs are extracted from `BurstGPT_without_fails_{1,2}.csv` for the selected stable A/B/C regimes. Timestamps are rebased to seconds from each segment's start while preserving within-segment arrival timing.

| Segment | File | Duration (s) | Requests | RPS | Original start (s) | Original end (s) |
|---|---|---:|---:|---:|---:|---:|
| A | `burstgpt_A.csv` | 600 | 131 | 0.218 | 10300200 | 10300800 |
| A1 | `burstgpt_A1.csv` | 300 | 63 | 0.210 | 10300200 | 10300500 |
| A2 | `burstgpt_A2.csv` | 300 | 68 | 0.227 | 10300500 | 10300800 |
| B | `burstgpt_B.csv` | 600 | 1,201 | 2.002 | 5389800 | 5390400 |
| B1 | `burstgpt_B1.csv` | 300 | 604 | 2.013 | 5389800 | 5390100 |
| B2 | `burstgpt_B2.csv` | 300 | 597 | 1.990 | 5390100 | 5390400 |
| C | `burstgpt_C.csv` | 600 | 6,204 | 10.340 | 849000 | 849600 |
| C1 | `burstgpt_C1.csv` | 300 | 3,029 | 10.097 | 849000 | 849300 |
| C2 | `burstgpt_C2.csv` | 300 | 3,175 | 10.583 | 849300 | 849600 |

Example:

```bash
aiperf profile \
  --model <MODEL> \
  --endpoint-type chat \
  --streaming \
  --url localhost:8000 \
  --input-file windows/burstgpt_A1.csv \
  --custom-dataset-type burst_gpt_trace \
  --fixed-schedule
```