feat: Multi-Backend Comparison Report (M114)

ROADMAP.md

@@ -1518,9 +1518,9 @@ Help users find the **optimal Prefill:Decode instance ratio** based on **real be
 - Programmatic `import_trtllm()` and `import_trtllm_data()` API
 - 25+ new tests
 
-### M113 🔄 TensorRT-LLM Benchmark Command Generator
+### M113 ✅ TensorRT-LLM Benchmark Command Generator
 
-*In progress*
+*Completed — PR #250*
 
 - `TRTLLMCommandGenerator` class in `trtllm_commands.py`
 - `TRTLLMCommandConfig`, `TRTLLMServerCommand`, `TRTLLMBenchmarkCommand`, `TRTLLMCommandSet` Pydantic models
@@ -1530,3 +1530,16 @@ Help users find the **optimal Prefill:Decode instance ratio** based on **real be
 - CLI `trtllm-commands` subcommand with table + JSON output
 - Programmatic `generate_trtllm_commands()` API
 - 29 new tests
+
+### M114 🔄 Multi-Backend Comparison Report
+
+*In progress*
+
+- `BackendComparator` class in `backend_compare.py`
+- `BackendComparisonConfig`, `BackendMetrics`, `BackendComparisonReport`, `BackendRanking`, `SLAResult` Pydantic models
+- Auto-detect input format (native, vLLM, SGLang, TensorRT-LLM)
+- Per-backend latency percentiles (P50/P95/P99), throughput, SLA compliance
+- Rank backends by configurable criteria (ttft_p99, tpot_p99, total_latency_p99, throughput)
+- CLI `compare-backends` subcommand with `--benchmark`, `--labels`, `--formats`, `--rank-by`, table + JSON output
+- Programmatic `compare_backends()` API
+- ~25 new tests

docs/iterations/current.md

@@ -65,4 +65,5 @@ The project has completed **110 milestones**, covering the full feature chain fr
 | 4 | 2026-04-06 | M110 SGLang Benchmark Format Importer | ✅ merged | PR #244 |
 | 5 | 2026-04-06 | M111 SGLang Benchmark Command Generator | ✅ merged | PR #246 |
 | 6 | 2026-04-06 | M112 TensorRT-LLM Benchmark Format Importer | ✅ merged | PR #248, both bots approved |
-| 7 | 2026-04-06 | M113 TensorRT-LLM Benchmark Command Generator | ⏳ pending review | Issue #249 |
+| 7 | 2026-04-06 | M113 TensorRT-LLM Benchmark Command Generator | ✅ merged | PR #250, both bots approved |
+| 8 | 2026-04-06 | M114 Multi-Backend Comparison Report | ⏳ pending review | Issue #251 |

src/xpyd_plan/__init__.py

@@ -1493,3 +1493,27 @@
     "import_trtllm",
     "import_trtllm_data",
 ]
+
+from xpyd_plan.backend_compare import (  # noqa: E402
+    BackendComparator,
+    BackendComparisonConfig,
+    BackendComparisonReport,
+    BackendFormat,
+    BackendMetrics,
+    BackendRanking,
+    RankCriteria,
+    SLAResult,
+    compare_backends,
+)
+
+__all__ += [
+    "BackendComparator",
+    "BackendComparisonConfig",
+    "BackendComparisonReport",
+    "BackendFormat",
+    "BackendMetrics",
+    "BackendRanking",
+    "RankCriteria",
+    "SLAResult",
+    "compare_backends",
+]