test(perf): Add some `Llama-3_3-Nemotron-Super-49B-v1` integration-perf-tests (TRT flow, trtllm-bench)

[venkywonka]

Description

• Add some Llama-3_3-Nemotron-Super-49B-v1 integration-perf-tests (cpp backend, trtllm-bench).
• This also exposes a --trust_remote_code flag in the trtllm-bench-build subcommand, that is required for transformers library to use Autoclasses to load DeciLM-based models (Llama-Nemotron-Super being one of them).
• This PR also changes config.py and model.py for the DeciLMForCausalLM classes to have trust_remote_code=True by default (it was False by default previously) for thing to work smoothly without extra parametrizations when run from top-level trtllm-bench.

Performance Summary – llama_v3.3_nemotron_super_49b

isl	osl	quant	con	backend	req/s	tps /gpu	avg latency ms	p50 latency ms
5000	500	none	1	cpp	0.1075	13.4317	9 306.1785	9 305.0552
5000	500	fp8	1	cpp	0.1485	18.5636	6 733.4385	6 730.6310
5000	500	none	250	cpp	0.6116	76.4499	317 885.8769	401 171.5739
5000	500	fp8	250	cpp	0.7220	90.2495	269 376.7776	340 154.1910
500	2000	none	1	cpp	0.0304	15.2075	32 878.3526	32 877.1050
500	2000	fp8	1	cpp	0.0435	21.7563	22 981.6188	22 975.2227
500	2000	none	250	cpp	0.3274	163.7098	589 062.8547	733 682.4485
500	2000	fp8	250	cpp	0.4158	207.8830	463 903.2804	577 812.6816

---

Run Invariants

• Model: llama_v3.3_nemotron_super_49b
• Backend: cpp (builds TensorRT engines)
• Precision: BF16 baseline, FP8 quantized variants
• Max batch size: 16  •  GPUs: 4 (per-GPU throughput shown above)
• Benchmark tool: trtllm-bench
• Synthetic dataset: 512 sequences per run

---

Execution Status Matrix

backend	isl	osl	quant	con	status
cpp	5000	500	none	1	TIMEOUT
cpp	5000	500	fp8	1	TIMEOUT
cpp	5000	500	none	250	PASS
cpp	5000	500	fp8	250	PASS
cpp	500	2000	none	1	TIMEOUT
cpp	500	2000	fp8	1	TIMEOUT
cpp	500	2000	none	250	PASS
cpp	500	2000	fp8	250	PASS

[venkywonka]

/bot run --disable-fail-fast

[tensorrt-cicd]

PR_Github #4410 [ run ] triggered by Bot

[venkywonka]

/bot run --disable-fail-fast

[tensorrt-cicd]

PR_Github #4421 [ run ] triggered by Bot

[tensorrt-cicd]

PR_Github #4410 [ run ] completed with state ABORTED

[tensorrt-cicd]

PR_Github #4421 [ run ] completed with state SUCCESS /LLM/main/L0_MergeRequest_PR pipeline #3183 completed with status: 'FAILURE'

[venkywonka]

/bot run --disable-fail-fast

[tensorrt-cicd]

PR_Github #4472 [ run ] triggered by Bot

[tensorrt-cicd]

PR_Github #4472 [ run ] completed with state SUCCESS /LLM/main/L0_MergeRequest_PR pipeline #3209 completed with status: 'FAILURE'

[venkywonka]

/bot run --disable-fail-fast

[tensorrt-cicd]

PR_Github #4584 [ run ] triggered by Bot

[tensorrt-cicd]

PR_Github #4584 [ run ] completed with state SUCCESS /LLM/main/L0_MergeRequest_PR pipeline #3298 completed with status: 'SUCCESS'

[venkywonka]

/bot run --disable-fail-fast

[tensorrt-cicd]

PR_Github #4737 [ run ] triggered by Bot

[tensorrt-cicd]

PR_Github #4737 [ run ] completed with state SUCCESS /LLM/main/L0_MergeRequest_PR pipeline #3419 completed with status: 'FAILURE'

[venkywonka]

/bot run --disable-fail-fast

[tensorrt-cicd]

PR_Github #4903 [ run ] triggered by Bot

[tensorrt-cicd]

PR_Github #4903 [ run ] completed with state SUCCESS /LLM/main/L0_MergeRequest_PR pipeline #3555 completed with status: 'SUCCESS'

[venkywonka]

/bot run --disable-fail-fast

[tensorrt-cicd]

PR_Github #5014 [ run ] triggered by Bot

[tensorrt-cicd]

PR_Github #5014 [ run ] completed with state SUCCESS /LLM/main/L0_MergeRequest_PR pipeline #3647 completed with status: 'FAILURE'

[venkywonka]

/bot run --disable-fail-fast

[tensorrt-cicd]

PR_Github #5109 [ run ] triggered by Bot

[tensorrt-cicd]

PR_Github #5109 [ run ] completed with state SUCCESS /LLM/main/L0_MergeRequest_PR pipeline #3721 completed with status: 'FAILURE'

[venkywonka]

/bot run --disable-fail-fast

[tensorrt-cicd]

PR_Github #5216 [ run ] triggered by Bot

[tensorrt-cicd]

PR_Github #5216 [ run ] completed with state SUCCESS /LLM/main/L0_MergeRequest_PR pipeline #3808 completed with status: 'SUCCESS'

[venkywonka]

/bot run --disable-fail-fast

[tensorrt-cicd]

PR_Github #5534 [ run ] triggered by Bot

[tensorrt-cicd]

PR_Github #5534 [ run ] completed with state SUCCESS /LLM/main/L0_MergeRequest_PR pipeline #4035 completed with status: 'FAILURE'