RedHatAI/gemma-4-31B-it-NVFP4

completed
Request throughput
2.29 req/s
Output throughput
585.4 tok/s
p95 TTFT
232.3 ms
p95 TPOT
20.46 ms

Manifest warnings

Review these published caveats before interpreting or comparing the metrics.

  • Review imported facts against the local model, hardware, launch command, and benchmark command before publication.

Server launch command

CUDA_VISIBLE_DEVICES=3 vllm serve RedHatAI/gemma-4-31B-it-NVFP4 \
  --revision 5df75ccad9b74546f5be758a2a68acc2e16bbc9b \
  --port 9003 \
  --host 127.0.0.1 \
  --dtype bfloat16 \
  --gpu-memory-utilization 0.88 \
  --max-model-len 9216 \
  --max-num-batched-tokens 16384 \
  --max-num-seqs 32 \
  --tensor-parallel-size 1 \
  --pipeline-parallel-size 1 \
  --kv-cache-dtype fp8 \
  --distributed-executor-backend mp \
  --no-enable-prefix-caching \
  --enable-chunked-prefill \
  --async-scheduling \
  --no-enforce-eager \
  --served-model-name RedHatAI/gemma-4-31B-it-NVFP4 \
  --disable-uvicorn-access-log \
  --language-model-only \
  --speculative-config '{"method":"mtp","model":"google/gemma-4-31B-it-assistant","revision":"627c5ec1458b9086b841a91e0512fd31fd2fbbf1","num_speculative_tokens":1}'

Run details

Workload
1024 → 256
Methodology
closed loop · streaming
Test load
8 concurrent requests
Endpoint
openai-compatible-completions
Requests
16 unmeasured warmups per invocation after one readiness request.
Dataset
sha256=7b3a595f0a42c9e568ebe49422d9a636f6f46f8c66bc8cb4557f4f2a17650e52 generator=runpile-sweeps-tokenizer-verified-jsonl synthetic=true generator_revision=cab078b273cceada260983886369dfc896718627d072e44f74a1cca08b67675c shared_prefix_tokens=0
Sampling
mode=greedy temperature=0
Stopping
ignore_eos=true max_output_tokens=256
Prefix and cache
cache_state=warm-host cache_enabled=false shared_prefix_tokens=0
Client
name=vllm bench serve backend=vllm version=0.28.0 placement=same-host
Hardware
1× NVIDIA GeForce RTX 5090
Runtime
vllm 0.28.0
Decoding method
Speculative decoding
mtp assistant model
google/gemma-4-31B-it-assistantrevision 627c5ec1458b9086b841a91e0512fd31fd2fbbf1
Runtime configuration
dtype=bfloat16 max_num_seqs=32 enforce_eager=false max_model_len=9216 kv_cache_dtype=fp8 prefix_caching=false chunked_prefill=true expert_parallel=false async_scheduling=true gpu_memory_utilization=0.88 max_num_batched_tokens=16384
Measurement
vllm bench serve 0.28.0
Timing
client-observed streaming request latency · client overhead included
Repetitions
1
Failure handling
Failed requests counted; no client retries.
Quantization
format=NVFP4 target plus BF16 MTP assistant checkpoint=compressed-tensors
Model revision
5df75ccad9b74546f5be758a2a68acc2e16bbc9b
Evidence
Raw benchmark evidence attached · eligible and ranked
Published
by anonymous-EVjB2S

Benchmark command

vllm bench serve \
  --backend vllm \
  --base-url http://127.0.0.1:9003 \
  --endpoint /v1/completions \
  --model RedHatAI/gemma-4-31B-it-NVFP4 \
  --tokenizer RedHatAI/gemma-4-31B-it-NVFP4 \
  --dataset-name custom \
  --dataset-path <EXACT_SYNTHETIC_DATASET> \
  --custom-output-len 256 \
  --skip-chat-template \
  --disable-shuffle \
  --num-prompts 256 \
  --max-concurrency 8 \
  --request-rate inf \
  --seed 0 \
  --ignore-eos \
  --temperature 0 \
  --percentile-metrics ttft,tpot,itl,e2el \
  --metric-percentiles 50,90,95,99 \
  --disable-tqdm \
  --save-result \
  --result-dir <local-path> \
  --result-filename gemma4-31b-nvfp4-mtp1-tp1__baseline__serve-fixed-1k-256-v1__c8__r1__attempt1.json \
  --num-warmups 16 \
  --metadata campaign=runpile-rtx5090-nemotron-gemma4-specdecode-pilot-20260830-v2 config_id=gemma4-31b-nvfp4-mtp1-tp1 server_variant=baseline workload_key=serve-fixed-1k-256-v1 repetition=1 aggregate_only=true