NVIDIA-Nemotron-3-Ultra-550B-A55B

workloads/nemotron_3_ultra_550b_b200.yaml

@@ -0,0 +1,50 @@
+# Nemotron-3-Ultra-550B-A55B (FP8) on H200
+# Hybrid Mamba-2 + LatentMoE architecture (550B total, 55B active).
+name: nemotron_3_ultra-b200
+gpu: B200
+num_gpus: 8
+nightly: true
+
+vllm:
+  model: nvidia/NVIDIA-Nemotron-3-Ultra-550B-A55B-NVFP4
+  serve_args: >-
+    --trust-remote-code
+    --kv-cache-dtype fp8
+    --max-num-seqs 16
+    --max-model-len 32768
+    --gpu-memory-utilization 0.9
+    --enable-flashinfer-autotune
+    --mamba-backend triton
+    --mamba-ssm-cache-dtype float32
+    -cc.pass_config.fuse_allreduce_rms=False
+    --tensor-parallel-size 8
+    --enable-auto-tool-choice
+    --tool-call-parser qwen3_coder
+    --reasoning-parser nemotron_v3
+    --speculative_config.method mtp
+    --speculative_config.num_speculative_tokens 5
+
+lm_eval:
+  model_args:
+    tokenized_requests: false
+    tokenizer_backend: null
+    timeout: 6000
+  tasks:
+    - name: gsm8k
+      num_fewshot: 5
+      model_args:
+        num_concurrent: 64
+        max_length: 32768
+        max_gen_toks: 8192
+
+vllm_bench:
+  configs:
+    - name: 8k-in-1k-out-conc-128
+      backend: openai-chat
+      dataset: speed_bench
+      input_len: 8192
+      output_len: 1024
+      num_prompts: 512
+      max_concurrency: 128
+      speed_bench_dataset_subset: throughput_8k
+      speed_bench_category: low_entropy

workloads/nemotron_3_ultra_550b_h200.yaml

@@ -0,0 +1,50 @@
+# Nemotron-3-Ultra-550B-A55B (FP8) on H200
+# Hybrid Mamba-2 + LatentMoE architecture (550B total, 55B active).
+name: nemotron_3_ultra-h200
+gpu: H200
+num_gpus: 8
+nightly: true
+
+vllm:
+  model: RedHatAI/NVIDIA-Nemotron-3-Ultra-550B-A55B-FP8-dynamic
+  serve_args: >-
+    --trust-remote-code
+    --kv-cache-dtype fp8
+    --max-num-seqs 16
+    --max-model-len 32768
+    --gpu-memory-utilization 0.9
+    --enable-flashinfer-autotune
+    --mamba-backend triton
+    --mamba-ssm-cache-dtype float32
+    -cc.pass_config.fuse_allreduce_rms=False
+    --tensor-parallel-size 8
+    --enable-auto-tool-choice
+    --tool-call-parser qwen3_coder
+    --reasoning-parser nemotron_v3
+    --speculative_config.method mtp
+    --speculative_config.num_speculative_tokens 5
+
+lm_eval:
+  model_args:
+    tokenized_requests: false
+    tokenizer_backend: null
+    timeout: 6000
+  tasks:
+    - name: gsm8k
+      num_fewshot: 5
+      model_args:
+        num_concurrent: 64
+        max_length: 32768
+        max_gen_toks: 8192
+
+vllm_bench:
+  configs:
+    - name: 8k-in-1k-out-conc-128
+      backend: openai-chat
+      dataset: speed_bench
+      input_len: 8192
+      output_len: 1024
+      num_prompts: 512
+      max_concurrency: 128
+      speed_bench_dataset_subset: throughput_8k
+      speed_bench_category: low_entropy