From 6336ca054fa33b64247f9199fd37e7119c5a3785 Mon Sep 17 00:00:00 2001
From: Tarun Kumar <takumar@redhat.com>
Date: Mon, 29 Jun 2026 16:43:55 +0530
Subject: [PATCH] Add Gemma4 and Nemotron-ultra model to H200

Signed-off-by: Tarun Kumar <takumar@redhat.com>
---
 workloads/gemma_4_31b_it_h200.yaml     | 46 ++++++++++++++++++++++++++
 workloads/nemotron_3_super_5_h200.yaml | 14 ++++++++
 2 files changed, 60 insertions(+)
 create mode 100644 workloads/gemma_4_31b_it_h200.yaml

diff --git a/workloads/gemma_4_31b_it_h200.yaml b/workloads/gemma_4_31b_it_h200.yaml
new file mode 100644
index 0000000..5a931aa
--- /dev/null
+++ b/workloads/gemma_4_31b_it_h200.yaml
@@ -0,0 +1,46 @@
+# Gemma4 31B (FP8) on H200 (8xH200, TP=1)
+name: gemma_4_31b_it_h200
+gpu: H200
+num_gpus: 8
+nightly: true
+
+vllm:
+  model: RedHatAI/gemma-4-31B-it-FP8-dynamic
+  serve_args: >-
+    --tensor-parallel-size 1
+    --kv-cache-dtype fp8
+    --tool-call-parser gemma4
+    --reasoning-parser gemma4
+    --enable-auto-tool-choice
+
+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
+
+bfcl:
+  test_categories:
+    - simple_python
+    - multiple
+    - parallel
+    - parallel_multiple
+    - multi_turn
+  num_threads: 16
+
+vllm_bench:
+  configs:
+    - name: 8k-in-1k-out-conc-128
+      backend: openai
+      dataset: random
+      input_len: 8192
+      output_len: 1024
+      num_prompts: 512
+      max_concurrency: 128
diff --git a/workloads/nemotron_3_super_5_h200.yaml b/workloads/nemotron_3_super_5_h200.yaml
index ff76e83..5633585 100644
--- a/workloads/nemotron_3_super_5_h200.yaml
+++ b/workloads/nemotron_3_super_5_h200.yaml
@@ -15,6 +15,9 @@ vllm:
     --speculative-config.method mtp
     --speculative-config.num_speculative_tokens 5
     --trust-remote-code
+    --enable-auto-tool-choice
+    --tool-call-parser qwen3_coder
+    --reasoning-parser nemotron_v3
 
 lm_eval:
   model_args:
@@ -29,6 +32,17 @@ lm_eval:
         max_length: 32768
         max_gen_toks: 8192
 
+bfcl:
+  test_categories:
+    - simple_python
+    - multiple
+    - parallel
+    - parallel_multiple
+    - multi_turn
+  num_threads: 16
+  max_test_cases:
+    multi_turn: 240
+
 vllm_bench:
   configs:
     - name: 8k-in-1k-out-conc-128