From 8fec9bed066cd89999a8b7f08931ccfe7ce50896 Mon Sep 17 00:00:00 2001
From: Eugene Rakhmatulin <eugr@eugr.com>
Date: Thu, 12 Mar 2026 13:30:15 -0700
Subject: [PATCH] Updated Nemotron to support dual sparks

---
 README.md                           | 10 ++++++++++
 recipes/nemotron-3-super-nvfp4.yaml |  9 +++++----
 2 files changed, 15 insertions(+), 4 deletions(-)

diff --git a/README.md b/README.md
index c3f1e52..da32fdf 100644
--- a/README.md
+++ b/README.md
@@ -149,6 +149,16 @@ Don't do it every time you rebuild, because it will slow down compilation times.
 
 For periodic maintenance, I recommend using a filter: `docker builder prune --filter until=72h`
 
+### 2026-03-12
+
+#### Nemotron-3-Super-120B NVFP4 Recipe
+
+Added a new recipe `nemotron-3-super-nvfp4` for running `nvidia/NVIDIA-Nemotron-3-Super-120B-A12B-NVFP4` with Marlin kernels. Supports both solo and cluster modes. Includes a custom reasoning parser (`super_v3_reasoning_parser.py`) fetched from the model repository. Supports both dual and single Spark configurations.
+
+```bash
+./run-recipe.sh nemotron-3-super-nvfp4
+```
+
 ### 2026-03-11
 
 #### Qwen3-Coder-Next INT4-AutoRound Recipe
diff --git a/recipes/nemotron-3-super-nvfp4.yaml b/recipes/nemotron-3-super-nvfp4.yaml
index 48120fe..a4de32d 100644
--- a/recipes/nemotron-3-super-nvfp4.yaml
+++ b/recipes/nemotron-3-super-nvfp4.yaml
@@ -7,8 +7,7 @@ description: vLLM serving Nemotron-3-Super-120B using Marlin kernels
 model: nvidia/NVIDIA-Nemotron-3-Super-120B-A12B-NVFP4
 container: vllm-node
 cluster_only: false
-# This model can only run on single node (solo)
-solo_only: true
+solo_only: false
 
 mods: 
   - mods/nemotron-super
@@ -17,7 +16,7 @@ container: vllm-node
 defaults:
   port: 8000
   host: 0.0.0.0
-  tensor_parallel: 1
+  tensor_parallel: 2
   gpu_memory_utilization: 0.7
   max_model_len: 262144
   max_num_seqs: 10
@@ -41,4 +40,6 @@ command: |
   --load-format fastsafetensors \
   --tool-call-parser qwen3_coder \
   --reasoning-parser-plugin super_v3_reasoning_parser.py \
-  --reasoning-parser super_v3
\ No newline at end of file
+  --reasoning-parser super_v3 \
+  --tensor-parallel-size {tensor_parallel} \
+  --distributed-executor-backend ray
\ No newline at end of file