Update model card

Apply BF16 model card updates while preserving base_model metadata.

README.md

@@ -1,16 +1,46 @@
 ---
 base_model:
 - nvidia/Nemotron-3-Nano-Omni-30B-A3B-Reasoning-BF16
+library_name: transformers
 license: other
 license_name: nvidia-open-model-agreement
 license_link: >-
   https://www.nvidia.com/en-us/agreements/enterprise-software/nvidia-open-model-agreement/
+pipeline_tag: any-to-any
 tags:
 - nvidia
+- pytorch
 - multimodal
-pipeline_tag: any-to-any
+datasets:
+- nvidia/Nemotron-Image-Training-v3
+track_downloads: true
 ---
 
+## At a Glance
+
+| | |
+|---|---|
+| **Total parameters** | 31B (Mamba2-Transformer hybrid MoE) |
+| **Active parameters** | ~3B per token |
+| **Max context** | 256k tokens |
+| **Modalities (in)** | Video, Audio, Image, Text |
+| **Modality (out)** | Text |
+| **Reasoning mode** | On by default; toggle via `enable_thinking` |
+| **Best for** | Video+speech analysis, document intelligence (OCR/charts/long docs), GUI/agentic workflows, ASR |
+| **Minimum GPU (BF16)** | 1× H100 80GB (single-GPU); 1× B200 / 1× H200 recommended |
+| **Minimum GPU (FP8)** | 1× L40S 48GB; 1× RTX Pro 6000 / 1× B200 recommended |
+| **Minimum GPU (NVFP4)** | 1× RTX 5090 32GB; 1× DGX Spark / 1× Jetson Thor also supported |
+| **Precisions** | [BF16](https://huggingface.co/nvidia/Nemotron-3-Nano-Omni-30B-A3B-Reasoning-BF16) (62 GB) · [FP8](https://huggingface.co/nvidia/Nemotron-3-Nano-Omni-30B-A3B-Reasoning-FP8) (33 GB) · [NVFP4](https://huggingface.co/nvidia/Nemotron-3-Nano-Omni-30B-A3B-Reasoning-NVFP4) (21 GB) |
+
+## Quick Start Guide
+
+### Model Parameters
+
+| Mode | temperature | top_p | top_k | max_tokens | reasoning_budget | grace_period |
+|------|-------------|-------|-------|------------|------------------|--------------|
+| **Thinking mode** | 0.6 | 0.95 | — | 20480 | 16384 | 1024 |
+| **Instruct mode** | 0.2 | — | 1 | 1024 | — | — |
+
 # Model Overview
 
 ### Description:
@@ -49,10 +79,9 @@ NGC 04/28/2026  via [URL](https://catalog.ngc.nvidia.com/orgs/nim/teams/nvidia/c
 **Architecture Type:** Mamba2-Transformer Hybrid Mixture of Experts (MoE) <br>
 
 **Network Architecture:**
-- [Nemotron 3 Nano LLM (30B A3B)](https://huggingface.co/nvidia/NVIDIA-Nemotron-3-Nano-30B-A3B-BF16)
-- CRADIO v4-H vision encoder
-- Parakeet speech encoder
-
+- [Nemotron 3 Nano LLM (30B A3B)](https://huggingface.co/nvidia/NVIDIA-Nemotron-3-Nano-30B-A3B-BF16) — 31B-parameter Mamba2-Transformer hybrid MoE backbone with ~3B active parameters per token.
+- [CRADIO v4-H](https://huggingface.co/nvidia/C-RADIOv4-H) — vision encoder for image and video frames.
+- [Parakeet](https://huggingface.co/nvidia/parakeet-tdt-0.6b-v2) — speech encoder for audio inputs.
 
 **Number of model parameters:** 3.1 x 10^10 (31B A3B) <br>
 
@@ -130,15 +159,6 @@ Nemotron-3-Nano-Omni-30B-A3B-Reasoning <br>
 
 ---
  
-## Quick Start Guide
-
-### Model Parameters
- 
-| Mode | temperature | top_p | top_k | max_tokens | reasoning_budget | grace_period |
-|------|-------------|-------|-------|------------|------------------|--------------|
-| **Thinking mode** | 0.6 | 0.95 | — | 20480 | 16384 | 1024 |
-| **Instruct mode** | 0.2 | — | 1 | 1024 | — | — |
- 
 ### Download Model Weights
  
 | Precision | Technical Name | HuggingFace URL |
@@ -159,7 +179,7 @@ hf auth login
 hf auth whoami
 ```
 
-#### Download the weights
+<!-- #### Download the weights
  
 Pick a target directory on a volume with ≥70 GB free (the model is ~62 GB).
  
@@ -183,7 +203,7 @@ hf download nvidia/Nemotron-3-Nano-Omni-30B-A3B-Reasoning-BF16 \
 ls "$WEIGHTS" | head
 du -sh "$WEIGHTS"  # expect ~62 GB
 test -f "$WEIGHTS/config.json" && echo OK
-```
+``` -->
  
 ---
  
@@ -225,6 +245,7 @@ vllm serve nvidia/Nemotron-3-Nano-Omni-30B-A3B-Reasoning-NVFP4 \
   --tool-call-parser qwen3_coder \
   --kv-cache-dtype fp8 # Omit this for BF16
 ```
+Efficient Video Sampling: video-pruning-rate=0.5 drops 50% of redundant video tokens; halves video-prefill VRAM/TTFT.
  
 #### Platform-Specific Notes
 
@@ -424,7 +445,7 @@ response = client.chat.completions.create(
         }
     ],
     max_tokens=1024,
-    temperature=1.0,
+    temperature=0.2,
     extra_body={"top_k": 1, "chat_template_kwargs": {"enable_thinking": False}},
 )
 print(response.choices[0].message.content)
@@ -449,7 +470,7 @@ response = client.chat.completions.create(
         }
     ],
     max_tokens=1024,
-    temperature=1.0,
+    temperature=0.2,
     extra_body={"top_k": 1, "chat_template_kwargs": {"enable_thinking": False}},
 )
 print(response.choices[0].message.content)
@@ -496,7 +517,7 @@ print(response.choices[0].message.content)
 ```bash
 curl -sS http://localhost:8000/v1/chat/completions \
   -H "Content-Type: application/json" \
-  -d '{"model":"nvidia/Nemotron-3-Nano-Omni-30B-A3B-Reasoning-NVFP4","messages":[{"role":"user","content":"Hello, what can you do?"}],"temperature":1.0,"top_k":1}' \
+  -d '{"model":"nvidia/Nemotron-3-Nano-Omni-30B-A3B-Reasoning-NVFP4","messages":[{"role":"user","content":"Hello, what can you do?"}],"temperature":0.2,"top_k":1}' \
   | python3 -c "import sys,json; print(json.load(sys.stdin)['choices'][0]['message']['content'])"
 ```
  
@@ -556,7 +577,7 @@ def chat(url, model, b64, text, max_tokens):
         ]}],
         "max_tokens": max_tokens,
         "stream": False,
-        "temperature": 1.0,
+        "temperature": 0.2,
         "chat_template_kwargs": {"enable_thinking": False},
     }, timeout=120)
     r.raise_for_status()
@@ -738,7 +759,6 @@ Higher values improve temporal coverage but increase VRAM and prefill time. Star
 
 ### Notes
 
-
 1. **Reasoning default:** Reasoning is on by default. If you omit `chat_template_kwargs`, the model will produce chain-of-thought traces in `content`. This is appropriate for text and image inputs.
 2. **Video frame sampling:** The default (~32 frames) is too conservative for most real videos. Set `--media-io-kwargs` at server launch.
 3. **PDF input format:** The API does not accept raw PDF uploads. Render pages to PNG and send as base64 (see PDF Example above).
@@ -1019,14 +1039,14 @@ We recommend following settings for reaching the optimal performance.
 ### Sampling Parameters
 We suggest the following sampling parameters based on the mode and tasks.
 * Thinking mode for long document analysis and multimodal reasoning tasks: <br>
- `temperature=0.5-0.7`, `top_p=0.95`, `grace_period=1024`, `reasoning_budget=16384`, `max_token=20480`, and `max_model_len=210000`<br>
+ `temperature=0.6`, `top_p=0.95`, `grace_period=1024`, `reasoning_budget=16384`, `max_token=20480`, and `max_model_len=210000`<br>
 * Instruct mode (non-thinking) for general tasks:<br>
  `temperature=0.2`, `top_k=1`<br>
-* For ASR tasks, we recommend non-thinking mode with
-`temperature=0.2`, `top_k=1`<br>
+* For ASR tasks, we recommend non-thinking mode with <br>
+`temperature=1.0`, `top_k=1`<br>
 
 ### Model output length
-For most multimodel reasoning tasks, we recommend using output length of at least 20480. For complex reasoning questions especially in math and programing increasing the maximum output length to 131072 tokens can give the model enough room to produce more detailed and correct answers. We also found the proposed Budget-Controlled Reasoning effectiveness in answering complex reasoning questions.
+For most multimodel reasoning tasks, we recommend using output length of at least 20480. For complex reasoning questions especially in math and programing increasing the maximum output length to 210000 tokens can give the model enough room to produce more detailed and correct answers. We also found the proposed Budget-Controlled Reasoning effectiveness in answering complex reasoning questions.
 
 ## Ethical Considerations:
 NVIDIA believes Trustworthy AI is a shared responsibility and we have established policies and practices to enable development for a wide array of AI applications. When downloaded or used in accordance with our terms of service, developers should work with their internal model team to ensure this model meets requirements for the relevant industry and use case and addresses unforeseen product misuse. <br>
@@ -1040,12 +1060,12 @@ Please report model quality, risk, security vulnerabilities or NVIDIA AI Concern
 # Citation:
 ```
 @misc{nvidia2026nemotron3nanoomni,
-      title={Nemotron 3 Nano Omni: Efficient and Open Multimodal Intelligence},
-      author={NVIDIA},
-      year={2026},
-      eprint={2604.24954},
-      archivePrefix={arXiv},
-      primaryClass={cs.LG},
-      url={https://arxiv.org/abs/2604.24954},
+      title={Nemotron 3 Nano Omni: Efficient and Open Multimodal Intelligence},
+      author={NVIDIA},
+      year={2026},
+      eprint={2604.24954},
+      archivePrefix={arXiv},
+      primaryClass={cs.LG},
+      url={https://arxiv.org/abs/2604.24954},
 }
 ```