NVIDIA-Nemotron-3.5-Bolt-74B-A9B-BF16: Ask about images and video
Added · Updated
Made by NVIDIA
View profile on Hugging Face (opens in a new tab)Send text, images or video and get written answers.
For teams with NVIDIA A100, H100 or Blackwell GPUs who want to customize or post-train the model; NVIDIA points others to its 24GB versions.
Model details
Original model nvidia/NVIDIA-Nemotron-3.5-Bolt-74B-A9B-BF16
Input Image or text Output Text
Ways to use it Run locally
One reviewed local configuration. Add My computer details to compare them.
BF16 vLLM on NVIDIA · vLLM · The pinned card documents vLLM on NVIDIA Ampere, Hopper, or Blackwell GPUs. The 192 GB VRAM threshold is a BuildTube estimate for the BF16 checkpoint at 4K.
How this configuration was checked
192 GB VRAM · reviewed estimate Source lists 149.8 GB (file-size). Estimate uses 149.78 GB exact BF16 checkpoint bytes, 0.034 GB 4K KV cache (8 attention blocks of 40 Mamba blocks plus attention blocks, from the config layer map × 2 KV heads × 128 head width × K/V × 2-byte values), 4 GB runtime buffers, and 38.186 GB remaining in the 192 GB VRAM tier.
BF16 · 4,096 token context · source context 262,144 tokens · batch size 1 · concurrency 1 · Single-user 4K comparison; the card supports up to 256K context and recommends deployment settings based on GPU memory.
Source checked 2026-10-21 · revision 567021037ff88364143e24ea05e6abd0f72086ff. pinned card specifies BF16 model, vLLM, supported NVIDIA GPU generations, and 256K context; exact shard bytes and attention-block config supply estimate inputs
Role in this family: Fine-tuned version.
nvidia/NVIDIA-Nemotron-3.5-Super-VL-120B-A12B-BF16 (opens in a new tab)What it does
Nemotron 3.5 Bolt is a vision language model, meaning an AI that reads text, images and video and replies in text. NVIDIA designed it to run everyday agent tasks locally. This is the full-precision BF16 release, intended mainly for customization and post-training (further training). A separate 24GB version uses 2-bit quantization, a compression method, to fit in 24GB of memory.
How to use it
Input → action → output: load the model with vLLM on a supported NVIDIA GPU, send text, an image or a video, and read the text reply. The chat template lets you turn thinking mode on or off with enable thinking. For faster everyday inference, the source points to the 24GB versions instead.
Common use cases
- Send an image with a question and read a written answer about it.
- Send a video with a question and read a text reply.
Can I use this?
Requirements The model is not gated and is governed by the OpenMDW 1.1 license; the source says it is ready for commercial or non-commercial use. It requires a recent vLLM version, but the exact version is not given in the model card. Deployment is listed on NVIDIA Ampere (A100), Hopper (H100 80GB) or Blackwell GPUs. Memory needs for the BF16 version are not documented.
Caveat NVIDIA recommends the 24GB versions rather than this one for fast inference. Other license terms are not summarized here, and benchmark tables are the creator's own results.
Worth knowing This full-precision version is intended mainly for customization and post-training, not direct production inference. The OpenMDW 1.1 terms are not summarized here. Licensed under OpenMDW 1.1.
Why it's here
Not trending this week
Last seen on the Hugging Face trending list on 2 November 2026. It stays on BuildTube so you can still find it.
- 52 people have liked it on Hugging Face.
- It was downloaded 8,753 times in the last 30 days.
Numbers checked on 8 November 2026; not refreshed since.
Behind it
View model pageBuildTube has not run or verified this project. Everything above is written from what the creator published.
Made this?
Quick questions
- What is NVIDIA-Nemotron-3.5-Bolt-74B-A9B-BF16?
- Nemotron 3.5 Bolt is a vision language model, meaning an AI that reads text, images and video and replies in text.
- What can I use NVIDIA-Nemotron-3.5-Bolt-74B-A9B-BF16 for?
- Send an image with a question and read a written answer about it; Send a video with a question and read a text reply
- Can I run NVIDIA-Nemotron-3.5-Bolt-74B-A9B-BF16 on my computer?
- It runs on your own hardware with a model runtime. Deployment is listed on NVIDIA Ampere (A100), Hopper (H100 80GB) or Blackwell GPUs. NVIDIA recommends the 24GB versions rather than this one for fast inference. Memory needs for the BF16 version are not documented.
- What do I need to use NVIDIA-Nemotron-3.5-Bolt-74B-A9B-BF16?
- OpenMDW 1.1 license and not gated. Deployment is listed with vLLM (a recent enough version) on NVIDIA Ampere (A100), Hopper (H100 80GB) or Blackwell GPUs.
- What license does NVIDIA-Nemotron-3.5-Bolt-74B-A9B-BF16 use?
- Licensed under OpenMDW 1.1. Other license terms are not summarized here, and benchmark tables are the creator's own results.
More from this week
- NVKE Run managed Kubernetes clusters on your own NVIDIA GPU machines
Set up a service that creates and manages separate Kubernetes clusters on your own NVIDIA GPU servers.
- P-H3-Image-Comfy Make and edit 8-megapixel images in ComfyUI
Turn short text requests into 8-megapixel images and edit pictures on your own graphics card.
- Idol Turn one picture of an object into a 3D Gaussian splat
Idol turns one photo, render or generated image of an object into a 3D Gaussian splat file.
Explore by topic
Post this
Download share imageFor makers
Made this? Show it’s on BuildTube — add the badge to your README or embed the card on your site.
README badge (Markdown)
[](https://buildtube.app/p/hf-nvidia-nvidia-nemotron-3-5-bolt-74b-a9b-bf16/)Embed card (HTML)
<iframe src="https://buildtube.app/embed/hf-nvidia-nvidia-nemotron-3-5-bolt-74b-a9b-bf16/" width="560" height="240" style="border:0;max-width:100%" loading="lazy" title="NVIDIA-Nemotron-3.5-Bolt-74B-A9B-BF16 on BuildTube"></iframe>