NVIDIA-Nemotron-3-Nano-4B-BF16 is a small language model (SLM) trained from scratch by NVIDIA, and designed as a unified model for both reasoning and non-reasoning tasks. It responds to user queries and tasks by first generating a reasoning trace and then concluding with a final response.
granite-4.1-3b-llmon-preview
ibm-granite·Instruct·released 1 Oct 2026
vllm serve ibm-granite/granite-4.1-3b-llmon-previewModel Summary: Granite-4.1-3B-LLMON-preview is a 3B parameter instruct model fine-tuned from Granite-4.1-3B-base to consume prompts in LLMON (LLM Object Notation) instead of a conventional chat template.
How much VRAM does granite-4.1-3b-llmon-preview need?
At Q4_K_M granite-4.1-3b-llmon-preview needs about 3.3 GB with an 8K context (3.3 GB at Q4, 8.3 GB at full precision). Pick your GPU to see what fits and how fast it runs.
Needs: ≈3.3 GB · Q4_K_M · 8K context — Pick a GPU to check the fit.
Advanced: context 8K · KV cache 16-bit
Smallest GPU for each quant and context (KV cache 16-bit)
| Quantisation | 4K | 8K | 16K | 32K | 64K | 128K |
|---|---|---|---|---|---|---|
| Q3_K_M | ||||||
| Q4_K_M | ||||||
| Q5_K_M | ||||||
| Q6_K | ||||||
| Q8_0 | ||||||
| FP16 |
≤16 GB card24–32 GB48–96 GBmore
| Quantisation | Total |
|---|---|
| Q2_KHeavy lossSmallest, noticeably worse quality | ≈2.3 GB |
| Q3_K_MNoticeable lossSmall, some quality loss | ≈2.9 GB |
| Q4_K_MrecommendedSmall lossRecommended default for local use | ≈3.3 GB |
| Q5_K_MNear-losslessNear-lossless for most tasks | ≈3.7 GB |
| Q6_KNear-losslessPractically lossless | ≈4.1 GB |
| Q8_0LosslessLossless in practice, twice the size of Q4 | ≈4.9 GB |
| FP16 / BF16LosslessFull precision, as released | ≈8.3 GB |
Estimates for 8K tokens of context: weights + KV cache + runtime overhead. Speed assumes memory-bound decoding at your GPU’s bandwidth. Pick a GPU above and the fit column checks every row against it.
How to pick a quant
Take the biggest quant that fits with room left for context. Q4_K_M is the usual sweet spot. Below Q4 quality drops fast, above Q6 you mostly pay memory for nothing. Big models (70B+, large MoE) tolerate low quants better than small ones.
- LosslessF16, Q8_0
- Near-losslessQ6_K, Q5_K_M
- Small lossQ5_K_S, Q4_K_M, IQ4_NL, IQ4_XS, Q4_K_S
- Noticeable lossQ4_0, Q3_K_L, Q3_K_M, IQ3_M
- Heavy lossIQ3_XXS, Q3_K_S, IQ2_M, Q2_K
Typical quality loss, not measured per model. Tiers come from KL divergence against FP16 on Llama 3 8B: llama.cpp perplexity README: LLaMA 3 8B scoreboard (KLD and ΔPPL vs FP16, Wikitext-2); llama.cpp quantize tool: ΔPPL per quant type @ Llama-3-8B
How much context fits?
At 8K context, Q4_K_M needs ≈3.3 GB(2.1 GB weights + 0.6 GB KV cache + 0.6 GB overhead). Pick a GPU above to see how much context fits.
Weights + overheadKV cache
Can the RTX 4060 run granite-4.1-3b-llmon-preview? →
Speed vs quality per quant
Smaller quants run faster and fit in less memory, but lose quality.
Fits in VRAMTight fitCPU offload✕ won't runQuality is typical KL divergence vs FP16, not measured for this model.
Real-world speed of granite-4.1-3b-llmon-preview
Tokens per second measured by people running it at home. A GPU shows a median once 3 reports agree on the same setup. Reports far from our estimate wait for a review. Measure yours with llama-bench →
No medians yet. Be one of the first 3 to report a setup.
All reports
Run granite-4.1-3b-llmon-preview locally
These commands load the original weights from Hugging Face.
vllm serve ibm-granite/granite-4.1-3b-llmon-previewservices:
vllm:
image: vllm/vllm-openai:latest
command: ibm-granite/granite-4.1-3b-llmon-preview
ports:
- "8000:8000"
ipc: host
volumes:
- ~/.cache/huggingface:/root/.cache/huggingface
deploy:
resources:
reservations:
devices:
- driver: nvidia
count: all
capabilities: [gpu]API on port 8000.
sglang serve ibm-granite/granite-4.1-3b-llmon-preview --host 0.0.0.0 --port 30000services:
sglang:
image: lmsysorg/sglang:latest
command: sglang serve ibm-granite/granite-4.1-3b-llmon-preview --host 0.0.0.0 --port 30000
ports:
- "30000:30000"
ipc: host
volumes:
- ~/.cache/huggingface:/root/.cache/huggingface
deploy:
resources:
reservations:
devices:
- driver: nvidia
count: all
capabilities: [gpu]API on port 30000.
from transformers import pipeline
pipe = pipeline("text-generation", model="ibm-granite/granite-4.1-3b-llmon-preview", dtype="auto", device_map="auto")
messages = [{"role": "user", "content": "Explain self-hosting in one paragraph."}]
out = pipe(messages, max_new_tokens=512)
print(out[0]["generated_text"][-1]["content"])About granite-4.1-3b-llmon-preview
Model Summary: Granite-4.1-3B-LLMON-preview is a 3B parameter instruct model fine-tuned from Granite-4.1-3B-base to consume prompts in LLMON (LLM Object Notation) instead of a conventional chat template. LLMON wraps every part of a prompt (system policy, instructions, data, documents, tool definitions, and tool results) in an individually addressable block and adds an exec block that names which instruction to run, which inputs it operates on, and where the output goes. • Developers: Granite Team, IBM • HF Collection: Granite Experiments HF Collection • Other sizes: Granite-4.1-8B-LLMON-preview • Technical Blog: LLMON: An LLM-Native Markup Language, now in Granite 4.1 3B and 8B • GitHub Repository: generative-computing/llmon • Visualization Demo: LLMON on Hugging Face Spaces • Release Date: November 12, 2026 • License: Apache 2.0 Supported Languages: English, German, Spanish, French, Japanese, Portuguese, Arabic, Czech, Italian, Korean, Dutch, and Chinese. Users may fine-tune the model for languages beyond these. Intended use: The model is designed to follow general instructions and can serve as the foundation for AI assistants across diverse domains, including business applications, as well as for LLM agents equipped with tool-use capabilities. It is particularly suited to settings where a prompt mixes trusted instructions with untrusted material (retrieved documents, pasted content, or tool results) and the boundary between the two must be explicit.
Full model card on Hugging Face →Other Granite 4.1 Llmon models
Whole family →Alternatives to granite-4.1-3b-llmon-preview
All alternatives →Model Summary: Granite-4.1-3B is a 3B parameter long-context instruct model finetuned from Granite-4.1-3B-Base using a combination of open source instruction datasets with permissive license and internally collected synthetic datasets.
We introduce the updated version of the Qwen3-4B non-thinking mode, named Qwen3-4B-Instruct-2507, featuring the following key enhancements:
Granite is a family of open-source large language models developed by IBM, designed for enterprise and research use. Granite models are built to be versatile, safe, and efficient — covering a range of sizes and capabilities from compact edge-deployable models to large-scale reasoning systems.
We are releasing MiniCPM5-2B, the second model in the MiniCPM5 series, following MiniCPM5-1B. It is a dense 2B Transformer that scales up the same training recipe, built for on-device, local deployment, and resource-constrained scenarios, reaching 2B-class open-source SOTA.
Qwen3 is the latest generation of large language models in Qwen series, offering a comprehensive suite of dense and mixture-of-experts (MoE) models.
granite-4.1-3b-llmon-preview FAQ
- How much VRAM does granite-4.1-3b-llmon-preview need?
- About 3.3 GB at Q4_K_M with an 8K context. Q8_0 needs about 4.9 GB and full precision about 8.3 GB. Longer contexts need more memory for the KV cache.
- Can granite-4.1-3b-llmon-preview run on the RTX 4060?
- Yes. It fits at Q8_0 (about 4.9 GB), at an estimated 38 tokens/s. It fits comfortably on the Apple M4 (16 GB), Radeon RX 6700 XT, RTX 3060 12GB.
- Can I use granite-4.1-3b-llmon-preview commercially?
- Yes. It is released under Apache 2.0. Permissive: use, modify and sell freely; keep the notice and licence text.
- What is the context length of granite-4.1-3b-llmon-preview?
- 131,072 tokens (128K). The full window needs far more memory than an 8K context.
- How do I run granite-4.1-3b-llmon-preview locally?
- With vLLM: vllm serve ibm-granite/granite-4.1-3b-llmon-preview.
Rate this model
Data from Hugging Face, the model card and public leaderboards. Memory and speed figures are estimates; real numbers depend on the runtime, context length and drivers. Always check the licence yourself before commercial use.
0 comments