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.
Raon-1-nano
KRAFTON·Instruct·released 29 Oct 2026
ollama run hf.co/KRAFTON/Raon-1-nano-gguf:Q4_K_MRaon-1 nano is a dense language model with 1.5B effective parameters (3.1B total, including a 1.5B per-layer embedding table), developed by KRAFTON AI. It is designed to follow instructions, maintain conversational context, and call the right tools in both Korean and English.
How much VRAM does Raon-1-nano need?
At Q4_K_M Raon-1-nano needs about 5.6 GB with an 8K context (5.6 GB at Q4, 10 GB at full precision). Pick your GPU to see what fits and how fast it runs.
Needs: ≈5.5 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 | 8K | 16K | 32K | 64K | 128K | 256K |
|---|---|---|---|---|---|---|
| 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 | ≈4.6 GB |
| Q3_K_MNoticeable lossSmall, some quality loss | ≈5.2 GB |
| Q4_K_MrecommendedSmall lossRecommended default for local use | ≈5.5 GB |
| Q5_K_MNear-losslessNear-lossless for most tasks | ≈6.0 GB |
| Q6_KNear-losslessPractically lossless | ≈6.1 GB |
| Q8_0LosslessLossless in practice, twice the size of Q4 | ≈6.8 GB |
| FP16 / BF16LosslessFull precision, as released | ≈9.6 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. Weights marked “file” use the real GGUF download size; “est.” rows are computed from the parameter count.
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 ≈5.5 GB(1.9 GB weights + 3.0 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 Raon-1-nano? →
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 Raon-1-nano
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 Raon-1-nano locally
--quantQ4_K_Mdefault; pick your GPU to tailor itKRAFTON/Raon-1-nano-gguf
ollama run hf.co/KRAFTON/Raon-1-nano-gguf:Q4_K_M llama-server -hf KRAFTON/Raon-1-nano-gguf:Q4_K_MDownloads the GGUF on first run. Installed through llama.app? The same flags work with `llama serve`.
lms get KRAFTON/Raon-1-nano-gguf@Q4_K_MThen load it from the LM Studio app or with `lms load`.
pip install mlx-lm
mlx_lm.generate --model KRAFTON/Raon-1-nano-mlx --prompt "Hello"Use `mlx_lm.chat` with the same --model for an interactive chat.
mlx_lm.server --model KRAFTON/Raon-1-nano-mlx --port 8080API on http://localhost:8080/v1.
from transformers import pipeline
pipe = pipeline("text-generation", model="KRAFTON/Raon-1-nano", 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"])Quantised versions and downloads
| Repository | Format | Quant | Size |
|---|---|---|---|
| KRAFTON/Raon-1-nano-gguf | GGUF | Q4_K_M | 2.1 GB |
| webmp3/Sakura-Raon-1-nano-GGUF | GGUF | Q4_K_M | 2.1 GB |
| webmp3/Sakura-Raon-1-nano-GGUF | GGUF | Q5_K_M | 2.6 GB |
| KRAFTON/Raon-1-nano-gguf | GGUF | Q6_K | 2.6 GB |
| webmp3/Sakura-Raon-1-nano-GGUF | GGUF | Q6_K | 2.6 GB |
| KRAFTON/Raon-1-nano-gguf | GGUF | Q6_K_MTP | 2.7 GB |
| KRAFTON/Raon-1-nano-gguf | GGUF | Q8_0 | 3.4 GB |
| webmp3/Sakura-Raon-1-nano-GGUF | GGUF | Q8_0 | 3.4 GB |
| KRAFTON/Raon-1-nano-gguf | GGUF | Q8_0_MTP | 3.5 GB |
| KRAFTON/Raon-1-nano-gguf | GGUF | F16 | 6.3 GB |
About Raon-1-nano
Raon-1 nano is a dense language model with 1.5B effective parameters (3.1B total, including a 1.5B per-layer embedding table), developed by KRAFTON AI. It is designed to follow instructions, maintain conversational context, and call the right tools in both Korean and English. As part of the Raon open model family, it is released under the Apache 2.0 license. • Faithful, natural conversation: Leads the comparison models on the instruction-following, context-following, and Korean conversation benchmarks shown below. • Korean and English: Trained on Korean and English, with solid English capabilities alongside its Korean strengths. • Tool calling: Tool calling within conversations, for everyday tasks such as office work and mobile control. • On-device deployment: Available in GGUF, MLX, and LiteRT-LM formats for local inference, including on Apple silicon. • Long context: Natively supports up to 262,144 tokens of context. Raon-1 nano uses a hybrid dense architecture, consisting of 18 linear attention layers and 6 full attention layers.
Full model card on Hugging Face →Alternatives to Raon-1-nano
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 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.
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.
Qwen3 is the latest generation of large language models in Qwen series, offering a comprehensive suite of dense and mixture-of-experts (MoE) models.
Raon-1-nano FAQ
- How much VRAM does Raon-1-nano need?
- About 5.6 GB at Q4_K_M with an 8K context. Q8_0 needs about 7.2 GB and full precision about 10 GB. Longer contexts need more memory for the KV cache.
- Can Raon-1-nano run on the RTX 4060?
- Yes. It fits at Q8_0 (about 6.8 GB), at an estimated 40 tokens/s. It fits comfortably on the Apple M4 (16 GB), Radeon RX 6700 XT, RTX 3060 12GB.
- Can I use Raon-1-nano 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 Raon-1-nano?
- 262,144 tokens (256K). The full window needs far more memory than an 8K context.
- How do I run Raon-1-nano with Ollama or llama.cpp?
- With Ollama: ollama run hf.co/KRAFTON/Raon-1-nano-gguf:Q4_K_M. With llama.cpp: llama-server -hf KRAFTON/Raon-1-nano-gguf:Q4_K_M, which starts an OpenAI-compatible server.
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