Gemma is a family of open models built by Google DeepMind. Gemma 4 models are multimodal, handling text and image input (with audio supported on E2B, E4B, and 12B) and generating text output. This release includes open-weights models in both pre-trained and instruction-tuned variants.
InternLumina-U2
internlm·Instruct·released 1 Sept 2026
hf download internlm/InternLumina-U2A Multi-Codebook Diffusion Large Language Model for Omni-Visual Understanding, Image Generation and Editing.
How much VRAM does InternLumina-U2 need?
At Q4_K_M InternLumina-U2 needs about 11 GB with an 8K context (11 GB at Q4, 33 GB at full precision). Pick your GPU to see what fits and how fast it runs.
Needs: ≈11 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 | ≈7.1 GB |
| Q3_K_MNoticeable lossSmall, some quality loss | ≈9.6 GB |
| Q4_K_MrecommendedSmall lossRecommended default for local use | ≈11 GB |
| Q5_K_MNear-losslessNear-lossless for most tasks | ≈13 GB |
| Q6_KNear-losslessPractically lossless | ≈15 GB |
| Q8_0LosslessLossless in practice, twice the size of Q4 | ≈19 GB |
| FP16 / BF16LosslessFull precision, as released | ≈33 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 ≈11 GB(9.0 GB weights + 1.4 GB KV cache + 0.9 GB overhead). Pick a GPU above to see how much context fits.
Weights + overheadKV cache
Can the Tesla T4 run InternLumina-U2? →
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 InternLumina-U2
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 InternLumina-U2 locally
About InternLumina-U2
A Multi-Codebook Diffusion Large Language Model for Omni-Visual Understanding, Image Generation and Editing. 📑 Paper · 🌐 Project Page · 💻 Code Intern Lumina U2 is a unified multimodal model that brings language, image, video and 3D into a single framework, covering text QA, text-to-image generation, image understanding, image editing, video understanding and 3D understanding with one model. It is a 16B-parameter MoE with 1B active parameters (16B-A1B), pairing an efficient sparse backbone with an 8-codebook fully-discrete visual representation built on AToken. This repo hosts checkpoints of the same architecture trained on different hardware stacks: • ascend/ — trained on Huawei Ascend NPUs. • nvidia/ — trained on NVIDIA GPUs.
Full model card on Hugging Face →Alternatives to InternLumina-U2
All alternatives →Gemma is a family of open models built by Google DeepMind. Gemma 4 models are multimodal, handling text and image input (with audio supported on E2B, E4B, and 12B) and generating text output. This release includes open-weights models in both pre-trained and instruction-tuned variants.
SenseNova-U1.5-8B-MoT is our latest native unified multimodal checkpoint for more accurate, consistent, reliable, and aesthetically compelling visual creation.
Gemma is a family of open models built by Google DeepMind. Gemma 4 models are multimodal, handling text and image input (with audio supported on E2B, E4B, and 12B) and generating text output. This release includes open-weights models in both pre-trained and instruction-tuned variants.
LLaDA2.0-Uni: Unifying Multimodal Understanding and Generation with Diffusion Large Language Model
NLD-Uni is a unified multimodal model for image understanding and image generation. It combines a diffusion language model with specialized visual components and a mixture-of-transformers design.
InternLumina-U2 FAQ
- How much VRAM does InternLumina-U2 need?
- About 11 GB at Q4_K_M with an 8K context. Q8_0 needs about 19 GB and full precision about 33 GB. Longer contexts need more memory for the KV cache.
- Can InternLumina-U2 run on the Tesla T4?
- Yes. It fits at Q6_K (about 15 GB), at an estimated 13 tokens/s. It fits comfortably on the Arc A770 16GB, Radeon RX 7800 XT, Tesla P40.
- Can I use InternLumina-U2 commercially?
- Yes. It is released under Apache 2.0. Permissive: use, modify and sell freely; keep the notice and licence text.
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