selfhost.directory

The index of self-hostable projects — ranked, categorized and served with live updates.
Own your data and stop giving it to big tech.

inclusionAI logo

ZwZ2-27B

inclusionAI·Instruct·released 16 Oct 2026

Unknown licence27.4B256K ctx ≈19 GB
Your GPU
llama-server -hf inclusionAI/ZwZ2-27B-GGUF:Q4_K_M
All 4 runners ↓

ZwZ2-27B is the Qwen3.6-27B-based model in Zooming-without-Zooming-v2 (ZwZ2). It learns to connect visual evidence across regions through comparison, aggregation, matching, and multi-step reasoning. Training uses full images; inference requires no zooming tools.

Notify me when

How much VRAM does ZwZ2-27B need?

At Q4_K_M ZwZ2-27B needs about 19 GB with an 8K context (19 GB at Q4, 56 GB at full precision). Pick your GPU to see what fits and how fast it runs.

--vram
any
--gpus
--quant

Needs: ≈19 GB · Q4_K_M · 8K context — Pick a GPU to check the fit.

Change quant, context or KV cache to compare
Advanced: context 8K · KV cache 16-bit
Context
KV cache

Smallest GPU for each quant and context (KV cache 16-bit)

Quantisation8K16K32K64K128K256K
Q3_K_M
Q4_K_M
Q5_K_M
Q6_K
Q8_0
FP16

≤16 GB card24–32 GB48–96 GBmore

QuantisationQualityWeightsKV cacheTotal
Q2_KHeavy lossSmallest, noticeably worse qualityHeavy loss8.4 GB est.2.0 GB≈11 GB
Q3_K_MNoticeable lossSmall, some quality lossNoticeable loss12 GB est.2.0 GB≈16 GB
Q4_K_MrecommendedSmall lossRecommended default for local useSmall loss15 GB file2.0 GB≈19 GB
Q5_K_MNear-losslessNear-lossless for most tasksNear-lossless18 GB file2.0 GB≈21 GB
Q6_KNear-losslessPractically losslessNear-lossless21 GB file2.0 GB≈24 GB
Q8_0LosslessLossless in practice, twice the size of Q4Lossless27 GB file2.0 GB≈30 GB
FP16 / BF16LosslessFull precision, as releasedLossless50 GB file2.0 GB≈55 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?

Model quant
KV cache

At 8K context, Q4_K_M needs ≈19 GB(15 GB weights + 2.0 GB KV cache + 1.3 GB overhead). Pick a GPU above to see how much context fits.

0204060801001K2K4K8K16K32K64K128K256KGB

Weights + overheadKV cache

Can the Tesla P40 run ZwZ2-27B? →

Speed vs quality per quant

Smaller quants run faster and fit in less memory, but lose quality.

Q2_KQ3_K_MQ4_K_MQ5_K_MQ6_KQ8_0F16

Fits in VRAMTight fitCPU offload✕ won't runQuality is typical KL divergence vs FP16, not measured for this model.

Real-world speed of ZwZ2-27B

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 ZwZ2-27B locally

--quantQ4_K_Mdefault; pick your GPU to tailor itinclusionAI/ZwZ2-27B-GGUF

OpenAI-compatible server with a web UIdocs
llama-server -hf inclusionAI/ZwZ2-27B-GGUF:Q4_K_M

Downloads the GGUF on first run. Installed through llama.app? The same flags work with `llama serve`.

Install llama.cpp

Serve an OpenAI-compatible APIdocs
vllm serve inclusionAI/ZwZ2-27B
Docker Compose (NVIDIA)docs
yaml
services:
  vllm:
    image: vllm/vllm-openai:latest
    command: inclusionAI/ZwZ2-27B
    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.

Install vLLM

Serve an OpenAI-compatible APIdocs
sglang serve inclusionAI/ZwZ2-27B --host 0.0.0.0 --port 30000
Docker Compose (NVIDIA)docs
yaml
services:
  sglang:
    image: lmsysorg/sglang:latest
    command: sglang serve inclusionAI/ZwZ2-27B --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.

Install SGLang

Ask about an imagedocs
python
from transformers import pipeline

pipe = pipeline("image-text-to-text", model="inclusionAI/ZwZ2-27B", dtype="auto", device_map="auto")
messages = [
    {
        "role": "user",
        "content": [
            {"type": "image", "url": "https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/bee.jpg"},
            {"type": "text", "text": "Describe this image."},
        ],
    }
]
out = pipe(text=messages, max_new_tokens=256)
print(out[0]["generated_text"][-1]["content"])

Install Transformers

Quantised versions and downloads

About ZwZ2-27B

ZwZ2-27B is the Qwen3.6-27B-based model in Zooming-without-Zooming-v2 (ZwZ2). It learns to connect visual evidence across regions through comparison, aggregation, matching, and multi-step reasoning. Training uses full images; inference requires no zooming tools. On 25 benchmarks across six domains, it improves over its base model on every benchmark, increasing overall accuracy from 55.52 to 60.51 (+4.99 points). Gains extend from fine-grained perception to spatial reasoning, long-context understanding, multimodal agents, and general visual reasoning. EvidenceConnect builds verified multi-region questions from SA-1B images: • Group 3–7 related regions and generate questions using privileged zoomed views. • Check answers through cross-model agreement and drop-one region tests. • Convert verified questions to four-option tasks and train with GRPO, without an additional supervised fine-tuning stage. All three model sizes use approximately 20K training examples. The policy sees full images with colored region outlines; privileged crop layouts are used only for data construction. See ZwZ2-RL-VQA for the data schema and synthesis process. Editable vector PDF. The figure shows eight representative benchmarks; the complete results follow.

Full model card on Hugging Face →

Other ZwZ2 models

Whole family →
inclusionAI logo
ZwZ2-4BinclusionAI
4.5BCheck licence≈4.2 GB↓ 1.1K
inclusionAI logo
ZwZ2-9BinclusionAI
9.4BCheck licence≈7.1 GB↓ 1.5K

Alternatives to ZwZ2-27B

All alternatives →
Qwen logo

Qwen3.8-27B

Qwen·Vision

≈19 GB

Following the widespread community adoption of the Qwen3.5 and Qwen3.6 series, we are pleased to introduce Qwen3.8, the most capable generation in the Qwen open-model family to date.

27.8B256K ctx Apache 2.0
6.3MAug 2026ollama
google logo

gemma-4-26B-A4B-it

google·Vision

≈18 GB

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.

25.8BA4B256K ctx Apache 2.0
4.8MMar 2026ollama
Qwen logo

Qwen3.6-27B

Qwen·Vision

≈19 GB

Following the February release of the Qwen3.5 series, we're pleased to share the first open-weight variant of Qwen3.6. Built on direct feedback from the community, Qwen3.6 prioritizes stability and real-world utility, offering developers a more intuitive, responsive, and genuinely productive coding…

27.8B256K ctx Apache 2.0
2.4MApr 2026ollama
google logo

gemma-4-31B-it

google·Vision

≈27 GB

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.

31.3B256K ctx Apache 2.0
6MMar 2026ollama
Qwen logo

Qwen3.5-27B

Qwen·Vision

≈19 GB

Over recent months, we have intensified our focus on developing foundation models that deliver exceptional utility and performance.

27.8B256K ctx Apache 2.0
1.6MFeb 2026ollama
google logo≈18 GB

DiffusionGemma is a generative model built by Google DeepMind. Based on the 26B A4B Mixture-of-Experts (MoE) Gemma 4 architecture, DiffusionGemma generates tokens using discrete diffusion. This open-weights model is multimodal, handling text, image, and video inputs to generate text output.

25.8BA4B256K ctx Apache 2.0
563.7KJun 2026gguf

ZwZ2-27B FAQ

How much VRAM does ZwZ2-27B need?
About 19 GB at Q4_K_M with an 8K context. Q8_0 needs about 31 GB and full precision about 56 GB. Longer contexts need more memory for the KV cache.
Can ZwZ2-27B run on the Tesla P40?
Yes. It fits at Q5_K_M (about 21 GB), at an estimated 9.8 tokens/s. It fits comfortably on the Radeon RX 7900 XTX, Instinct MI50 32GB, RTX 5090.
Can I use ZwZ2-27B commercially?
Unclear. It is released under Unknown licence. No licence stated: assume all rights reserved. Check the licence on the model card before using it in a product.
What is the context length of ZwZ2-27B?
262,144 tokens (256K). The full window needs far more memory than an 8K context.
How do I run ZwZ2-27B with llama.cpp?
Run llama-server -hf inclusionAI/ZwZ2-27B-GGUF:Q4_K_M. It downloads the GGUF and starts an OpenAI-compatible server.

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.