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.

tencent logo

Hy4-flash-base

tencent·Base model·released 17 Nov 2026

tencent-hy-license219BMoE1M ctx ≈133 GB
Your GPU
hf download tencent/Hy4-flash-base
Setup ↓

Hy4 flash is our most cost-effective Hy model to date, featuring 220B total parameters (including 64B Engram parameters) with just 12B active parameters per token. It natively accepts text and image inputs, generates text, and supports a context window of up to 1M tokens.

Notify me when

How much VRAM does Hy4-flash-base need?

At Q4_K_M Hy4-flash-base needs about 133 GB with an 8K context (133 GB at Q4, 432 GB at full precision). Pick your GPU to see what fits and how fast it runs.

--vram
any
--gpus
--quant

Needs: ≈133 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)

Quantisation8K16K32K64K128K1M
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 loss67 GB 2.6 GB≈74 GB
Q3_K_MNoticeable lossSmall, some quality lossNoticeable loss100 GB 2.6 GB≈108 GB
Q4_K_MrecommendedSmall lossRecommended default for local useSmall loss124 GB 2.6 GB≈133 GB
Q5_K_MNear-losslessNear-lossless for most tasksNear-lossless145 GB 2.6 GB≈156 GB
Q6_KNear-losslessPractically losslessNear-lossless168 GB 2.6 GB≈180 GB
Q8_0LosslessLossless in practice, twice the size of Q4Lossless217 GB 2.6 GB≈231 GB
FP16 / BF16LosslessFull precision, as releasedLossless409 GB 2.6 GB≈432 GB

Estimates for 8K tokens of context: weights + KV cache + runtime overhead. All experts stay in memory, but only the active ones are read per token. 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?

Model quant
KV cache

At 8K context, Q4_K_M needs ≈133 GB(124 GB weights + 2.6 GB KV cache + 6.7 GB overhead). Pick a GPU above to see how much context fits.

01002003004005001K2K4K8K16K32K64K128K256K512K1MGB

Weights + overheadKV cache

Can the Apple M2 Ultra (192 GB) run Hy4-flash-base? →

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 Hy4-flash-base

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 Hy4-flash-base locally

These commands load the original weights from Hugging Face.

Ask about an imagedocs
python
from transformers import pipeline

pipe = pipeline("image-text-to-text", model="tencent/Hy4-flash-base", 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

About Hy4-flash-base

Hy4 flash is our most cost-effective Hy model to date, featuring 220B total parameters (including 64B Engram parameters) with just 12B active parameters per token. It natively accepts text and image inputs, generates text, and supports a context window of up to 1M tokens. Its 42-layer language backbone consists of one dense feed-forward layer and 41 MoE layers, with identity Hyper-Connections (iHC) supporting inter-layer information flow. A vision encoder (adapted from SigLIP2-So400m-Patch16-Res512) is connected to the backbone through an MLP projector. In addition to the backbone, a built-in 5-layer drafter, DFly2 (0.85B parameters), reuses the target model's KV cache and sparse indices to support long-context speculative decoding at a low, constant cost. Compared with Hy4, Hy4 Flash introduces three notable architectural changes: • Hybrid attention. Sparse attention (IndexPool) and sliding-window attention (SWA) are interleaved at a 1:3 ratio. • LatentMoE. Routed experts operate in a lower-dimensional latent space. Each token activates 16 of 512 routed experts, alongside one shared expert. • N-gram embedding (Engram). The backbone is augmented with learned bigram and trigram representations, retrieved via multi-head hash lookups. Bits-per-byte (BPB) on internal code & agent validation, lower is better: Few-shot pretrain benchmark evaluation:

Full model card on Hugging Face →

Alternatives to Hy4-flash-base

All alternatives →
Qwen logo

Qwen3.8-Flash-Next

Qwen·Vision

≈108 GB

As the frontier of foundation models pushes toward ever-larger parameter counts and ever-longer context windows, the question is no longer just how much we can scale, but how efficiently we can do so.

180BA61.6B256K ctx qwen-community-1.0
2.7MAug 2026ollama
MiniMaxAI logo

MiniMax-M3.1-Flash

MiniMaxAI·Vision

≈132 GB

MiniMax-M3.1-Flash is a native multimodal mixture-of-experts model for coding, agentic workflows, computer use, and cybersecurity. It supports text, image, and video inputs with a context window of up to 1M tokens.

220B1M ctx minimax-community
150KOct 2026ollama
zai-org logo

GLM-5.3-Flash

zai-org·Vision

≈213 GB

👋 Join our WeChat or Discord community. 📖 Check out the GLM-5.3-Flash blog and GLM-5 Technical report . 📍 Use GLM-5.3-Flash API services on Z.ai API Platform.

321BA18.3B1M ctx MIT
6.4MAug 2026ollama
thinkingmachines logo

Inkling-Small

thinkingmachines·Vision

≈159 GB

Inkling-Small is a general-purpose multimodal model that accepts text, image and audio inputs and generates text outputs. It is intended for use in English and other languages, and across multiple coding languages.

266BA1.7B1M ctx Apache 2.0
916.1KJul 2026gguf
Qwen logo≈142 GB

Meet Qwen3-VL — the most powerful vision-language model in the Qwen series to date.

236BA22B256K ctx Apache 2.0
318.9KSept 2025ollama
tencent logo

Hy4-flash

tencent·Vision

≈134 GB

Hy4 flash is our most cost-effective Hy model to date, featuring 220B total parameters (including 64B Engram parameters) with just 12B active parameters per token. It natively accepts text and image inputs, generates text, and supports a context window of up to 1M tokens.

220B1M ctx tencent-hy-license
3.9KNov 2026gguf

Hy4-flash-base FAQ

How much VRAM does Hy4-flash-base need?
About 133 GB at Q4_K_M with an 8K context. Q8_0 needs about 231 GB and full precision about 432 GB. Longer contexts need more memory for the KV cache.
Can Hy4-flash-base run on the Apple M2 Ultra (192 GB)?
Yes. It fits at Q4_K_M (about 133 GB), at an estimated 3.3 tokens/s. It fits comfortably on the Apple M3 Ultra (512 GB).
Can I use Hy4-flash-base commercially?
Unclear. It is released under tencent-hy-license. Custom licence: read the model card before using it commercially. Check the licence on the model card before using it in a product.
What is the context length of Hy4-flash-base?
1,048,576 tokens (1M). The full window needs far more memory than an 8K context.
How do I run Hy4-flash-base locally?
It runs with Transformers. Each has copy-paste examples in its docs.

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.