Clover-35B-A3B-Preview

35B-A3B

DiffuSpace Clover

DiffuSpace's 35B block-diffusion model, built on the Qwen3.6-35B-A3B backbone under Apache 2.0 with a 128K window: 256 experts, 8 used for each token, so about 3B parameters are read per token. Only 10 of its 40 layers keep a cache that grows with context — the other 30 are Gated DeltaNet linear attention — so each token adds about 20 KB, a quarter of the usual shape. llama.cpp has run it left to right with self-speculative decoding since 2 November; block diffusion itself needs DiffuSpace's own SGLang path. Sizes here come from bartowski's GGUF files: Q4_K_M is 23.4 GB at 4K, tight on a 24 GB card, while Q3_K_M is 18.2 GB at 4K, comfortable on a 24 GB card and still comfortable at its full window. No quality loss has been published for any level.

⬇ 9.7K HF downloads♥ 4 likesbartowski/Diffuspace_Clover-35B-A3B-Preview-GGUF· stats from 11/13/2026
Consumer GPUPro GPUMac / Apple Silicon

131K

Max Context

5

Quant Variants

GGUF Q8_0

Best Quality

—

Perplexity rise (best build)

Calculate VRAMHugging FaceCompare

Quantization Variants

Per-quant VRAM, quality loss, and inference speed on RTX 4090

Measured = run on this site · Estimated = not run here: calculated, or a figure this site has not verified · Community = public reports

FormatLevelBPWVRAMPPL LossSpeedSourceActions
GGUFQ3_K_M418.2 GB——Estimated
CalcHF
GGUFQ4_04.620.9 GB——Estimated
CalcHF
GGUFQ4_K_M5.1523.4 GB——Estimated
CalcHF
GGUFQ5_K_M6.2228.3 GB——Estimated
CalcHF
GGUFQ8_08.5138.6 GB——Estimated
CalcHF

Which file each size is for

A file size is not a memory figure: the VRAM column is the calculator's total at 4K context — weights plus KV cache plus a 10% activation buffer. Units differ on purpose: file sizes are decimal GB, as Hugging Face lists them; the VRAM column is GiB (1024³ bytes) with 2% added to the weights, so 7.12 GB of file counts as about 6.8 GiB there. A vision projector (mmproj), where the repo ships one, is a separate download and is in neither number.

Similar models

Compare with Qwen3-VL 30B-A3B
30B-A3B

Qwen3-VL 30B-A3B Instruct

Alibaba Qwen3-VL

Consumer GPUPro GPU
Q4_K_M19.9 GB+2.6%perplexity rise

Multimodal MoE with only ~3B active parameters, so it stays responsive on Apple unified memory and survives CPU offload far better than a dense 30B. At Q4_K_M it needs about 19.9 GB at 4K, comfortable on a 24 GB card; Q3_K_M brings that to 15.1 GB. The vision encoder is a separate 1.1 GB file.

36B

Seed-OSS 36B Instruct

ByteDance Seed

Consumer GPUPro GPU
Q4_K_M23.9 GB+2.7%perplexity rise

Dense 36B with a native 512K context. The Q4_K_M file is 21.8 GB, so with cache and buffers it needs about 23.9 GB at 4K: tight on a 24 GB card, where Q3_K_M (19.5 GB) is the comfortable choice, or a 2×16GB split. The AWQ build is 21.1 GB, not the 4-bit rate the name suggests. Context is the real cost: 512K of KV cache is roughly 128GB on its own, so budget context first and weights second.

27B

Qwen3.8 27B

Alibaba Qwen3.8

Consumer GPUPro GPU
Q4_K_M18.5 GB—perplexity rise

Hybrid attention: only 16 of its 64 layers keep a KV cache, which is why a 27B model holds a 262K context in about 16GB of cache rather than 64GB. Sizing it as a conventional stack overstates the cache fourfold — the estimates here count the 16 and match published measurements at 8K, 32K and 262K. Quality loss per level has not been published.

30B-A3B

GLM-4.7-Flash

Zhipu GLM-4.7

Consumer GPUPro GPU
Q4_K_M19.5 GB—perplexity rise

Zhipu's lightweight member of GLM-4.7: 30B total, about 3B active per token, MIT licence. Its attention is MLA, which caches one compressed 576-value vector per token per layer instead of full keys and values — so at 32K context the cache is under 2 GB where a conventional 47-layer model would need several times that. llama.cpp runs it as a DeepSeek-2-style model. Sizes below come from bartowski's GGUF files; quality loss per level has not been published.

Running Clover-35B-A3B-Preview locally

At Q4_K_M and 4K of context, Clover-35B-A3B-Preview needs about 23.4 GB — 21.2 GB of weights, 0.08 GB of KV cache and a 2.1 GB activation buffer. The smallest cards in this index that clear that comfortably are 32 GB ones — 4 of them, from the RTX 5090 up — and 26 of the 78 cards here do. These are calculated figures, not measurements: the estimate stops counting a card as comfortable at 88% of its VRAM, which is roughly the room a desktop session needs.

What longer context costs

Going from 4K to 32K adds about 0.55 GB, taking the total to 24.0 GB. That is gentler than the parameter count suggests: only 10 of 40 layers keep a KV cache that grows with context, so the cache scales at a fraction of the usual rate. The model's native window is 128K; holding all of it at Q4_K_M would need about 26 GB.

Which build to download

This index only tracks Clover-35B-A3B-Preview in GGUF, across 5 levels (Q3_K_M, Q4_0, Q4_K_M, Q5_K_M, Q8_0). No per-level perplexity sweep has been published for this model, so the quality column is empty rather than estimated — within one model, more bits per weight is the only ordering the data supports. No throughput has been measured for this model on this site's hardware. GGUF runs on llama.cpp and Ollama across NVIDIA, AMD and Apple silicon; AWQ and GPTQ target vLLM on CUDA and ROCm; EXL2 is ExLlamaV2 — CUDA only, and now archived, so local use rather than serving.

Common questions

How much VRAM does Clover-35B-A3B-Preview need?
About 23.4 GB at Q4_K_M with 4K of context and batch 1, rising to roughly 24.0 GB at 32K. That figure is weights plus KV cache plus a 10% activation buffer, calculated from the model's architecture rather than measured on a card.
Will Clover-35B-A3B-Preview run on a 32GB GPU?
Yes — at Q4_K_M and 4K context it needs about 23.4 GB, which leaves 8.6 GB spare on a RTX 5090, or on any of the other 32 GB cards here (4 in all). That is the smallest size in this index that clears it comfortably; 26 of 78 do.
Which quantization of Clover-35B-A3B-Preview should I use?
No published quality comparison exists for this model, so pick by footprint: Q4_K_M is the level most people run, and a higher bits-per-weight level is more faithful. All 5 levels in the index are Q3_K_M, Q4_0, Q4_K_M, Q5_K_M, Q8_0. Note this is a mixture-of-experts model — all parameters must be resident even though only a fraction are active per token, so the memory cost follows the total, not the active count.

Where this model fits

Sized at Q4_K_M with a 4K context window, smallest card first. Comfortable means the estimate uses at most 88% of the memory.

Runs comfortably on
RTX 509032GB · +8.6Arc Pro B70 32G32GB · +8.6Radeon AI PRO R9700 32G32GB · +8.6Instinct MI100 32G32GB · +8.6Mac M3 Pro 36G36GB · +3.6
Tight but possible
RTX 309024GBRadeon RX 7900 XTX24GBRTX 409024GB
Just misses
Radeon RX 7900 XT+3.4GB overMac M4 24G+5.4GB over
Ships in
GGUF
Try it yourself
Size it in the calculatorGenerate the run command

How to actually run this

Deployment guides for this model and this class of hardware.

This page's figures change when the model or the runtime does.Last updated 2026-11-13 RSS → /feed.xml