4B XHToken Spark
Spark-X2.6 4B is a small XHToken model with a nominal 1M-token window. Only 9 of its 36 layers keep a cache that grows with context; the other 27 use a 512-token sliding window. Q4_K_M needs about 3.0 GB at 4K, comfortable on an 8 GB card, and 4.1 GB at 32K, comfortable on an 8 GB card. At 128K it needs about 7.8 GB, tight on an 8 GB card, and the full 1M window needs over 40 GB. llama.cpp runs it as spark2-5. XHToken's own Q4_K_M file is 3.5 GB because it keeps the embeddings and attention gates at 16 bits; sizes here use bartowski's standard 2.6 GB file. No per-level quality loss has been published.
1049K
Max Context
2
Quant Variants
GGUF Q8_0
Best Quality
—
Perplexity rise (best build)
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
| Format | Level | BPW | VRAM | PPL Loss | Speed | Source | Actions |
|---|---|---|---|---|---|---|---|
| GGUF | Q4_K_M | 5.1 | 3.0 GB | — | — | Estimated | |
| GGUF | Q8_0 | 8.52 | 4.8 GB | — | — | Estimated |
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.
- GGUF Q4_K_M[bartowski/Spark-X2.6-4B-GGUF](https://huggingface.co/bartowski/Spark-X2.6-4B-GGUF) · Spark-X2.6-4B-Q4_K_M.gguf · 2.62 GB (2,621,104,768 bytes) · listing read 2026-10-26
- GGUF Q8_0[bartowski/Spark-X2.6-4B-GGUF](https://huggingface.co/bartowski/Spark-X2.6-4B-GGUF) · Spark-X2.6-4B-Q8_0.gguf · 4.38 GB (4,375,022,208 bytes) · listing read 2026-10-26
Similar models #
[Compare with Llama 3.1](https://quantized.uk/tools/compare/?a=spark-x2.6-4b&b=llama-3.1-8b)
[8BConsumer GPUMac / Apple Silicon3.3 GBmin VRAM+0.1%perplexity rise](https://quantized.uk/quant-hub/llama-3.1-8b/)
Llama 3.1 8B Instruct
Meta Llama 3.1
Meta's flagship 8B model with 128K context. Best-in-class for local deployment.
7BConsumer GPUMac / Apple Silicon4.2 GBmin VRAM+0.7%perplexity rise
Qwen2.5 7B Instruct
Alibaba Qwen2.5
Alibaba's highly optimized 7B. Punches well above its weight, especially in coding.
3.8BConsumer GPUMac / Apple Silicon3.6 GBmin VRAM+0.2%perplexity rise
Phi-3.5 Mini Instruct
Microsoft Phi
Microsoft's tiny powerhouse. Best 4B model for on-device deployment.
8BConsumer GPUMac / Apple Silicon5.4 GBmin VRAM+2.3%perplexity rise
Nous Hermes 3 Llama 3.1 8B
NousResearch
Fine-tuned Llama 3.1 8B with improved roleplay and instruction following.
Running Spark-X2.6 4B locally #
At Q4_K_M and 4K of context, Spark-X2.6 4B needs about 3.0 GB — 2.5 GB of weights, 0.25 GB of KV cache and a 0.3 GB activation buffer. The smallest cards in this index that clear that comfortably are 8 GB ones — 11 of them, from the Mac M1 8G up — and 78 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 0.98 GB, taking the total to 4.1 GB. That is gentler than the parameter count suggests: only 9 of 36 layers keep a KV cache that grows with context, and 27 more cap theirs at a sliding window, so the cache scales at a fraction of the usual rate. The model's native window is 1M; holding all of it at Q4_K_M would need about 42 GB.
Which build to download #
This index only tracks Spark-X2.6 4B in GGUF, across 2 levels (Q4_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 Spark-X2.6 4B need?
- About 3.0 GB at Q4_K_M with 4K of context and batch 1, rising to roughly 4.1 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 Spark-X2.6 4B run on a 8GB GPU?
- Yes — at Q4_K_M and 4K context it needs about 3.0 GB, which leaves 3.0 GB spare on a Mac M1 8G, or on any of the other 8 GB cards here (11 in all). That is the smallest size in this index that clears it comfortably; 78 of 78 do.
- Which quantization of Spark-X2.6 4B 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 2 levels in the index are Q4_K_M, Q8_0.
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
-
Mac M1 8G8GB · +3.0Mac M3 8G8GB · +3.0RTX 40608GB · +5.0RTX 4060 Ti 8G8GB · +5.0Radeon RX 9060 XT 8G8GB · +5.0
-
Ships in
-
Try it yourself
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 2026-10-26 RSS → /feed.xml