gemma 4 21b a4b it REAP needs about 22.4 GB of memory at Q4_K_M (a 13.8 GB file) with an 8K context window, so it fits on a 24 GB GPU. This repository has 11 quants, from 8.8 GB (Q2_K) to 41.6 GB (BF16). Longer context needs more: at 32K tokens the KV cache alone adds about 32.2 GB.
Gemma‑4‑21B A4B‑IT REAP, a 20% expert‑pruned Gemma‑4 model
Gemma‑4‑21B A4B‑IT REAP is a 20% expert‑pruned variant of Google Gemma‑4‑26B‑A4B‑IT. It retains the same routing behavior and active parameter count per token while reducing disk and memory footprint by ~18%. The model supports reasoning, tools, and vision.
Good at
- Reasoning and math tasks
- Tool‑calling and agentic workflows
- Vision‑enabled generation
Limits
- No vision support in quantized files
- Requires large VRAM for higher‑precision quants
- License restricts commercial use
Best for
- Local inference on high‑VRAM GPUs
- Fine‑tuning with limited disk space
- Rapid prototyping of multimodal applications
Who it’s for. Developers and researchers needing a compact Gemma‑4 model for local deployment. The REAP pruning keeps routing behavior intact, making it suitable for tasks that rely on expert selection.
Quants & memory
| Quant | File | Est. memory | vs VRAM | Result | |
|---|---|---|---|---|---|
| Q2_K | 8.8 GB | 17.3 GB | Too big | Get file → | |
| Q3_K_S | 10.1 GB | 18.7 GB | Too big | Get file → | |
| Q3_K_M | 11.0 GB | 19.5 GB | Too big | Get file → | |
| Q3_K_L | 11.4 GB | 20.0 GB | Too big | Get file → | |
| Q4_K_S | 12.8 GB | 21.3 GB | Too big | Get file → | |
| Q4_K_M | 13.8 GB | 22.4 GB | Too big | Get file → | |
| Q5_K_S | 14.8 GB | 23.4 GB | Too big | Get file → | |
| Q5_K_M | 15.7 GB | 24.3 GB | Too big | Get file → | |
| Q6_K | 18.6 GB | 27.2 GB | Too big | Get file → | |
| Q8_0 | 22.1 GB | 30.7 GB | Too big | Get file → | |
| BF16 | 41.6 GB | 50.1 GB | Too big | Get file → |
For 8 GB VRAM, use Q4_K_S (file size 12.75 GB, estimated 21.3 GB memory). With 12 GB VRAM, Q4_K_M (13.83 GB, 22.4 GB memory) is feasible. For 16 GB VRAM, Q5_K_M (15.75 GB, 24.3 GB memory) offers higher quality. Higher‑precision quants require more VRAM.
What GPU do you need for gemma 4 21b a4b it REAP?
These figures add the file size, an fp16 KV cache at an 8K context window (worked out from this model’s own header) and about 0.5 GB for the runtime. They are estimates, so leave a little headroom, and expect to use less context or a smaller quant if you are right at the limit.
- 4 GB VRAM: no quant fits at an 8K context.
- 6 GB VRAM: no quant fits at an 8K context.
- 8 GB VRAM: no quant fits at an 8K context.
- 12 GB VRAM: no quant fits at an 8K context.
- 16 GB VRAM: no quant fits at an 8K context.
- 24 GB VRAM: the best quant that fits is Q5_K_S (14.8 GB file, about 23.4 GB in use).
- 32 GB VRAM: the best quant that fits is Q8_0 (22.1 GB file, about 30.7 GB in use).
- 48 GB VRAM: the best quant that fits is Q8_0 (22.1 GB file, about 30.7 GB in use).
Which gemma 4 21b a4b it REAP quant should you choose?
Q4_K_M is the usual starting point for gemma 4 21b a4b it REAP: it balances file size and quality well. Here it is a 13.8 GB file.
The Q8_0 file is 22.1 GB, about 1.6 times the size of Q4_K_M. It keeps the weights closer to the original, which only matters if you have the memory to spare.
If memory is tight, the smallest file is Q2_K at 8.8 GB, with the quality trade-offs that come with fewer bits. Use it only when memory is very tight. Expect a clear drop in quality, which is usually worse on small models and easier to live with on large ones.
New to quant names? Read GGUF quantization explained.
Specifications
Architecture from the GGUF header
Source
Frequently asked questions about gemma 4 21b a4b it REAP
How much VRAM does gemma 4 21b a4b it REAP need?
About 22.4 GB at Q4_K_M with an 8K context window. That is the 13.8 GB file, plus the KV cache and roughly 0.5 GB of runtime overhead. Smaller quants need less and longer context needs more.
Can I run gemma 4 21b a4b it REAP on an 8 GB GPU?
Not at an 8K context. The smallest file needs about 17.3 GB, so you would need a larger GPU or CPU offloading.
What is the file size of gemma 4 21b a4b it REAP GGUF?
It depends on the quant: Q2_K is 8.8 GB, Q3_K_S is 10.1 GB, Q3_K_M is 11.0 GB, Q3_K_L is 11.4 GB, Q4_K_S is 12.8 GB, Q4_K_M is 13.8 GB and 5 more.
What context length does gemma 4 21b a4b it REAP support?
The GGUF header lists a trained context length of 262,144 tokens (256K). The KV cache grows with the context you use: about 8.1 GB at 8K and 32.2 GB at 32K.
Where can I download gemma 4 21b a4b it REAP GGUF?
The files are on Hugging Face in the barozp/gemma-4-21b-a4b-it-REAP-GGUF repository. The download buttons on this page open the file on Hugging Face. ggufmodels.com does not host any model files.
What license does gemma 4 21b a4b it REAP use?
Hugging Face lists the license as gemma. Licenses can differ between the quantized files and the original model, so check the repository and the base model before commercial use.
Can I run Gemma‑4‑21B A4B‑IT REAP on 8 GB VRAM?
Use the Q4_K_S quant. It requires approximately 21.3 GB memory at 8k context, so 8 GB VRAM is insufficient for full inference.
Does the model support vision tasks?
The quantized files do not include vision support; the model is text‑only in this configuration.
What is the active parameter count per token?
The active parameter count per token remains unchanged at ~4B after REAP pruning.