KAT Coder V2.5 Dev 35B A3B MTP ABLITERATED needs about 22.9 GB of memory at Q4_K_M (a 21.7 GB file) with an 8K context window, so it fits on a 24 GB GPU. This repository has 2 quants, from 21.7 GB (Q4_K_M) to 71.1 GB (BF16). Longer context needs more: at 32K tokens the KV cache alone adds about 2.8 GB.
KAT Coder V2.5 Dev 35B A3B MTP Abliterated – 35.5B parameter coding model
KAT Coder V2.5 Dev 35B A3B MTP Abliterated is a 35.5B parameter model built on the qwen35moe architecture. It includes a transplanted Multi‑Token Prediction layer for speculative decoding. The model is released under the Apache‑2.0 license.
Good at
- coding
- reasoning
- tools
Limits
- trained context 262144 tokens
- no vision support
- requires large VRAM
Best for
- local coding assistants
- agentic coding workflows
- speculative decoding experiments
Who it’s for. Developers and researchers who need a large language model for coding tasks and want to experiment with Multi‑Token Prediction in llama.cpp. The model is suitable for local deployment on systems with sufficient GPU memory.
Quants & memory
| Quant | File | Est. memory | vs VRAM | Result | |
|---|---|---|---|---|---|
| Q4_K_M | 21.7 GB | 22.9 GB | Too big | Get file → | |
| BF16 | 71.1 GB | 72.3 GB | Too big | Get file → |
For 8 GB VRAM, use the Q4_K_M quantization (21.71 GB file, ~22.9 GB VRAM at 8 k context). For higher quality with more memory, BF16 (71.07 GB file, ~72.3 GB VRAM at 8 k context) is available. Memory estimates are approximate.
What GPU do you need for KAT Coder V2.5 Dev 35B A3B MTP ABLITERATED?
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 Q4_K_M (21.7 GB file, about 22.9 GB in use).
- 32 GB VRAM: the best quant that fits is Q4_K_M (21.7 GB file, about 22.9 GB in use).
- 48 GB VRAM: the best quant that fits is Q4_K_M (21.7 GB file, about 22.9 GB in use).
Which KAT Coder V2.5 Dev 35B A3B MTP ABLITERATED quant should you choose?
Q4_K_M is the usual starting point for KAT Coder V2.5 Dev 35B A3B MTP ABLITERATED: it balances file size and quality well. Here it is a 21.7 GB file.
New to quant names? Read GGUF quantization explained.
Specifications
Architecture from the GGUF header
Frequently asked questions about KAT Coder V2.5 Dev 35B A3B MTP ABLITERATED
How much VRAM does KAT Coder V2.5 Dev 35B A3B MTP ABLITERATED need?
About 22.9 GB at Q4_K_M with an 8K context window. That is the 21.7 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 KAT Coder V2.5 Dev 35B A3B MTP ABLITERATED on an 8 GB GPU?
Not at an 8K context. The smallest file needs about 22.9 GB, so you would need a larger GPU or CPU offloading.
What is the file size of KAT Coder V2.5 Dev 35B A3B MTP ABLITERATED GGUF?
It depends on the quant: Q4_K_M is 21.7 GB, BF16 is 71.1 GB.
What context length does KAT Coder V2.5 Dev 35B A3B MTP ABLITERATED support?
The GGUF header lists a trained context length of 262,144 tokens (256K). The KV cache grows with the context you use: about 0.7 GB at 8K and 2.8 GB at 32K.
Where can I download KAT Coder V2.5 Dev 35B A3B MTP ABLITERATED GGUF?
The files are on Hugging Face in the jakeroxs/KAT-Coder-V2.5-Dev-35B-A3B-MTP-ABLITERATED-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 KAT Coder V2.5 Dev 35B A3B MTP ABLITERATED use?
Hugging Face lists the license as apache-2.0. 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 this model on 8 GB VRAM?
Yes, using the Q4_K_M quantization which requires about 22.9 GB VRAM at 8 k context, but actual VRAM usage may be lower with optimizations.
Does the model support vision?
No, the model does not have vision capabilities.
What license applies?
The model is released under the Apache‑2.0 license.