Nemotron Cascade 14B Thinking needs about 10.8 GB of memory at Q4_K_M (a 9.0 GB file) with an 8K context window, so it fits on a 12 GB GPU. This repository has 27 quants, from 0.0 GB to 29.5 GB (BF16). Longer context needs more: at 32K tokens the KV cache alone adds about 5.4 GB.
14.8B‑parameter Nemotron Cascade model focused on reasoning
Nemotron Cascade 14B Thinking is a 14.8B‑parameter model built on the Qwen3 architecture. It is trained with sequential and domain‑wise reinforcement learning for advanced reasoning tasks. The model supports a 32,768‑token context window and is optimized for long‑form reasoning.
Good at
- Advanced reasoning
- Long‑context problem solving
- Domain‑specific instruction following
Limits
- No vision support
- Requires RoPE scaling for extended context
- License is other
Best for
- Complex reasoning workflows
- Long‑form content generation
- Domain‑specific instruction tasks
Who it’s for. Developers and researchers needing a large, reasoning‑oriented model for local deployment. It is suitable for applications that require deep logical analysis and extended context handling.
Quants & memory
| Quant | File | Est. memory | vs VRAM | Result | |
|---|---|---|---|---|---|
| — | 0.0 GB | 1.8 GB | Fits | Get file → | |
| IQ2_XS | 4.7 GB | 6.5 GB | Fits | Get file → | |
| IQ2_S | 5.0 GB | 6.8 GB | Fits | Get file → | |
| IQ2_M | 5.3 GB | 7.2 GB | Fits | Get file → | |
| Q2_K | 5.8 GB | 7.6 GB | Tight | Get file → | |
| IQ3_XXS | 5.9 GB | 7.8 GB | Tight | Get file → | |
| IQ3_XS | 6.4 GB | 8.2 GB | Too big | Get file → | |
| Q2_K_L | 6.5 GB | 8.4 GB | Too big | Get file → | |
| Q3_K_S | 6.7 GB | 8.5 GB | Too big | Get file → | |
| IQ3_M | 6.9 GB | 8.7 GB | Too big | Get file → | |
| Q3_K_M | 7.3 GB | 9.2 GB | Too big | Get file → | |
| Q3_K_L | 7.9 GB | 9.7 GB | Too big | Get file → | |
| IQ4_XS | 8.1 GB | 10.0 GB | Too big | Get file → | |
| IQ4_NL | 8.5 GB | 10.4 GB | Too big | Get file → | |
| Q4_0 | 8.5 GB | 10.4 GB | Too big | Get file → | |
| Q4_K_S | 8.6 GB | 10.4 GB | Too big | Get file → | |
| Q3_K_XL | 8.6 GB | 10.4 GB | Too big | Get file → | |
| Q4_K_M | 9.0 GB | 10.8 GB | Too big | Get file → | |
| Q4_1 | 9.4 GB | 11.2 GB | Too big | Get file → | |
| Q4_K_L | 9.6 GB | 11.4 GB | Too big | Get file → | |
| Q5_K_S | 10.3 GB | 12.1 GB | Too big | Get file → | |
| Q5_K_M | 10.5 GB | 12.4 GB | Too big | Get file → | |
| Q5_K_L | 11.0 GB | 12.8 GB | Too big | Get file → | |
| Q6_K | 12.1 GB | 14.0 GB | Too big | Get file → | |
| Q6_K_L | 12.5 GB | 14.3 GB | Too big | Get file → | |
| Q8_0 | 15.7 GB | 17.5 GB | Too big | Get file → | |
| BF16 | 29.5 GB | 31.4 GB | Too big | Get file → |
For 8 GB VRAM, use IQ2_XS (4.69 GB, 6.5 GB memory at 8K context). For 12 GB VRAM, IQ4_K_M (9 GB, 10.8 GB memory) is a good balance. For 24 GB VRAM, Q6_K (12.12 GB, 14 GB memory) provides higher quality with manageable memory usage. Estimates are approximate.
What GPU do you need for Nemotron Cascade 14B Thinking?
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: the best quant that fits is the main file (0.0 GB file, about 1.8 GB in use).
- 6 GB VRAM: the best quant that fits is the main file (0.0 GB file, about 1.8 GB in use).
- 8 GB VRAM: the best quant that fits is IQ3_XXS (5.9 GB file, about 7.8 GB in use).
- 12 GB VRAM: the best quant that fits is Q4_K_L (9.6 GB file, about 11.4 GB in use).
- 16 GB VRAM: the best quant that fits is Q6_K_L (12.5 GB file, about 14.3 GB in use).
- 24 GB VRAM: the best quant that fits is Q8_0 (15.7 GB file, about 17.5 GB in use).
- 32 GB VRAM: the best quant that fits is BF16 (29.5 GB file, about 31.4 GB in use).
Which Nemotron Cascade 14B Thinking quant should you choose?
Q4_K_M is the usual starting point for Nemotron Cascade 14B Thinking: it balances file size and quality well. Here it is a 9.0 GB file.
The Q8_0 file is 15.7 GB, about 1.7 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 the first one in the table at 0.0 GB, with the quality trade-offs that come with fewer bits.
New to quant names? Read GGUF quantization explained.
Specifications
Architecture from the GGUF header
Source
Frequently asked questions about Nemotron Cascade 14B Thinking
How much VRAM does Nemotron Cascade 14B Thinking need?
About 10.8 GB at Q4_K_M with an 8K context window. That is the 9.0 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 Nemotron Cascade 14B Thinking on an 8 GB GPU?
Yes, with IQ3_XXS: the 5.9 GB file needs about 7.8 GB at 8K context. Larger quants in this repository do not fit in 8 GB.
What is the file size of Nemotron Cascade 14B Thinking GGUF?
It depends on the quant: file is 0.0 GB, IQ2_XS is 4.7 GB, IQ2_S is 5.0 GB, IQ2_M is 5.3 GB, Q2_K is 5.8 GB, IQ3_XXS is 5.9 GB and 21 more.
What context length does Nemotron Cascade 14B Thinking support?
The GGUF header lists a trained context length of 32,768 tokens (32K). The KV cache grows with the context you use: about 1.3 GB at 8K and 5.4 GB at 32K.
Where can I download Nemotron Cascade 14B Thinking GGUF?
The files are on Hugging Face in the bartowski/nvidia_Nemotron-Cascade-14B-Thinking-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 Nemotron Cascade 14B Thinking use?
Hugging Face lists the license as other. 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 Nemotron Cascade 14B Thinking on 8 GB VRAM?
Yes, the IQ2_XS quant file is 4.69 GB and uses about 6.5 GB of VRAM at 8K context.
Does the model support vision or image input?
No, the model has no vision capabilities.
What context length does the model support?
The base model supports 32,768 tokens; RoPE scaling can extend this to 90K tokens with the appropriate factor.