DFLASH needs about 3.3 GB of memory at BF16 (a 2.6 GB file) with an 8K context window, so it fits on a 4 GB GPU. This repository has 8 quants, from 0.7 GB to 213.6 GB. Longer context needs more: at 32K tokens the KV cache alone adds about 0.8 GB.
DFLASH is a 396B-parameter Ornith model in GGUF format, quantized by eemin. Its trained context length is 256K tokens. The repository has 8 quant files, from 0.7 GB to 213.6 GB.
Quants & memory
| Quant | File | Est. memory | vs VRAM | Result | |
|---|---|---|---|---|---|
| — | 0.7 GB | 1.4 GB | Fits | Get file → | |
| — | 0.9 GB | 1.6 GB | Fits | Get file → | |
| — | 1.4 GB | 2.1 GB | Fits | Get file → | |
| BF16 | 2.6 GB | 3.3 GB | Fits | Get file → | |
| — | 97.8 GB (5 parts) | 98.5 GB | Too big | Get file → | |
| — | 148.5 GB (8 parts) | 149.2 GB | Too big | Get file → | |
| — | 204.6 GB (11 parts) | 205.3 GB | Too big | Get file → | |
| — | 213.6 GB (11 parts) | 214.3 GB | Too big | Get file → |
What GPU do you need for DFLASH?
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 BF16 (2.6 GB file, about 3.3 GB in use).
- 6 GB VRAM: the best quant that fits is BF16 (2.6 GB file, about 3.3 GB in use).
- 8 GB VRAM: the best quant that fits is BF16 (2.6 GB file, about 3.3 GB in use).
- 12 GB VRAM: the best quant that fits is BF16 (2.6 GB file, about 3.3 GB in use).
- 16 GB VRAM: the best quant that fits is BF16 (2.6 GB file, about 3.3 GB in use).
- 24 GB VRAM: the best quant that fits is BF16 (2.6 GB file, about 3.3 GB in use).
- 32 GB VRAM: the best quant that fits is BF16 (2.6 GB file, about 3.3 GB in use).
- 48 GB VRAM: the best quant that fits is BF16 (2.6 GB file, about 3.3 GB in use).
Which DFLASH quant should you choose?
BF16 (2.6 GB) is the middle option in this repository. The unquantized model. It gives original quality and needs the most memory, so it is mainly for conversion or reference.
If memory is tight, the smallest file is the first one in the table at 0.7 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 DFLASH
How much VRAM does DFLASH need?
About 3.3 GB at BF16 with an 8K context window. That is the 2.6 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 DFLASH on an 8 GB GPU?
Yes, with BF16: the 2.6 GB file needs about 3.3 GB at 8K context. Larger quants in this repository do not fit in 8 GB.
What is the file size of DFLASH GGUF?
It depends on the quant: file is 0.7 GB, file is 0.9 GB, file is 1.4 GB, BF16 is 2.6 GB, file is 97.8 GB, file is 148.5 GB and 2 more.
What context length does DFLASH support?
The GGUF header lists a trained context length of 262,144 tokens (256K). The KV cache grows with the context you use: about 0.2 GB at 8K and 0.8 GB at 32K.
Where can I download DFLASH GGUF?
The files are on Hugging Face in the eemin/Ornith-1.0-397B-DFLASH-GGUF repository. The download buttons on this page open the file on Hugging Face. ggufmodels.com does not host any model files.