Swallow-70b-NVE-instruct-hf needs about 44.6 GB of memory at Q4_K_M (a 41.4 GB file) with an 8K context window, so it fits on a 48 GB GPU. This repository has 22 quants, from 25.5 GB (Q2_K) to 73.3 GB (Q8_0).
Swallow‑70b‑NVE‑instruct‑hf – a 70‑billion‑parameter LLaMA‑2 based instruction model with Japanese and English support
Swallow‑70b‑NVE‑instruct‑hf is a 70‑billion‑parameter LLaMA‑2 derived model fine‑tuned for instruction following. It supports Japanese and English text and is available in multiple GGUF quantizations. The model was released on January 26, 2024 as part of the Swallow‑NVE series.
Good at
- Instruction following in Japanese and English
- Large context handling up to 4096 tokens
- Fine‑tuned for conversational tasks
Limits
- No vision or multimodal support
- License not specified
- Requires large VRAM for higher‑quality quants
Best for
- Developers building Japanese‑English chatbots
- Researchers needing a large instruction model
- Enterprise applications requiring high‑capacity language generation
Who it’s for. The model is intended for developers and researchers who need a large, instruction‑tuned LLaMA‑2 variant with Japanese and English capabilities. It is suitable for applications that can accommodate the memory requirements of a 70‑billion‑parameter model.
Quants & memory
| Quant | File | Est. memory | vs VRAM | Result | |
|---|---|---|---|---|---|
| Q2_K | 25.5 GB | 27.3 GB | Too big | Get file → | |
| IQ3_XS | 28.3 GB | 30.2 GB | Too big | Get file → | |
| IQ3_S | 29.9 GB | 31.8 GB | Too big | Get file → | |
| Q3_K_S | 29.9 GB | 31.8 GB | Too big | Get file → | |
| IQ3_M | 30.9 GB | 32.8 GB | Too big | Get file → | |
| Q3_K | 33.3 GB | 35.1 GB | Too big | Get file → | |
| Q3_K_M | 33.3 GB | 35.1 GB | Too big | Get file → | |
| Q3_K_L | 36.1 GB | 38.0 GB | Too big | Get file → | |
| IQ4_XS | 37.2 GB | 39.0 GB | Too big | Get file → | |
| Q4_0 | 38.9 GB | 40.7 GB | Too big | Get file → | |
| IQ4_NL | 39.2 GB | 41.1 GB | Too big | Get file → | |
| Q4_K_S | 39.2 GB | 41.1 GB | Too big | Get file → | |
| Q4_K | 41.4 GB (2 parts) | 43.3 GB | Too big | Get file → | |
| Q4_K_M | 41.4 GB (2 parts) | 43.3 GB | Too big | Get file → | |
| Q4_1 | 43.2 GB (2 parts) | 45.0 GB | Too big | Get file → | |
| Q5_0 | 47.5 GB (2 parts) | 49.3 GB | Too big | Get file → | |
| Q5_K_S | 47.5 GB (2 parts) | 49.3 GB | Too big | Get file → | |
| Q5_K | 48.8 GB (2 parts) | 50.6 GB | Too big | Get file → | |
| Q5_K_M | 48.8 GB (2 parts) | 50.6 GB | Too big | Get file → | |
| Q5_1 | 51.8 GB (2 parts) | 53.6 GB | Too big | Get file → | |
| Q6_K | 56.6 GB (2 parts) | 58.4 GB | Too big | Get file → | |
| Q8_0 | 73.3 GB (2 parts) | 75.1 GB | Too big | Get file → |
For systems with 8 GB VRAM, use Q2_K (25.46 GB file, 28.6 GB memory at 8k context). For 16 GB VRAM, Q4_0 (38.87 GB file, 42.1 GB memory) or Q4_K (41.42 GB file, 44.6 GB memory). For 24 GB VRAM, Q5_0 (47.46 GB file, 50.6 GB memory) or Q5_K (48.75 GB file, 51.9 GB memory).
What GPU do you need for Swallow-70b-NVE-instruct-hf?
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: no quant fits at an 8K context.
- 32 GB VRAM: the best quant that fits is IQ3_XS (28.3 GB file, about 31.5 GB in use).
- 48 GB VRAM: the best quant that fits is Q4_1 (43.2 GB file, about 46.4 GB in use).
Which Swallow-70b-NVE-instruct-hf quant should you choose?
Q4_K_M is the usual starting point for Swallow-70b-NVE-instruct-hf: it balances file size and quality well. Here it is a 41.4 GB file.
The Q8_0 file is 73.3 GB, about 1.8 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 25.5 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
Frequently asked questions about Swallow-70b-NVE-instruct-hf
How much VRAM does Swallow-70b-NVE-instruct-hf need?
About 44.6 GB at Q4_K_M with an 8K context window. That is the 41.4 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 Swallow-70b-NVE-instruct-hf on an 8 GB GPU?
Not at an 8K context. The smallest file needs about 28.6 GB, so you would need a larger GPU or CPU offloading.
What is the file size of Swallow-70b-NVE-instruct-hf GGUF?
It depends on the quant: Q2_K is 25.5 GB, IQ3_XS is 28.3 GB, IQ3_S is 29.9 GB, Q3_K_S is 29.9 GB, IQ3_M is 30.9 GB, Q3_K is 33.3 GB and 16 more.
What context length does Swallow-70b-NVE-instruct-hf support?
The GGUF header lists a trained context length of 4,096 tokens (4K). The KV cache grows with the context you use: about 2.7 GB at 8K and 1.3 GB at 4K.
Where can I download Swallow-70b-NVE-instruct-hf GGUF?
The files are on Hugging Face in the RichardErkhov/tokyotech-llm_-_Swallow-70b-NVE-instruct-hf-gguf repository. The download buttons on this page open the file on Hugging Face. ggufmodels.com does not host any model files.
Can I run Swallow‑70b‑NVE‑instruct‑hf on 8 GB VRAM?
Only the lowest‑quality quantizations such as Q2_K are feasible on 8 GB VRAM, but performance may be limited.
Does the model support vision or multimodal inputs?
No, the model is purely text‑based and does not support vision or multimodal inputs.
What languages does the model understand?
The model supports Japanese and English text.