Deepseek-Prover-V2-671B-Bf16 needs about 406.1 GB of memory at Q4_K_M (a 404.5 GB file) with an 8K context window, which is more than a 48 GB GPU holds. This repository has 26 quants, from 184.8 GB (IQ1_S) to 1342.3 GB (BF16). Longer context needs more: at 32K tokens the KV cache alone adds about 4.3 GB.
Deepseek-Prover-V2-671B-Bf16 – a 671‑billion‑parameter formal theorem‑proving model
Deepseek-Prover-V2-671B-Bf16 is a 671‑billion‑parameter LLM trained for formal theorem proving in Lean 4. It uses a recursive proof‑search pipeline and reinforcement learning to combine informal reasoning with formal proofs. The model is released under the MIT license.
Good at
- Formal theorem proving
- Lean 4 proof synthesis
- Chain‑of‑thought reasoning
Limits
- No vision support
- Requires large VRAM for full context
- No explicit language support list
Best for
- Academic research in formal methods
- Automated proof assistants
- Mathematical education tools
Who it’s for. Researchers and developers working on formal verification, theorem proving, and mathematical AI. The model is suitable for teams with access to high‑memory GPUs and a need for deep formal reasoning capabilities.
Quants & memory
| Quant | File | Est. memory | vs VRAM | Result | |
|---|---|---|---|---|---|
| IQ1_S | 184.8 GB (4 parts) | 186.4 GB | Too big | Get file → | |
| IQ1_M | 199.8 GB (5 parts) | 201.4 GB | Too big | Get file → | |
| IQ2_XXS | 216.0 GB (5 parts) | 217.5 GB | Too big | Get file → | |
| IQ2_M | 227.7 GB (5 parts) | 229.3 GB | Too big | Get file → | |
| Q2_K | 244.2 GB (5 parts) | 245.8 GB | Too big | Get file → | |
| Q2_K_L | 244.4 GB (5 parts) | 246.0 GB | Too big | Get file → | |
| Q2_K_XL | 250.4 GB (6 parts) | 252.0 GB | Too big | Get file → | |
| IQ3_XXS | 272.4 GB (6 parts) | 274.0 GB | Too big | Get file → | |
| Q3_K_S | 289.2 GB (6 parts) | 290.8 GB | Too big | Get file → | |
| Q3_K_M | 319.3 GB (7 parts) | 320.9 GB | Too big | Get file → | |
| Q6_K_XL | 336.4 GB (8 parts) | 338.0 GB | Too big | Get file → | |
| Q8_K_XL | 336.9 GB (7 parts) | 338.5 GB | Too big | Get file → | |
| IQ4_XS | 357.3 GB (8 parts) | 358.8 GB | Too big | Get file → | |
| IQ4_NL | 378.1 GB (8 parts) | 379.7 GB | Too big | Get file → | |
| Q4_0 | 379.3 GB (8 parts) | 380.9 GB | Too big | Get file → | |
| Q4_K_S | 380.1 GB (8 parts) | 381.7 GB | Too big | Get file → | |
| Q4_K_XL | 383.5 GB (8 parts) | 385.1 GB | Too big | Get file → | |
| Q4_K_M | 404.5 GB (9 parts) | 406.1 GB | Too big | Get file → | |
| Q4_1 | 419.9 GB (9 parts) | 421.5 GB | Too big | Get file → | |
| Q5_K_S | 423.9 GB (9 parts) | 425.5 GB | Too big | Get file → | |
| Q5_K_M | 475.4 GB (10 parts) | 477.0 GB | Too big | Get file → | |
| Q5_K_XL | 480.5 GB (10 parts) | 482.1 GB | Too big | Get file → | |
| Q6_K | 550.9 GB (12 parts) | 552.5 GB | Too big | Get file → | |
| Q3_K_XL | 591.0 GB (13 parts) | 592.6 GB | Too big | Get file → | |
| Q8_0 | 713.3 GB (15 parts) | 714.9 GB | Too big | Get file → | |
| BF16 | 1342.3 GB (30 parts) | 1343.9 GB | Too big | Get file → |
For 8 GB VRAM, use IQ1_S (file size 184.79 GB, memory 186.4 GB at 8k context). For 16 GB VRAM, IQ1_M (199.78 GB, 201.4 GB). For 24 GB VRAM, IQ2_XXS (215.96 GB, 217.5 GB). Higher memory budgets allow larger quants such as Q4_K_M (404.49 GB, 406.1 GB).
What GPU do you need for Deepseek-Prover-V2-671B-Bf16?
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: no quant fits at an 8K context.
- 48 GB VRAM: no quant fits at an 8K context.
Which Deepseek-Prover-V2-671B-Bf16 quant should you choose?
Q4_K_M is the usual starting point for Deepseek-Prover-V2-671B-Bf16: it balances file size and quality well. Here it is a 404.5 GB file.
The Q8_0 file is 713.3 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 IQ1_S at 184.8 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 Deepseek-Prover-V2-671B-Bf16
How much VRAM does Deepseek-Prover-V2-671B-Bf16 need?
About 406.1 GB at Q4_K_M with an 8K context window. That is the 404.5 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 Deepseek-Prover-V2-671B-Bf16 on an 8 GB GPU?
Not at an 8K context. The smallest file needs about 186.4 GB, so you would need a larger GPU or CPU offloading.
What is the file size of Deepseek-Prover-V2-671B-Bf16 GGUF?
It depends on the quant: IQ1_S is 184.8 GB, IQ1_M is 199.8 GB, IQ2_XXS is 216.0 GB, IQ2_M is 227.7 GB, Q2_K is 244.2 GB, Q2_K_L is 244.4 GB and 20 more.
What context length does Deepseek-Prover-V2-671B-Bf16 support?
The GGUF header lists a trained context length of 163,840 tokens (160K). The KV cache grows with the context you use: about 1.1 GB at 8K and 4.3 GB at 32K.
Where can I download Deepseek-Prover-V2-671B-Bf16 GGUF?
The files are on Hugging Face in the unsloth/DeepSeek-Prover-V2-671B-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 Deepseek-Prover-V2-671B-Bf16 use?
Hugging Face lists the license as mit. 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 Deepseek-Prover-V2-671B-Bf16 on 8 GB VRAM?
The smallest quant, IQ1_S, requires about 186.4 GB of VRAM at 8k context, so 8 GB is insufficient.
Does the model support image inputs?
No, the model has no vision capabilities.
What license does the model use?
The model is released under the MIT license.