DeepSeek V4 Pro
DeepSeek V4 Pro is an open-source Mixture-of-Experts text-generation model hosted at deepseek-ai/DeepSeek-V4-Pro under the MIT license and intended for use with the transformers library. Its official model card describes a hybrid attention architecture combining Compressed Sparse Attention and Heavily Compressed Attention, plus Manifold-Constrained Hyper-Connections and the Muon optimizer, with support for a one-million-token context length. The card states 1.6T parameters with 49B activated. The supplied information does not include numerical hardware requirements or benchmark results for this model, and no testing or compatibility is implied.
At a glance
- Parameters
- 1598.84B
- Architecture
- deepseek_v4
- Context length
- 1,048,576
- License
- mit
- Software
- Publisher PyTorch reference
Will it run on your machine?
Save your machine to see a personalized rating and its reasoning.
Add my hardwareBest for
- Long-context text generation with up to one million tokens of context
- Working with long documents or other extended text inputs
- Text-generation tasks via the transformers pipeline
- Experimental exploration of the model's hybrid attention mechanism
- Research involving million-token context intelligence
Tradeoffs
- Supplied source is a preview model and the official model card is incomplete text.
- No numerical hardware requirements are provided.
- No evaluation results or performance benchmarks for this model are included.
- Compatibility with specific software or hardware is not stated.
- No testing or validation by this guide is claimed.
Ways to run it
Publisher PyTorch reference · Linux / rented GPU
Repository-specific command in publisher documentation; confirm dependencies and hardware in the source.
torchrun --nproc-per-node ${MP} generate.py --ckpt-path ${SAVE_PATH} --config ${CONFIG} --interactiveExplore its uses
Keep exploring
Bonsai 2 · 27B
A compressed 27B-class reasoning model with publisher-provided GGUF packs and a dedicated llama.cpp fork for CUDA, Metal, and CPU.
DeepSeek V4.1 Flash
A multimodal reasoning model with a compressed key-value cache, published as open weights.
Qwen3.8-Flash-Next
An experimental open-weight multimodal model with sparse attention and 262K native context.