Deepseek-V4-Flash-oQ2.5e-mtp

Deepseek V4 Flash quantized to 2.5 bpw using oQe dynamic imatrix-weighted quantization in oMLX 0.5.1. Native MTP weights are preserved (use oMLX Lightning MTP). Fits in 128GB of unified memory. Successfully tested with Openclaw multi-turn tool-calling (including some basic coding tasks), chat, and creative writing. For Openclaw, make sure to set blockstreamingdefault to "on" or else multi-turn tool calling will fail.

oMLX's highly efficient caching leads to significantly higher speeds than ds4.c for agentic tasks, in my experience. Prompt processing and decode are marginally faster than ds4.c in my testing, due to MLX architecture and MTP.

<!-- markdownlint-disable first-line-h1 --> <!-- markdownlint-disable html --> <!-- markdownlint-disable no-duplicate-header --> <div align="center"> <img src="https://github.com/deepseek-ai/DeepSeek-V2/blob/main/figures/logo.svg?raw=true" width="60%" alt="DeepSeek-V4" /> </div> <hr> <div align="center" style="line-height: 1;"> <a href="https://www.deepseek.com/" target="_blank" style="margin: 2px;"> <img alt="Homepage" src="https://github.com/deepseek-ai/DeepSeek-V2/blob/main/figures/badge.svg?raw=true" style="display: inline-block; vertical-align: middle;"/> </a> <a href="https://chat.deepseek.com/" target="_blank" style="margin: 2px;"> <img alt="Chat" src="https://img.shields.io/badge/🤖%20Chat-DeepSeek%20V4-536af5?color=536af5&logoColor=white" style="display: inline-block; vertical-align: middle;"/> </a> </div> <div align="center" style="line-height: 1;"> <a href="https://huggingface.co/deepseek-ai" target="_blank" style="margin: 2px;"> <img alt="Hugging Face" src="https://img.shields.io/badge/%F0%9F%A4%97%20Hugging%20Face-DeepSeek%20AI-ffc107?color=ffc107&logoColor=white" style="display: inline-block; vertical-align: middle;"/> </a> <a href="https://twitter.com/deepseek_ai" target="_blank" style="margin: 2px;"> <img alt="Twitter Follow" src="https://img.shields.io/badge/Twitter-deepseek_ai-white?logo=x&logoColor=white" style="display: inline-block; vertical-align: middle;"/> </a> </div> <div align="center" style="line-height: 1;"> <a href="LICENSE" style="margin: 2px;"> <img alt="License" src="https://img.shields.io/badge/License-MIT-f5de53?&color=f5de53" style="display: inline-block; vertical-align: middle;"/> </a> </div> <p align="center"> <a href="https://arxiv.org/abs/2606.19348"><b>Technical Report</b>👁️</a> </p>

Introduction

We present a preview version of DeepSeek-V4 series, including two strong Mixture-of-Experts (MoE) language models — DeepSeek-V4-Pro with 1.6T parameters (49B activated) and DeepSeek-V4-Flash with 284B parameters (13B activated) — both supporting a context length of one million tokens.

DeepSeek-V4 series incorporate several key upgrades in architecture and optimization:

1. Hybrid Attention Architecture: We design a hybrid attention mechanism combining Compressed Sparse Attention (CSA) and Heavily Compressed Attention (HCA) to dramatically improve long-context efficiency. In the 1M-token context setting, DeepSeek-V4-Pro requires only 27% of single-token inference FLOPs and 10% of KV cache compared with DeepSeek-V3.2.
2. Manifold-Constrained Hyper-Connections (mHC): We incorporate mHC to strengthen conventional residual connections, enhancing stability of signal propagation across layers while preserving model expressivity.
3. Muon Optimizer: We employ the Muon optimizer for faster convergence and greater training stability.

We pre-train both models on more than 32T diverse and high-quality tokens, followed by a comprehensive post-training pipeline. The post-training features a two-stage paradigm: independent cultivation of domain-specific experts (through SFT and RL with GRPO), followed by unified model consolidation via on-policy distillation, integrating distinct proficiencies across diverse domains into a single model.

DeepSeek-V4-Pro-Max, the maximum reasoning effort mode of DeepSeek-V4-Pro, significantly advances the knowledge capabilities of open-source models, firmly establishing itself as the best open-source model available today. It achieves top-tier performance in coding benchmarks and significantly bridges the gap with leading closed-source models on reasoning and agentic tasks. Meanwhile, DeepSeek-V4-Flash-Max achieves comparable reasoning performance to the Pro version when given a larger thinking budget, though its smaller parameter scale naturally places it slightly behind on pure knowledge tasks and the most complex agentic workflows.

<div align="center"> <img src="assets/dsv4_performance.png" > </div>

Model Downloads

<div align="center"> </div>

*FP4 + FP8 Mixed: MoE expert parameters use FP4 precision; most other parameters use FP8.

Evaluation Results

Base Model

<div align="center"> </div>

Instruct Model

DeepSeek-V4-Pro and DeepSeek-V4-Flash both support three reasoning effort modes:

DeepSeek-V4-Pro-Max vs Frontier Models

<div align="center"> </div>

Comparison across Modes

<div align="center"> </div>

Chat Template

This release does not include a Jinja-format chat template. Instead, we provide a dedicated encoding folder with Python scripts and test cases demonstrating how to encode messages in OpenAI-compatible format into input strings for the model, and how to parse the model's text output. Please refer to the encoding folder for full documentation.

A brief example:

from encoding_dsv4 import encode_messages, parse_message_from_completion_text

messages = [
    {"role": "user", "content": "hello"},
    {"role": "assistant", "content": "Hello! I am DeepSeek.", "reasoning_content": "thinking..."},
    {"role": "user", "content": "1+1=?"}
]

# messages -> string
prompt = encode_messages(messages, thinking_mode="thinking")

# string -> tokens
import transformers
tokenizer = transformers.AutoTokenizer.from_pretrained("deepseek-ai/DeepSeek-V4-Pro")
tokens = tokenizer.encode(prompt)

How to Run Locally

Please refer to the inference folder for detailed instructions on running DeepSeek-V4 locally, including model weight conversion and interactive chat demos.

For local deployment, we recommend setting the sampling parameters to temperature = 1.0, top_p = 1.0. For the Think Max reasoning mode, we recommend setting the context window to at least 384K tokens.

License

This repository and the model weights are licensed under the MIT License.

Citation

@misc{deepseekai2026deepseekv4,
      title={DeepSeek-V4: Towards Highly Efficient Million-Token Context Intelligence},
      author={DeepSeek-AI},
      year={2026},
}

Contact

If you have any questions, please raise an issue or contact us at [email protected].