this post was submitted on 26 Apr 2025
57 points (98.3% liked)

LocalLLaMA

2953 readers
17 users here now

Welcome to LocalLLaMA! Here we discuss running and developing machine learning models at home. Lets explore cutting edge open source neural network technology together.

Get support from the community! Ask questions, share prompts, discuss benchmarks, get hyped at the latest and greatest model releases! Enjoy talking about our awesome hobby.

As ambassadors of the self-hosting machine learning community, we strive to support each other and share our enthusiasm in a positive constructive way.

founded 2 years ago
MODERATORS
[email protected]
[email protected]
[email protected]
[email protected]
[email protected]
57
DeepSeek R2 AI Model Rumors Begin to Swirl Online; Reported to Feature 97% Lower Costs Compared to GPT-4 & Fully Trained on Huawei's Ascend Chips (wccftech.com)
submitted 2 weeks ago by [email protected] to c/[email protected]
3 comments fedilink hide all child comments
you are viewing a single comment's thread
view the rest of the comments
[–] [email protected] 8 points 2 weeks ago* (last edited 2 weeks ago) (1 children)

1.2T param, 78B active, hybrid MoE

That's enormous, very much not local, heh.

Here's the actual article translation (which seems right comparing to other translations):

TranslationDeepSeek R2: Unit Cost Drops 97.3%, Imminent Release + Core Specifications

Author: Chasing Trends Observer
Veteran Crypto Investor Watching from Afar
2025-04-25 12:06:16 Sichuan

Three Core Technological Breakthroughs of DeepSeek R2:

  1. Architectural Innovation
    Adopts proprietary Hybrid MoE 3.0 architecture, achieving 1.2 trillion dynamically activated parameters (actual computational consumption: 78 billion parameters).
    Validated by Alibaba Cloud tests:
  • 97.3% reduction in per-token cost compared to GPT-4 Turbo for long-text inference tasks
    (Data source: IDC Computing Power Economic Model)

  1. Data Engineering
    Constructed 5.2PB high-quality corpus covering finance, law, patents, and vertical domains.
    Multi-stage semantic distillation boosts instruction compliance accuracy to 89.7%
    (Benchmark: C-Eval 2.0 test set)

  2. Hardware Optimization
    Proprietary distributed training framework achieves:

  • 82% utilization rate on Ascend 910B chip clusters
  • 512 PetaFLOPS actual computing power at FP16 precision
  • 91% efficiency of equivalent-scale A100 clusters
    (Validated by Huawei Labs)

Application Layer Advancements - Three Multimodal Breakthroughs:

  1. Vision Understanding
    ViT-Transformer hybrid architecture achieves:
  • 92.4 mAP on COCO dataset object segmentation
  • 11.6% improvement over CLIP models

  1. Industrial Inspection
    Adaptive feature fusion algorithm reduces false detection rate to 7.2E-6 in photovoltaic EL defect detection
    (Field data from LONGi Green Energy production lines)

  2. Medical Diagnostics
    Knowledge graph-enhanced chest X-ray multi-disease recognition:

  • 98.1% accuracy vs. 96.3% average of senior radiologist panels
    (Blind test results from Peking Union Medical College Hospital)

Key Highlight:
8-bit quantization compression achieves:

  • 83% model size reduction
  • <2% accuracy loss
    (Enables edge device deployment - Technical White Paper Chapter 4.2)

Others translate it as 'sub-8-bit' quantization, which is interesting too.

[–] [email protected] 6 points 2 weeks ago (1 children)

I'm sad to see how many mentions of "proprietary" there are in there. I didn't think that was DeepSeek's way of doing things.

[–] [email protected] 4 points 2 weeks ago

The rumor is probably total BS, heh.

That being said, it’s not surprising if Deepseek goes more commercial since it’s basically China’s ChatGPT now.