m
DeepSeek mHC
Back to home

mHC Config Generator

Generate training configurations based on your model setup and goals.

Model Configuration

12128

Training Context

LOW RISK
# mHC Training Configuration
# Generated by deepseekmhc.org
# Risk Level: LOW

model:
  size: 7B
  depth: 32
  architecture: dense
  mhc_enabled: true

mhc:
  residual_width_multiplier: 1.2
  constraint_strength: 0.18
  projection_enabled: false

training:
  learning_rate: 0.00023999999999999998
  lr_multiplier: 0.8
  gradient_clip: 1
  warmup_steps: 2000

stability:
  mixed_precision: bf16
  activation_checkpointing: false
  dedicated_compute_stream: false

Learn

← Read mHC Explained

Next

Collapse Diagnostics →