AI Infra Interviews logo

ERNIE / Lab profile

Baidu

A base release opens an adaptation path.

ERNIE 5.5 Flash Base is a pretrained text checkpoint. Its training stage and custom license are first-order selection constraints before any serving comparison.

Architecture trajectory

What connects these models?

The 43-layer backbone mixes 37 hierarchical compressed-attention layers with six sparse latent-attention layers. Compression applies to particular representations, not automatically to all cache allocations.

Compression applies to a particular state representation.

  1. HCA layer37 backbone layers
    Compressed history128-token grouping
  2. MLA-DSA layer6 backbone layers
    Latent historyselection and retained state
HCA layer37 backbone layersCompressed history128-token groupingMLA-DSA layer6 backbone layersLatent historyselection and retainedstate

Sparse reads do not imply proportional storage savings.

Sum both layer families before adding runtime overhead.

ERNIE 5.5 Flash Base backbone schematic. Thirty-seven HCA layers compress groups of token history; six MLA-DSA layers use latent history and sparse selection. ECHO and runtime buffers add separate state. This is not a literal execution ordering.

Newest first / grouped sequence, not a time scale

The ERNIE timeline

  1. ERNIE-5.5-Flash-Base

    Start with the training stage and license

    Thirty-seven hierarchical compressed-attention layers group history at a 128-token ratio; six MLA-DSA layers use a different retained representation. The whole cache does not shrink by 128 times.

    The applicable model license identifies November 5 as its release date; the repository existed earlier.

    Release evidence ↗

Which other labs should you compare?

DeepSeek and Ling are useful architectural comparisons, but their post-trained checkpoints serve a different starting point for application development.

DeepSeek V4 / V4.1

DeepSeek

Flash evolves from a post-training update into a different memory system.

Ling

InclusionAI

Hybrid state and checkpoint identity decide the rollout.

Interview preparation

What you should be able to explain

Distinguish total weights, active computation, extra ECHO parameters, training context and configured positions.

Study the underlying systems → · Get the illustrated report →

Sources and scope

Covered checkpoints carry their own research dates. Assessments and comparison candidates are our interpretation of the cited designs, not measured cross-lab rankings. Each timeline entry links to its release evidence; model pages preserve pinned card and configuration sources.