T5-v1.1 is a sequence-to-sequence language model variant used as Atlas’s reader. Unlike the original T5, it is pretrained only on unlabeled text, avoiding supervised-dataset contamination that could overlap with downstream evaluation tasks. section 6.1
Within retrieval-augmented inference, T5-v1.1 provides the reader stage:
- Receives documents selected by a dual-encoder retriever. section 6.1
- Generates responses as part of a modular system connected to an external knowledge base. section 6.1
- Supports more credible few-shot evaluation by excluding supervised pretraining data that might overlap with evaluation tasks. section 6.1
Table 6.1 Parametric vs non-parametric knowledge
|
Feature
|
Parametric Knowledge (e.g., GPT-3)
|
Non-Parametric Knowledge (e.g., Atlas)
|
|
Knowledge Freshness
|
Static; frozen at the time of training.
|
Dynamic; updatable in near real-time.
|
|
Verifiability
|
Opaque; "black box" reasoning.
|
Transparent; responses grounded in citable sources.
|
|
Cost to Update
|
Extremely high (full model retraining).
|
Low (re-index external corpus).
|
|
Domain Specialization
|
Requires expensive fine-tuning or prompting.
|
Achieved by curating the external corpus.
|
|
Hallucination Risk
|
High when knowledge is outdated or missing.
|
Low; the model can state that knowledge is not found.
|
|
Failure Mode
|
Factual hallucination when knowledge is missing/outdated.
|
Retrieval failure; can return "knowledge not
found."
|
Compare T5-v1.1’s reader role with the retriever and the broader tradeoffs between frozen parametric knowledge and updateable non-parametric knowledge. section 6.1
