[KongchangAI]
ConceptInduction Head

Induction Head

A type of specialized attention head in Transformers responsible for detecting previously occurring patterns and predicting their subsequent tokens. It is an important basis for in-context learning, but may also lead to repetition loops.

Timeline (last 90 days)

Nov 12

By scanning attention heads, the developer found that an induction head drives the loop, with an induction score of 0.425, while the ordinary copying score is only 0.012.

Unverified50%
Nov 12

An induction head is usually implemented by two attention heads working together: the previous-token head marks each token's predecessor, and the induction head uses that marker to retrieve the successor.

Unverified50%
Nov 12

Induction heads are a special class of attention heads first systematically described by Anthropic's mechanistic interpretability research.

Unverified50%
Nov 12

That induction head is also responsible for copying names into reading-comprehension answers, so simply ablating it would impair reading ability.

Unverified50%
Nov 12

After directly ablating that induction head, the loop rate fell from 29.4% to 25.4%.

Unverified50%
Nov 6

In experiment 3, the researchers removed the input and output layers and kept only four Transformer blocks, and the LLM correctly determined all of the languages the model was trained on (Python, English, and romaji Japanese), and was able to locate the attention head responsible for copying (the induction head).

Unverified50%
Sep 21

Mechanistic interpretability methods have been widely used in recent years to localize specific functional modules such as induction heads and fact retrieval circuits.

Unverified50%
Sep 12

Induction heads are an important clue for understanding the capability of in-context learning.

Unverified50%
Sep 12

Upon seeing [A][B]...[A], induction heads tend to predict that the next token is [B].

Unverified50%
Sep 12

The research identifies induction heads, which can recognize patterns that have appeared previously in a sequence and predict their repetition.

Unverified50%

1 more timeline events

All Facts (11)

Unverified

By scanning attention heads, the developer found that an induction head drives the loop, with an induction score of 0.425, while the ordinary copying score is only 0.012.

50%
Unverified

An induction head is usually implemented by two attention heads working together: the previous-token head marks each token's predecessor, and the induction head uses that marker to retrieve the successor.

50%
Unverified

Induction heads are a special class of attention heads first systematically described by Anthropic's mechanistic interpretability research.

50%
Unverified

That induction head is also responsible for copying names into reading-comprehension answers, so simply ablating it would impair reading ability.

50%
Unverified

After directly ablating that induction head, the loop rate fell from 29.4% to 25.4%.

50%
Unverified

In experiment 3, the researchers removed the input and output layers and kept only four Transformer blocks, and the LLM correctly determined all of the languages the model was trained on (Python, English, and romaji Japanese), and was able to locate the attention head responsible for copying (the induction head).

50%
Unverified

Mechanistic interpretability methods have been widely used in recent years to localize specific functional modules such as induction heads and fact retrieval circuits.

50%
Unverified

Induction heads are an important clue for understanding the capability of in-context learning.

50%
Unverified

Upon seeing [A][B]...[A], induction heads tend to predict that the next token is [B].

50%
Unverified

The research identifies induction heads, which can recognize patterns that have appeared previously in a sequence and predict their repetition.

50%
Unverified

OpenAI的研究团队通过可解释性方法发现了模型内部的归纳头(induction heads)等关键机制

50%

Source Articles