< Prev | Home | Model | Random | Next >

Model: None

Dataset: The Pile

Neuron 3 in Layer 4

Load this data into an Interactive Neuroscope

See Documentation here

Transformer Lens Loading: HookedTransformer.from_pretrained('pythia-350m')



Text #0

Max Range: 4.3432. Min Range: -4.3432

Max Act: 4.3432. Min Act: -0.1700

Data Index: 9649736 (The Pile)

Max Activating Token Index: 387

Click toggle to see full text

Truncated

 suppose $C_{p}(E) = C(E) = 2$ and such that $\langle P\rangle\subseteq E(\overline{{\mathbb{Q}}})_\text{tors}$ is a ${\mathbb{Q}}$-rational subgroup of order $p$. Both $E_1$ and

Full Text #0

<|endoftext|>ogeny $E\to E/\langle P\rangle$). 
 
1.  The graph consists of one vertex, the curve $E/{\mathbb{Q}}$ does not have any isogenous curves (other than itself), and $E({\mathbb{Q}})_{\text{tors}}$ is necessarily trivial as any torsion point $P\in E({\mathbb{Q}})$ would induce a rational isogeny.  

    0.2in 
 
2.  The case of two vertices, i.e., the case when $C(E) = C_{p}(E) = 2$ and $C_q(E)=1$ for all $q\neq p$, occurs for $p\in \{2, 3, 5, 7, 11, 13, 17, 19, 37, 43, 67, 163\}$:  

    E\_[1]{} \[r\] & E\_[2]{} = E\_1/P  

    Suppose first that $p>7$ and $C_p(E)=2$. If $q$ divides the order of the torsion subgroup of $E_1$ or $E_2$, then $q=p$. Now, Mazur’s theorem \[thm-mazur\] shows that $E_1({\mathbb{Q}})_{\text{tors}}=E_2({\mathbb{Q}})_{\text{tors}}$ must be trivial. Hence, the isogeny-torsion graph $L_2$ is of the form $([1],[1])$.  

    Now let $p$ be $3, 5,$ or $7$, suppose $C_{p}(E) = C(E) = 2$ and such that $\langle P\rangle\subseteq E(\overline{{\mathbb{Q}}})_\text{tors}$ is a ${\mathbb{Q}}$-rational subgroup of order $p$. Both $E_1$ and $E_2$ can have trivial torsion over ${\mathbb{Q}}$, simultaneously, giving $([1],[1])$ as its $L_2$ type. If $P\in E_1({\mathbb{Q}})_{\text{tors}}$, however, by Lemma \[lem-subsequent-rational-pts\], $E_2 = E / \langle P\rangle$ has no rational points of order $p$ and thus, $E_2({\mathbb{Q}})_{\text{tors}}$ is trivial. Thus, the $L_2$-type is $([p],[1])$.  

    Finally, if $C_{2}(E) = C(E) = 2$, then both isogenous curves have torsion subgroups isomorphic to ${\mathbb Z}/ 2 {\mathbb Z}$, by Lemma \[lem-orbit-of-rational-points\] and Lemma \[lem-2torspt-all-have-2torspt\].  

    0.2in 
 
3.  By our work in Section \[sec-whatgraphs\] (see also Cor. \[cor-uniquecyclicQrationalgroup\]), the case of $L_3$, that is, $C(E)=C_p(E)=3$ and $n(P)=p^2$, can only occur for $p=3$ or $5$. In such case, we have  

    E\_[1]{} \[r\] & E\_[2]{} = E\_1/P\[r\] & E\_[3]{}= E\_1/P  

    Since $E_2$ has two independent $p$-isogenies, for $p=3$ or $5$, it follows that the image of $\rho_{E,p}$ is contained in a split Cartan subgroup of ${\operatorname{GL}}(2,{\mathbb F}_p)$. If $E_2$ does not have rational $p$-torsion points, then neither $E_1$ or $E_3$ would have rational torsion points, by Lemma \[lem-subsequent-rational-pts\] (which would imply a rational $p$-torsion point on $E_2$), and the isogeny-torsion graph type is $([1],[1],[1])$. Otherwise, if $E_2$ has a rational $p$-torsion point, then Lemma \[p\^2 isogeny\] implies that either $E_1$ or $E_3$ have a rational point of $p$-torsion (and, clearly, a cyclic rational $p^2$-isogeny). Let us assume that $E_1$ has a rational $p$-torsion point. Since $E_1$ has a unique $p$-isogeny, it follows that $P_1=[p]P$ must be one of the $p$-torsion points defined over ${\mathbb{Q}}$. By Lemma \[lem-subsequent-rational-pts\], we

Text #1

Max Range: 4.3432. Min Range: -4.3432

Max Act: 4.2494. Min Act: -0.1700

Data Index: 938623 (The Pile)

Max Activating Token Index: 972

Click toggle to see full text

Truncated

Full Text #1


Text #2

Max Range: 4.3432. Min Range: -4.3432

Max Act: 4.2494. Min Act: -0.1700

Data Index: 4127425 (The Pile)

Max Activating Token Index: 822

Click toggle to see full text

Truncated

Full Text #2


Text #3

Max Range: 4.3432. Min Range: -4.3432

Max Act: 4.2494. Min Act: -0.1700

Data Index: 670635 (The Pile)

Max Activating Token Index: 787

Click toggle to see full text

Truncated

Full Text #3


Text #4

Max Range: 4.3432. Min Range: -4.3432

Max Act: 4.2494. Min Act: -0.1700

Data Index: 571208 (The Pile)

Max Activating Token Index: 618

Click toggle to see full text

Truncated

Full Text #4


Text #5

Max Range: 4.3432. Min Range: -4.3432

Max Act: 4.2494. Min Act: -0.1700

Data Index: 10275181 (The Pile)

Max Activating Token Index: 445

Click toggle to see full text

Truncated

Full Text #5


Text #6

Max Range: 4.3432. Min Range: -4.3432

Max Act: 4.2494. Min Act: -0.1700

Data Index: 7955703 (The Pile)

Max Activating Token Index: 117

Click toggle to see full text

Truncated

Full Text #6


Text #7

Max Range: 4.3432. Min Range: -4.3432

Max Act: 4.2494. Min Act: -0.1700

Data Index: 4401592 (The Pile)

Max Activating Token Index: 942

Click toggle to see full text

Truncated

Full Text #7


Text #8

Max Range: 4.3432. Min Range: -4.3432

Max Act: 4.2494. Min Act: -0.1700

Data Index: 7105218 (The Pile)

Max Activating Token Index: 435

Click toggle to see full text

Truncated

Full Text #8


Text #9

Max Range: 4.3432. Min Range: -4.3432

Max Act: 4.2181. Min Act: -0.1700

Data Index: 9350237 (The Pile)

Max Activating Token Index: 553

Click toggle to see full text

Truncated

Full Text #9


Text #10

Max Range: 4.3432. Min Range: -4.3432

Max Act: 4.2181. Min Act: -0.1700

Data Index: 4815566 (The Pile)

Max Activating Token Index: 542

Click toggle to see full text

Truncated

Full Text #10


Text #11

Max Range: 4.3432. Min Range: -4.3432

Max Act: 4.2181. Min Act: -0.1700

Data Index: 3393137 (The Pile)

Max Activating Token Index: 775

Click toggle to see full text

Truncated

Full Text #11


Text #12

Max Range: 4.3432. Min Range: -4.3432

Max Act: 4.2181. Min Act: -0.1700

Data Index: 5308171 (The Pile)

Max Activating Token Index: 932

Click toggle to see full text

Truncated

Full Text #12


Text #13

Max Range: 4.3432. Min Range: -4.3432

Max Act: 4.2181. Min Act: -0.1700

Data Index: 1454991 (The Pile)

Max Activating Token Index: 432

Click toggle to see full text

Truncated

Full Text #13


Text #14

Max Range: 4.3432. Min Range: -4.3432

Max Act: 4.2181. Min Act: -0.1700

Data Index: 2100437 (The Pile)

Max Activating Token Index: 605

Click toggle to see full text

Truncated

Full Text #14


Text #15

Max Range: 4.3432. Min Range: -4.3432

Max Act: 4.2181. Min Act: -0.1700

Data Index: 3939988 (The Pile)

Max Activating Token Index: 554

Click toggle to see full text

Truncated

Full Text #15


Text #16

Max Range: 4.3432. Min Range: -4.3432

Max Act: 4.1869. Min Act: -0.1700

Data Index: 5644272 (The Pile)

Max Activating Token Index: 687

Click toggle to see full text

Truncated

Full Text #16


Text #17

Max Range: 4.3432. Min Range: -4.3432

Max Act: 4.1869. Min Act: -0.1700

Data Index: 3862947 (The Pile)

Max Activating Token Index: 512

Click toggle to see full text

Truncated

Full Text #17


Text #18

Max Range: 4.3432. Min Range: -4.3432

Max Act: 4.1869. Min Act: -0.1700

Data Index: 2459670 (The Pile)

Max Activating Token Index: 298

Click toggle to see full text

Truncated

Full Text #18


Text #19

Max Range: 4.3432. Min Range: -4.3432

Max Act: 4.1869. Min Act: -0.1700

Data Index: 6620117 (The Pile)

Max Activating Token Index: 429

Click toggle to see full text

Truncated

Full Text #19