wassname/Brukino_AntiPaSTO_Appetizer
1
0
Fork 0
mirror of https://github.com/wassname/Brukino_AntiPaSTO_Appetizer.git synced 2026-06-27 17:13:50 +08:00
Code Issues Packages Projects Releases Wiki Activity

Increase S-space top_k to 1024

Browse Source
This commit is contained in:
wassname
2026-04-10 09:52:24 +08:00
parent 382ffc4315
commit e44dc0e74e
2 changed files with 7 additions and 7 deletions
Download Patch File Download Diff File Expand all files Collapse all files
experiment.ipynb
+6 -6
View File
@@ -2,7 +2,7 @@
"cells": [ "cells": [
{ {
"cell_type": "markdown", "cell_type": "markdown",
"id": "b8a288c6", "id": "62eec772",
"metadata": {}, "metadata": {},
"source": [ "source": [
"# Brukino's AntiPaSTO Appetizer: Guided CoT Eval & Frenet-Serret Curvature\n", "# Brukino's AntiPaSTO Appetizer: Guided CoT Eval & Frenet-Serret Curvature\n",
@@ -20,7 +20,7 @@
{ {
"cell_type": "code", "cell_type": "code",
"execution_count": null, "execution_count": null,
"id": "e8c081e6", "id": "47c7efe2",
"metadata": {}, "metadata": {},
"outputs": [], "outputs": [],
"source": [ "source": [
@@ -44,11 +44,11 @@
{ {
"cell_type": "code", "cell_type": "code",
"execution_count": null, "execution_count": null,
"id": "7b5b34e4", "id": "8c590e6c",
"metadata": {}, "metadata": {},
"outputs": [], "outputs": [],
"source": [ "source": [
"def get_s_space_projector(model, top_k=256):\n", "def get_s_space_projector(model, top_k=1024):\n",
" \"\"\"\n", " \"\"\"\n",
" Gathers all weight matrices that write to the residual stream\n", " Gathers all weight matrices that write to the residual stream\n",
" (o_proj from attention and down_proj from MLP) across all layers,\n", " (o_proj from attention and down_proj from MLP) across all layers,\n",
@@ -123,7 +123,7 @@
{ {
"cell_type": "code", "cell_type": "code",
"execution_count": null, "execution_count": null,
"id": "ab5130a9", "id": "6ae905b2",
"metadata": {}, "metadata": {},
"outputs": [], "outputs": [],
"source": [ "source": [
@@ -205,7 +205,7 @@
{ {
"cell_type": "code", "cell_type": "code",
"execution_count": null, "execution_count": null,
"id": "96c84bd3", "id": "30e7fb4e",
"metadata": { "metadata": {
"lines_to_next_cell": 2 "lines_to_next_cell": 2
}, },
experiment.py
+1 -1
View File
@@ -39,7 +39,7 @@ NUM_EXAMPLES = 3
# %% # %%
def get_s_space_projector(model, top_k=256): def get_s_space_projector(model, top_k=1024):
""" """
Gathers all weight matrices that write to the residual stream Gathers all weight matrices that write to the residual stream
(o_proj from attention and down_proj from MLP) across all layers, (o_proj from attention and down_proj from MLP) across all layers,