Spaces:
Sleeping
Sleeping
Update app.py
Browse files
app.py
CHANGED
|
@@ -10,22 +10,7 @@ import numpy as np
|
|
| 10 |
import plotly.express as px
|
| 11 |
import circuitsvis as cv
|
| 12 |
|
| 13 |
-
|
| 14 |
-
roneneldan/TinyStories-1M
|
| 15 |
-
roneneldan/TinyStories-3M
|
| 16 |
-
roneneldan/TinyStories-8M
|
| 17 |
-
roneneldan/TinyStories-28M
|
| 18 |
-
roneneldan/TinyStories-33M
|
| 19 |
-
roneneldan/TinyStories-1Layer-21M
|
| 20 |
-
roneneldan/TinyStories-2Layers-33M
|
| 21 |
-
roneneldan/TinyStories-Instruct-1M
|
| 22 |
-
roneneldan/TinyStories-Instruct-3M
|
| 23 |
-
roneneldan/TinyStories-Instruct-8M
|
| 24 |
-
roneneldan/TinyStories-Instruct-28M
|
| 25 |
-
roneneldan/TinyStories-Instruct-33M
|
| 26 |
-
roneneldan/TinyStories-Instuct-1Layer-21M
|
| 27 |
-
roneneldan/TinyStories-Instruct-2Layers-33M
|
| 28 |
-
"""
|
| 29 |
|
| 30 |
|
| 31 |
# Little bit of front end for model selector
|
|
@@ -62,6 +47,22 @@ model_name = st.sidebar.radio("Model (only use patching for\nsmall (<4L) models
|
|
| 62 |
model = HookedTransformer.from_pretrained(model_name)
|
| 63 |
|
| 64 |
def predict_next_token(prompt):
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 65 |
logits = model(prompt)[0,-1]
|
| 66 |
answer_index = logits.argmax()
|
| 67 |
answer = model.tokenizer.decode(answer_index)
|
|
|
|
| 10 |
import plotly.express as px
|
| 11 |
import circuitsvis as cv
|
| 12 |
|
| 13 |
+
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 14 |
|
| 15 |
|
| 16 |
# Little bit of front end for model selector
|
|
|
|
| 47 |
model = HookedTransformer.from_pretrained(model_name)
|
| 48 |
|
| 49 |
def predict_next_token(prompt):
|
| 50 |
+
"""
|
| 51 |
+
roneneldan/TinyStories-1M
|
| 52 |
+
roneneldan/TinyStories-3M
|
| 53 |
+
roneneldan/TinyStories-8M
|
| 54 |
+
roneneldan/TinyStories-28M
|
| 55 |
+
roneneldan/TinyStories-33M
|
| 56 |
+
roneneldan/TinyStories-1Layer-21M
|
| 57 |
+
roneneldan/TinyStories-2Layers-33M
|
| 58 |
+
roneneldan/TinyStories-Instruct-1M
|
| 59 |
+
roneneldan/TinyStories-Instruct-3M
|
| 60 |
+
roneneldan/TinyStories-Instruct-8M
|
| 61 |
+
roneneldan/TinyStories-Instruct-28M
|
| 62 |
+
roneneldan/TinyStories-Instruct-33M
|
| 63 |
+
roneneldan/TinyStories-Instuct-1Layer-21M
|
| 64 |
+
roneneldan/TinyStories-Instruct-2Layers-33M
|
| 65 |
+
"""
|
| 66 |
logits = model(prompt)[0,-1]
|
| 67 |
answer_index = logits.argmax()
|
| 68 |
answer = model.tokenizer.decode(answer_index)
|