Spaces:
Sleeping
Sleeping
John Graham Reynolds
commited on
Commit
·
3102077
1
Parent(s):
8673ccf
typo and model title
Browse files
app.py
CHANGED
@@ -17,14 +17,14 @@ EXAMPLE_PROMPTS = [
|
|
17 |
"Solve 49*l + 45*l - 125 - 63 = 0 for l.",
|
18 |
]
|
19 |
|
20 |
-
TITLE = "CyberSolve LinAlg 1.2"
|
21 |
DESCRIPTION= """Welcome to the CyberSolve LinAlg 1.2 demo! \n
|
22 |
|
23 |
**Overview and Usage**: This 🤗 Space is designed to demo the abilities of the **CyberSolve LinAlg 1.2** text-to-text language model.
|
24 |
|
25 |
Specifically, the **CyberSolve LinAlg 1.x** family of models
|
26 |
are downstream versions of the 783M parameter FLAN-T5 text-to-text transformer, fine-tuned on the Google DeepMind Mathematics dataset for the purpose of solving linear equations of a single variable.
|
27 |
-
To effectively query the model for its intended task, prompt the model solve an arbitrary linear equation of a single variable with a query of the form: *"Solve 24 = 1601c - 1605c for c."*; the model
|
28 |
will return its prediciton in a simple format. The algebraic capabailites of CyberSolve far exceed those of the base FLAN-T5 model. CyberSolve LinAlg 1.2 achieves a 90.7 percent exact match benchmark
|
29 |
on the DeepMind Mathematics evaluation dataset of 10,000 unique linear equations; the FLAN-T5 base model scores 9.6 percent.
|
30 |
|
|
|
17 |
"Solve 49*l + 45*l - 125 - 63 = 0 for l.",
|
18 |
]
|
19 |
|
20 |
+
TITLE = "🤖CyberSolve LinAlg 1.2🧠"
|
21 |
DESCRIPTION= """Welcome to the CyberSolve LinAlg 1.2 demo! \n
|
22 |
|
23 |
**Overview and Usage**: This 🤗 Space is designed to demo the abilities of the **CyberSolve LinAlg 1.2** text-to-text language model.
|
24 |
|
25 |
Specifically, the **CyberSolve LinAlg 1.x** family of models
|
26 |
are downstream versions of the 783M parameter FLAN-T5 text-to-text transformer, fine-tuned on the Google DeepMind Mathematics dataset for the purpose of solving linear equations of a single variable.
|
27 |
+
To effectively query the model for its intended task, prompt the model to solve an arbitrary linear equation of a single variable with a query of the form: *"Solve 24 = 1601c - 1605c for c."*; the model
|
28 |
will return its prediciton in a simple format. The algebraic capabailites of CyberSolve far exceed those of the base FLAN-T5 model. CyberSolve LinAlg 1.2 achieves a 90.7 percent exact match benchmark
|
29 |
on the DeepMind Mathematics evaluation dataset of 10,000 unique linear equations; the FLAN-T5 base model scores 9.6 percent.
|
30 |
|