Update evaluation_stats.json via script
Browse files- evaluation_stats.json +7 -7
evaluation_stats.json
CHANGED
|
@@ -1,6 +1,5 @@
|
|
| 1 |
{
|
| 2 |
"perplexity_history": [
|
| 3 |
-
291.7740173339844,
|
| 4 |
291.7562255859375,
|
| 5 |
291.74176025390625,
|
| 6 |
291.7083740234375,
|
|
@@ -19,7 +18,8 @@
|
|
| 19 |
291.44378662109375,
|
| 20 |
289.20306396484375,
|
| 21 |
288.07843017578125,
|
| 22 |
-
287.02593994140625
|
|
|
|
| 23 |
],
|
| 24 |
"last_examples": {
|
| 25 |
"Story Continuation": "How do I get the best picture of what we would want?\nI am getting a lot of awesome photos that can be amazing! Thanks for sharing!!!\nHi guys! We were happy!!! Thanks for all!!\nHave you ever gotten into this photo!!\nThanks so much luck!!\nThank you so much!!!!!!!!!!\nHi, I'",
|
|
@@ -27,11 +27,11 @@
|
|
| 27 |
"Creative Prompt": "We are pleased to announce the launch of the first prototype projector is bringing us to Mars Earth.\nThe mission is to be launching our new spacecrafts, which will come up with the launch of Marscraftia SpaceXicon Spacecraft and asteroid orbital spacecraft.\nWe have a mission to make Earthcraft asteroids through NASA Earth Day\u2122",
|
| 28 |
"Question Answering (Basic)": "As an example we've been using the same pattern threaded threading patches of new threads that can be removed from thread threads. These thread threads were used to thread thread thread threads removed them onto thread threads or thread threads. It is threadbare threads thread threads thread thread thread threads for thread thread thread threads when thread thread threads thread thread threadbone thread thread threads thread",
|
| 29 |
"Code Generation (Simple Python)": "We are looking forward to seeing us in the future today! This is one of our favourite gameplaying gameplayers! We have been lucky enough to have the opportunity to find new players who want to join them as well. This will be the first time we get back to you when you wish you were able to join us on this awesome journey so much easier",
|
| 30 |
-
"Reasoning (Simple)": "The time of day we were trying to get our latest updates. It\u2019s been amazed because we are going to see all the way
|
| 31 |
},
|
| 32 |
-
"last_update_time": "2025-05-08 15:
|
| 33 |
-
"datasets_processed_count":
|
| 34 |
-
"texts_processed_count":
|
| 35 |
-
"tokens_processed_count":
|
| 36 |
"lighteval_results": {}
|
| 37 |
}
|
|
|
|
| 1 |
{
|
| 2 |
"perplexity_history": [
|
|
|
|
| 3 |
291.7562255859375,
|
| 4 |
291.74176025390625,
|
| 5 |
291.7083740234375,
|
|
|
|
| 18 |
291.44378662109375,
|
| 19 |
289.20306396484375,
|
| 20 |
288.07843017578125,
|
| 21 |
+
287.02593994140625,
|
| 22 |
+
286.1083679199219
|
| 23 |
],
|
| 24 |
"last_examples": {
|
| 25 |
"Story Continuation": "How do I get the best picture of what we would want?\nI am getting a lot of awesome photos that can be amazing! Thanks for sharing!!!\nHi guys! We were happy!!! Thanks for all!!\nHave you ever gotten into this photo!!\nThanks so much luck!!\nThank you so much!!!!!!!!!!\nHi, I'",
|
|
|
|
| 27 |
"Creative Prompt": "We are pleased to announce the launch of the first prototype projector is bringing us to Mars Earth.\nThe mission is to be launching our new spacecrafts, which will come up with the launch of Marscraftia SpaceXicon Spacecraft and asteroid orbital spacecraft.\nWe have a mission to make Earthcraft asteroids through NASA Earth Day\u2122",
|
| 28 |
"Question Answering (Basic)": "As an example we've been using the same pattern threaded threading patches of new threads that can be removed from thread threads. These thread threads were used to thread thread thread threads removed them onto thread threads or thread threads. It is threadbare threads thread threads thread thread thread threads for thread thread thread threads when thread thread threads thread thread threadbone thread thread threads thread",
|
| 29 |
"Code Generation (Simple Python)": "We are looking forward to seeing us in the future today! This is one of our favourite gameplaying gameplayers! We have been lucky enough to have the opportunity to find new players who want to join them as well. This will be the first time we get back to you when you wish you were able to join us on this awesome journey so much easier",
|
| 30 |
+
"Reasoning (Simple)": "The time of day we were trying to get our latest updates. It\u2019s been amazed because we are going to see all the way up! It is my journey to find us in a safe place for people who love us!!\nI am not happy that you have had to be lucky enough to try and improve yourself but I am very grateful to myself for taking"
|
| 31 |
},
|
| 32 |
+
"last_update_time": "2025-05-08 15:40:27 UTC",
|
| 33 |
+
"datasets_processed_count": 4,
|
| 34 |
+
"texts_processed_count": 24,
|
| 35 |
+
"tokens_processed_count": 12288,
|
| 36 |
"lighteval_results": {}
|
| 37 |
}
|