Add official IFEval summary JSON
Browse files- ifeval-summary.json +29 -0
ifeval-summary.json
ADDED
|
@@ -0,0 +1,29 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"model_run": "/home/empix/llm-runs/tercet-sft",
|
| 3 |
+
"checkpoint": "/home/empix/llm-runs/tercet-sft/checkpoint-00012702",
|
| 4 |
+
"checkpoint_name": "checkpoint-00012702",
|
| 5 |
+
"weights": "ema",
|
| 6 |
+
"device": "cuda",
|
| 7 |
+
"prompts": 541,
|
| 8 |
+
"max_new_tokens": 1280,
|
| 9 |
+
"temperature": 0.0,
|
| 10 |
+
"created_at": "2026-08-26T11:00:49Z",
|
| 11 |
+
"metrics": {
|
| 12 |
+
"strict": {
|
| 13 |
+
"prompt_level_accuracy": 0.1922365988909427,
|
| 14 |
+
"instruction_level_accuracy": 0.328537170263789,
|
| 15 |
+
"prompts_passed": 104,
|
| 16 |
+
"prompts_total": 541,
|
| 17 |
+
"instructions_passed": 274,
|
| 18 |
+
"instructions_total": 834
|
| 19 |
+
},
|
| 20 |
+
"loose": {
|
| 21 |
+
"prompt_level_accuracy": 0.21256931608133087,
|
| 22 |
+
"instruction_level_accuracy": 0.3501199040767386,
|
| 23 |
+
"prompts_passed": 115,
|
| 24 |
+
"prompts_total": 541,
|
| 25 |
+
"instructions_passed": 292,
|
| 26 |
+
"instructions_total": 834
|
| 27 |
+
}
|
| 28 |
+
}
|
| 29 |
+
}
|