Model card, license notes, training records
Browse files- LICENSE.md +14 -0
- README.md +306 -0
- training/data_manifest.json +109 -0
- training/laya.yaml +211 -0
- training/run_stats.json +759 -0
- training/train_log.jsonl +384 -0
LICENSE.md
ADDED
|
@@ -0,0 +1,14 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
# Lisans notları
|
| 2 |
+
|
| 3 |
+
Bu model `convaiinnovations/laya-multilingual` (Apache-2.0) üzerine aşağıdaki verilerle eğitildi. Model ağırlıklarının
|
| 4 |
+
kullanım koşulları bu kaynakların koşullarından etkilenebilir.
|
| 5 |
+
|
| 6 |
+
| Kaynak | Kullanım | Hugging Face kartındaki lisans |
|
| 7 |
+
|---|---|---|
|
| 8 |
+
| convaiinnovations/laya-multilingual | taban model | Apache-2.0 |
|
| 9 |
+
| avbiswas/bev-decision-150K (ve Türkçesi hayriyigit/bev-decision-150K-tr) | eğitim + test | belirtilmemiş ("unknown") |
|
| 10 |
+
| mteb/amazon_massive_intent (MASSIVE tr-TR / en-US) | eğitim + test | Apache-2.0 (orijinal MASSIVE: CC BY 4.0) |
|
| 11 |
+
| facebook/xnli (tr / en) | eğitim (validation) + test | kartta belirtilmemiş; orijinal XNLI lisansını kontrol edin |
|
| 12 |
+
| turkish-nlp-suite/beyazperde-all-movie-reviews | eğitim + test | CC BY-SA 4.0 |
|
| 13 |
+
| Praveenrajus/jev-bench (ve Türkçesi hayriyigit/jev-bench-tr) | yalnızca test | other |
|
| 14 |
+
| rajpurkar/squad_v2 (ve Türkçesi hayriyigit/squad_v2-tr) | eğitim + test | CC BY-SA 4.0 |
|
README.md
ADDED
|
@@ -0,0 +1,306 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
---
|
| 2 |
+
language:
|
| 3 |
+
- tr
|
| 4 |
+
- en
|
| 5 |
+
license: other
|
| 6 |
+
license_name: mixed-training-data
|
| 7 |
+
license_link: LICENSE.md
|
| 8 |
+
base_model: convaiinnovations/laya-multilingual
|
| 9 |
+
pipeline_tag: text-classification
|
| 10 |
+
library_name: laya
|
| 11 |
+
tags:
|
| 12 |
+
- laya
|
| 13 |
+
- decision-model
|
| 14 |
+
- calibration
|
| 15 |
+
- turkish
|
| 16 |
+
- choice
|
| 17 |
+
- noul
|
| 18 |
+
- score
|
| 19 |
+
datasets:
|
| 20 |
+
- hayriyigit/bev-decision-150K-tr
|
| 21 |
+
- avbiswas/bev-decision-150K
|
| 22 |
+
- mteb/amazon_massive_intent
|
| 23 |
+
- facebook/xnli
|
| 24 |
+
- turkish-nlp-suite/beyazperde-all-movie-reviews
|
| 25 |
+
- hayriyigit/jev-bench-tr
|
| 26 |
+
- hayriyigit/squad_v2-tr
|
| 27 |
+
- rajpurkar/squad_v2
|
| 28 |
+
---
|
| 29 |
+
|
| 30 |
+
# Laya TR v2: Türkçe karar modeli (cevaplanabilirlik eklendi)
|
| 31 |
+
|
| 32 |
+
**English summary.** A Turkish fine-tune of [`convaiinnovations/laya-multilingual`](https://huggingface.co/convaiinnovations/laya-multilingual),
|
| 33 |
+
the non-autoregressive "System 1" decision model of [Laya](https://github.com/NandhaKishorM/laya): it answers typed
|
| 34 |
+
`choice`, `noul` (yes/no) and `score` questions about a text in a single forward pass and returns calibrated
|
| 35 |
+
probabilities. Trained for 2 epochs on 305,812 decisions (machine-translated bev-decision-150K, its English
|
| 36 |
+
originals, machine-translated SQuAD 2.0 for answerability, and native Turkish MASSIVE, XNLI and movie reviews). On 101,716 held-out test decisions accuracy goes
|
| 37 |
+
from 0.443 to 0.686 and ECE from 0.282 to 0.083. Compared with
|
| 38 |
+
[`hayriyigit/laya-tr-test`](https://huggingface.co/hayriyigit/laya-tr-test), answerability on the Turkish SQuAD 2.0 dev set
|
| 39 |
+
(can the document answer the question?) improves from 0.565 to 0.703 accuracy
|
| 40 |
+
(AUROC 0.596 → 0.794); the other test sets stay within about one point.
|
| 41 |
+
|
| 42 |
+
Bu model, [Laya](https://github.com/NandhaKishorM/laya)'nın çok dilli kontrol noktası
|
| 43 |
+
[`convaiinnovations/laya-multilingual`](https://huggingface.co/convaiinnovations/laya-multilingual) (mmBERT tabanlı) üzerine
|
| 44 |
+
Türkçe karar verisiyle ince ayar yapılmış halidir. [`hayriyigit/laya-tr-test`](https://huggingface.co/hayriyigit/laya-tr-test)'in ikinci sürümüdür: aynı veriye
|
| 45 |
+
SQuAD 2.0'ın Türkçe çevirisi eklendi, böylece model bir belgenin bir soruyu gerçekten cevaplayıp cevaplayamadığına
|
| 46 |
+
(cevaplanabilirlik) daha iyi karar veriyor. Laya metin üretmez: bir metin (state) ve tipli sorular alır, her soru
|
| 47 |
+
için tek ileri geçişte kalibre edilmiş olasılıklar döndürür.
|
| 48 |
+
|
| 49 |
+
- **choice:** seçeneklerden biri (ör. hangi departman, hangi niyet),
|
| 50 |
+
- **noul:** bir ifadenin doğru olma olasılığı (evet/hayır),
|
| 51 |
+
- **score:** sıralı seviyelerden biri (ör. 1-5 memnuniyet).
|
| 52 |
+
|
| 53 |
+
## Kullanım
|
| 54 |
+
|
| 55 |
+
```python
|
| 56 |
+
import laya
|
| 57 |
+
|
| 58 |
+
agent = laya.load("hayriyigit/laya-tr-test-v2", token="hf_...") # private repo: okuma yetkili bir token
|
| 59 |
+
|
| 60 |
+
state = "Siparişim iki haftadır gelmedi, kargo takip numarası da çalışmıyor. Paramı geri istiyorum."
|
| 61 |
+
questions = {
|
| 62 |
+
"birim": {
|
| 63 |
+
"type": "choice",
|
| 64 |
+
"instructions": "Bu talebi hangi birim ele almalı?",
|
| 65 |
+
"criteria": {"fatura": "ödeme, iade, fatura", "kargo": "teslimat ve takip", "teknik": "uygulama veya site hataları"},
|
| 66 |
+
},
|
| 67 |
+
"iade": {"type": "noul", "instructions": "Müşteri para iadesi istiyor."},
|
| 68 |
+
"memnuniyet": {
|
| 69 |
+
"type": "score",
|
| 70 |
+
"instructions": "Müşteri ne kadar memnun?",
|
| 71 |
+
"criteria": ["Çok memnuniyetsiz", "Memnuniyetsiz", "Nötr", "Memnun", "Çok memnun"],
|
| 72 |
+
},
|
| 73 |
+
}
|
| 74 |
+
answers = agent.predict(state, questions)["answers"]
|
| 75 |
+
```
|
| 76 |
+
|
| 77 |
+
Bu modelin bu örnekteki cevabı:
|
| 78 |
+
|
| 79 |
+
| Soru | Cevap | Olasılık |
|
| 80 |
+
|---|---|---:|
|
| 81 |
+
| birim | `kargo` | 0,958 |
|
| 82 |
+
| iade | evet | 0,909 |
|
| 83 |
+
| memnuniyet | Çok memnuniyetsiz (seviye 0) | 0,791 |
|
| 84 |
+
|
| 85 |
+
Her cevapta tüm seçeneklerin olasılıkları (`probabilities`) ve eşik koymak için `answer_confidence` da döner.
|
| 86 |
+
|
| 87 |
+
Soru başına bağlam 8.192 token (soru ve seçenekler için 768); kısa girdiler yalnızca kendi uzunlukları kadar
|
| 88 |
+
işlenir. Sıcaklıklar (choice, score, noul) eğitimde görülmemiş Türkçe kararlarda kalibre edildi: 0,898, 1,078, 0,950.
|
| 89 |
+
|
| 90 |
+
## Değerlendirme
|
| 91 |
+
|
| 92 |
+
Test kararlarının hiçbiri eğitimde görülmedi; metni bir test satırında da geçen eğitim satırları, İngilizce orijinaller
|
| 93 |
+
üzerinden eşleştirilerek eğitimden çıkarıldı (aşağıda). "Önce" sütunu ince ayar öncesi `laya-multilingual`'dır.
|
| 94 |
+
|
| 95 |
+
### Kaynağa göre
|
| 96 |
+
|
| 97 |
+
| Test seti | Karar | Doğruluk (önce) | Doğruluk (sonra) | Fark (puan) | ECE (önce) | ECE (sonra) |
|
| 98 |
+
|---|---:|---:|---:|---:|---:|---:|
|
| 99 |
+
| bev-decision-150K-tr test (makine çevirisi) | 46.320 | 0,444 | **0,774** | +33,1 | 0,275 | **0,049** |
|
| 100 |
+
| bev-decision-150K test, İngilizce orijinal | 5.766 | 0,473 | **0,772** | +29,9 | 0,243 | **0,068** |
|
| 101 |
+
| jev-bench-tr (benchmark, makine çevirisi) | 22.773 | 0,402 | **0,496** | +9,4 | 0,295 | **0,175** |
|
| 102 |
+
| jev-bench, İngilizce orijinal | 2.000 | 0,499 | **0,584** | +8,4 | 0,246 | **0,109** |
|
| 103 |
+
| MASSIVE tr-TR (ana dili Türkçe) | 2.974 | 0,353 | **0,773** | +42,0 | 0,286 | **0,069** |
|
| 104 |
+
| XNLI Türkçe test (insan çevirisi) | 5.010 | 0,691 | **0,733** | +4,2 | 0,212 | **0,097** |
|
| 105 |
+
| beyazperde film yorumları (ana dili Türkçe) | 5.000 | 0,180 | **0,533** | +35,4 | 0,330 | **0,119** |
|
| 106 |
+
| SQuAD 2.0 dev, cevaplanabilirlik (makine çevirisi) | 11.873 | 0,520 | **0,703** | +18,3 | 0,319 | **0,133** |
|
| 107 |
+
| **Hepsi** | **101.716** | 0,443 | **0,686** | +24,3 | 0,282 | **0,083** |
|
| 108 |
+
|
| 109 |
+
### Soru türüne göre
|
| 110 |
+
|
| 111 |
+
| Tür | Karar | Doğruluk (önce) | Doğruluk (sonra) | ECE (önce) | ECE (sonra) |
|
| 112 |
+
|---|---:|---:|---:|---:|---:|
|
| 113 |
+
| choice | 39.616 | 0,448 | **0,660** | 0,247 | **0,098** |
|
| 114 |
+
| noul | 39.371 | 0,575 | **0,797** | 0,292 | **0,060** |
|
| 115 |
+
| score | 22.729 | 0,204 | **0,538** | 0,325 | **0,108** |
|
| 116 |
+
|
| 117 |
+
### Cevaplanabilirlik (SQuAD 2.0 dev, Türkçe)
|
| 118 |
+
|
| 119 |
+
Soru: *"Verilen `document`, `question` sorusunun TÜM kısımlarını doğru ve eksiksiz şekilde cevaplamak için yeterli bilgiyi
|
| 120 |
+
açıkça içeriyor mu? …"* (noul). State `{"question": …, "document": …}`. 11.873 soru, yarısı cevaplanamaz; dev ile
|
| 121 |
+
train farklı Vikipedi makalelerinden gelir.
|
| 122 |
+
|
| 123 |
+
| Model | Doğruluk | AUROC | Cevaplanabilirde doğru | Cevaplanamazda doğru | Ort. P(evet): cevaplanabilir / cevaplanamaz |
|
| 124 |
+
|---|---:|---:|---:|---:|---:|
|
| 125 |
+
| Temel model | 0,520 | 0,540 | 0,858 | 0,182 | 0,785 / 0,749 |
|
| 126 |
+
| laya-tr-test (v1) | 0,565 | 0,596 | 0,473 | 0,657 | 0,496 / 0,401 |
|
| 127 |
+
| **Bu model (v2)** | 0,703 | 0,794 | 0,869 | 0,537 | 0,783 / 0,465 |
|
| 128 |
+
|
| 129 |
+
Model artık iki tür soruyu ayırt ediyor, ama hâlâ kolay "evet" diyor: cevaplanamaz soruların yaklaşık yarısını yakalıyor.
|
| 130 |
+
Eşik 0,5 yerine 0,790 alınırsa doğruluk 0,722 olur. Cevaplanabilirlik kararlarında yüksek bir eşik kullanın.
|
| 131 |
+
|
| 132 |
+
### Önceki sürümle karşılaştırma
|
| 133 |
+
|
| 134 |
+
| Test seti | Doğruluk (v1) | Doğruluk (v2) | ECE (v1) | ECE (v2) |
|
| 135 |
+
|---|---:|---:|---:|---:|
|
| 136 |
+
| bev-decision-150K-tr test (makine çevirisi) | 0,771 | 0,774 | 0,031 | 0,049 |
|
| 137 |
+
| bev-decision-150K test, İngilizce orijinal | 0,775 | 0,772 | 0,036 | 0,068 |
|
| 138 |
+
| jev-bench-tr (benchmark, makine çevirisi) | 0,490 | 0,496 | 0,171 | 0,175 |
|
| 139 |
+
| jev-bench, İngilizce orijinal | 0,578 | 0,584 | 0,105 | 0,109 |
|
| 140 |
+
| MASSIVE tr-TR (ana dili Türkçe) | 0,781 | 0,773 | 0,063 | 0,069 |
|
| 141 |
+
| XNLI Türkçe test (insan çevirisi) | 0,738 | 0,733 | 0,095 | 0,097 |
|
| 142 |
+
| beyazperde film yorumları (ana dili Türkçe) | 0,531 | 0,533 | 0,067 | 0,119 |
|
| 143 |
+
| SQuAD 2.0 dev, cevaplanabilirlik (makine çevirisi) | – | 0,703 | – | 0,133 |
|
| 144 |
+
|
| 145 |
+
v1'in test setinde SQuAD yoktu; SQuAD için v1 değeri yukarıdaki tablodadır (aynı soru ve aynı dev seti). Toplam ECE v1'de
|
| 146 |
+
0,070, v2'de 0,083: sıcaklıklar yalnızca 1.571 kararlık bir doğrulama diliminde ayarlandığı için
|
| 147 |
+
sonuç gürültülüdür; en çok `score` sorularının kalibrasyonu kötüleşti.
|
| 148 |
+
|
| 149 |
+
### Nasıl okunmalı
|
| 150 |
+
|
| 151 |
+
- **Kalibrasyon** en büyük kazanç: ECE 0,282 → 0,083. Modelin verdiği olasılıklar eşik koymak için
|
| 152 |
+
kullanılabilir hale geldi.
|
| 153 |
+
- **İngilizce unutulmadı**: eğitime İngilizce orijinallerin ~%25'i karıştırıldı; bev ve jev-bench'in İngilizce testlerinde de
|
| 154 |
+
doğruluk arttı. Diğer diller ölçülmedi.
|
| 155 |
+
- **En bağımsız sinyal XNLI** (+4,2 puan): ayrı kaynaklı, insan çevirisi bir test. MASSIVE ve beyazperde'deki
|
| 156 |
+
büyük artışlar kısmen bu kaynakların train bölümlerinin eğitimde olmasından gelir.
|
| 157 |
+
- **noul doğruluğu** (0,797) v1'deki değerden (0,823) düşük görünür; nedeni test setine SQuAD'ın zor
|
| 158 |
+
cevaplanabilirlik sorularının eklenmesidir, iki değer aynı test setinden gelmez.
|
| 159 |
+
- **Bazı jev-bench alt setleri kötüleşti:** civil_comments −5,7 puan. Nedeni ayrıca incelenmedi.
|
| 160 |
+
- **jev-bench** bev ile aynı görev ailelerinden (ör. MASSIVE, CLINC150, SST-5, BoolQ) beslenir; birebir aynı metinler
|
| 161 |
+
çıkarıldı, ama tamamen bağımsız bir test değildir.
|
| 162 |
+
|
| 163 |
+
### bev-decision alanları (Türkçe test)
|
| 164 |
+
|
| 165 |
+
| Alan | Karar | Doğruluk (önce) | Doğruluk (sonra) | ECE (sonra) |
|
| 166 |
+
|---|---:|---:|---:|---:|
|
| 167 |
+
| Sentiment, emotion, and moderation | 11.526 | 0,616 | **0,860** | 0,052 |
|
| 168 |
+
| Spatial and logical reasoning | 8.384 | 0,321 | **0,596** | 0,063 |
|
| 169 |
+
| Retail, product, and shopping | 6.601 | 0,288 | **0,929** | 0,067 |
|
| 170 |
+
| Response preference and quality | 3.176 | 0,321 | **0,561** | 0,071 |
|
| 171 |
+
| Support and intent routing | 2.858 | 0,678 | **0,969** | 0,053 |
|
| 172 |
+
| Tool and workflow decisions | 2.307 | 0,769 | **0,984** | 0,063 |
|
| 173 |
+
| Software security | 2.240 | 0,417 | **0,666** | 0,080 |
|
| 174 |
+
| Game-state decisions | 2.039 | 0,345 | **0,858** | 0,066 |
|
| 175 |
+
| Software engineering and code | 2.039 | 0,322 | **0,608** | 0,141 |
|
| 176 |
+
| Financial reporting and banking | 1.740 | 0,166 | **0,603** | 0,031 |
|
| 177 |
+
| Reading comprehension | 1.422 | 0,628 | **0,712** | 0,123 |
|
| 178 |
+
| Scientific and paper understanding | 508 | 0,374 | **0,848** | 0,058 |
|
| 179 |
+
| Contract evidence | 320 | 0,334 | **0,725** | 0,080 |
|
| 180 |
+
| Browser interaction | 286 | 0,273 | **0,724** | 0,133 |
|
| 181 |
+
| Spam detection | 271 | 0,550 | **0,956** | 0,079 |
|
| 182 |
+
| Civic and safety operations | 261 | 0,253 | **0,628** | 0,092 |
|
| 183 |
+
| Sensory quality rating | 260 | 0,073 | **0,454** | 0,091 |
|
| 184 |
+
| Engagement and ranking | 82 | 0,524 | **0,622** | 0,202 |
|
| 185 |
+
|
| 186 |
+
### jev-bench-tr alt setleri
|
| 187 |
+
|
| 188 |
+
| Alt set | Karar | Doğruluk (önce) | Doğruluk (sonra) |
|
| 189 |
+
|---|---:|---:|---:|
|
| 190 |
+
| arc_challenge | 1.000 | 0,251 | **0,249** |
|
| 191 |
+
| banking77 | 1.000 | 0,270 | **0,319** |
|
| 192 |
+
| boolq | 1.000 | 0,687 | **0,711** |
|
| 193 |
+
| chaosnli | 1.599 | 0,478 | **0,546** |
|
| 194 |
+
| civil_comments | 2.000 | 0,913 | **0,857** |
|
| 195 |
+
| clinc150 | 1.000 | 0,249 | **0,390** |
|
| 196 |
+
| fever_evidence | 1.000 | 0,551 | **0,813** |
|
| 197 |
+
| go_emotions | 1.000 | 0,277 | **0,307** |
|
| 198 |
+
| helpsteer2_helpfulness | 1.000 | 0,148 | **0,164** |
|
| 199 |
+
| helpsteer2_verbosity | 1.000 | 0,211 | **0,214** |
|
| 200 |
+
| ledgar | 1.000 | 0,087 | **0,092** |
|
| 201 |
+
| massive | 1.000 | 0,376 | **0,775** |
|
| 202 |
+
| measuring_hate_speech | 1.000 | 0,315 | **0,563** |
|
| 203 |
+
| mmlu | 1.000 | 0,274 | **0,267** |
|
| 204 |
+
| mnli | 1.000 | 0,719 | **0,757** |
|
| 205 |
+
| paws | 1.000 | 0,526 | **0,541** |
|
| 206 |
+
| sms_spam | 800 | 0,326 | **0,956** |
|
| 207 |
+
| sst5 | 1.000 | 0,283 | **0,385** |
|
| 208 |
+
| strategyqa_closed | 687 | 0,517 | **0,511** |
|
| 209 |
+
| strategyqa_grounded | 687 | 0,546 | **0,547** |
|
| 210 |
+
| stsb | 1.000 | 0,157 | **0,242** |
|
| 211 |
+
| yelp5 | 1.000 | 0,189 | **0,422** |
|
| 212 |
+
|
| 213 |
+
banking77 (77) ve clinc150 (150 seçenek) gibi çok seçenekli sorularda seçenekler 768 tokenlık soru bütçesine
|
| 214 |
+
sığmak için kısalır; bu, Laya'nın bilinen bir sınırıdır ([#394](https://github.com/NandhaKishorM/laya/issues/394)).
|
| 215 |
+
|
| 216 |
+
### Eğitim sırasında doğrulama
|
| 217 |
+
|
| 218 |
+
Eğitimden ayrılan, metne göre gruplanmış 1.571 Türkçe karar (sıcaklık kalibrasyonu öncesi):
|
| 219 |
+
|
| 220 |
+
| Epoch | Doğruluk | NLL | Brier | ECE |
|
| 221 |
+
|---:|---:|---:|---:|---:|
|
| 222 |
+
| 1 | 0,803 | 0,504 | 0,273 | 0,051 |
|
| 223 |
+
| 2 | 0,809 | 0,507 | 0,273 | 0,053 |
|
| 224 |
+
|
| 225 |
+
İkinci epoch çok az katkı verdi; daha fazla epoch önerilmez.
|
| 226 |
+
|
| 227 |
+
## Eğitim verisi
|
| 228 |
+
|
| 229 |
+
| Kaynak | Satır | Karar | choice / noul / score |
|
| 230 |
+
|---|---:|---:|---|
|
| 231 |
+
| bev-decision-150K-tr train (Türkçe, makine çevirisi) | 102.782 | 206.164 | 69.040 / 89.653 / 47.471 |
|
| 232 |
+
| bev-decision-150K train, İngilizce orijinal (Türkçe satırların ikizleri) | 34.261 | 69.158 | 22.610 / 30.587 / 15.961 |
|
| 233 |
+
| MASSIVE tr-TR train (ana dili Türkçe) | 5.000 | 5.000 | 5.000 / 0 / 0 |
|
| 234 |
+
| XNLI Türkçe validation (insan çevirisi) | 2.490 | 2.490 | 2.490 / 0 / 0 |
|
| 235 |
+
| beyazperde film yorumları train (ana dili Türkçe) | 5.000 | 5.000 | 0 / 0 / 5.000 |
|
| 236 |
+
| SQuAD 2.0-tr train, cevaplanabilirlik (makine çevirisi, 130K satırdan örneklem) | 20.000 | 20.000 | 0 / 20.000 / 0 |
|
| 237 |
+
| **Toplam** | **169.533** | **307.812** | |
|
| 238 |
+
|
| 239 |
+
Bunlardan metne göre gruplanmış 1.571 Türkçe karar doğrulama ve kalibrasyon için ayrıldı; bu kararların
|
| 240 |
+
İngilizce ikizleri de eğitimden çıkarıldı. Eğitimde 305.812 karar kullanıldı.
|
| 241 |
+
|
| 242 |
+
- **Hedefler:** Etiketler sorulardan çıkarılıp hedef olasılıklara çevrildi. Sert etiketler 0,1 ile yumuşatıldı (doğru
|
| 243 |
+
cevaba 0,9; `score` sorularında kalan pay komşu seviyelere). Laya'nın eğitim yöntemi (RLCD) bu dağılımları hedefler.
|
| 244 |
+
- **Alan dengesi:** Türkçenin sinyal taşımadığı ya da etiketlerin gürültülü olduğu bev alanlarından eğitim satırlarının
|
| 245 |
+
bir kısmı tutuldu:
|
| 246 |
+
|
| 247 |
+
| Alan | Tutulan oran | Çıkarılan satır |
|
| 248 |
+
|---|---:|---:|
|
| 249 |
+
| Software engineering and code | 0,3 | 4.376 |
|
| 250 |
+
| Spatial and logical reasoning | 0,3 | 11.581 |
|
| 251 |
+
| Browser interaction | 0,3 | 3.202 |
|
| 252 |
+
| Engagement and ranking | 0,2 | 3.318 |
|
| 253 |
+
|
| 254 |
+
- **Örtüşme temizliği:** Bir test satırında da geçen metni içeren eğitim satırları çıkarıldı:
|
| 255 |
+
|
| 256 |
+
| Eğitim kaynağı | Test kaynağı | Çıkarılan eğitim satırı |
|
| 257 |
+
|---|---|---:|
|
| 258 |
+
| bev-decision-tr | jev-bench-tr | 337 |
|
| 259 |
+
| bev-decision-tr | beyazperde | 6 |
|
| 260 |
+
| bev-decision-tr | SQuAD-tr | 2 |
|
| 261 |
+
| bev-decision-tr | MASSIVE tr-TR | 10 |
|
| 262 |
+
| MASSIVE tr-TR | bev-decision-tr | 477 |
|
| 263 |
+
| MASSIVE tr-TR | jev-bench-tr | 16 |
|
| 264 |
+
| MASSIVE tr-TR | MASSIVE tr-TR | 16 |
|
| 265 |
+
| beyazperde | beyazperde | 76 |
|
| 266 |
+
|
| 267 |
+
- **Bağlam bütçesi:** Hiçbir eğitim kararı 8.192 tokena sığmadığı ya da doğru seçeneği kısaltılınca
|
| 268 |
+
başka bir seçenekle aynı göründüğü için çıkarılmadı. Eğitim kararlarının token uzunluğu: medyan 188,
|
| 269 |
+
%90 521, %99 1.474, en fazla 7.430.
|
| 270 |
+
|
| 271 |
+
bev-decision-150K'nın Türkçesi bir LLM ile alan bazında çevrildi; kod blokları, tanımlayıcılar, URL'ler ve Mind2Web sayfa
|
| 272 |
+
öğeleri İngilizce bırakıldı. Ayrıntılar: [hayriyigit/bev-decision-150K-tr](https://huggingface.co/datasets/hayriyigit/bev-decision-150K-tr).
|
| 273 |
+
SQuAD 2.0'ın yalnızca paragraf ve soruları çevrildi; etiket `answers`'ın boş olup olmamasından gelir. Ayrıntılar:
|
| 274 |
+
[hayriyigit/squad_v2-tr](https://huggingface.co/datasets/hayriyigit/squad_v2-tr).
|
| 275 |
+
|
| 276 |
+
## Eğitim
|
| 277 |
+
|
| 278 |
+
| | |
|
| 279 |
+
|---|---|
|
| 280 |
+
| Taban model | `convaiinnovations/laya-multilingual` (mmBERT-base encoder + Laya karar başlığı) |
|
| 281 |
+
| Yöntem | Laya'nın RLCD tarifi: gürültülü logit örnekleri üzerinde proper scoring ödülüyle policy gradient + yumuşak cross-entropy |
|
| 282 |
+
| Epoch / güncelleme | 2 / 19.114 (her güncellemede 32 karar) |
|
| 283 |
+
| Öğrenme hızı | encoder 1e-5, karar başlığı 1e-4; AdamW, cosine |
|
| 284 |
+
| Hassasiyet | bf16, gradient checkpointing |
|
| 285 |
+
| Bağlam | max_len 8.192, head_max_len 768 |
|
| 286 |
+
| Donanım ve süre | 1 × NVIDIA GeForce RTX 5090 (32 GB); eğitim ~1 sa 34 dk, veri hazırlığı ve değerlendirmeyle ~1 sa 42 dk |
|
| 287 |
+
| Yazılım | torch 2.14.1, transformers 5.18.0, laya 0.3.24 (commit `fa9a2a7`) |
|
| 288 |
+
|
| 289 |
+
Eğitim config'i, eğitim kaydı ve veri manifest'i [`training/`](training/) klasöründe; tüm metrikler `eval_report.json`'da.
|
| 290 |
+
|
| 291 |
+
## Sınırlamalar
|
| 292 |
+
|
| 293 |
+
- Eğitim verisinin çoğu **makine çevirisidir**; model çeviri dilinin izlerini taşıyabilir. Ana dili Türkçe veri sınırlıdır
|
| 294 |
+
(~12 bin satır).
|
| 295 |
+
- Kod, teknik tanımlayıcılar ve web sayfası öğeleri eğitimde İngilizce kaldı; bu tür girdilerde karışık dil beklenir.
|
| 296 |
+
- Yalnızca Türkçe ve İngilizce değerlendirildi; diğer dillerdeki performans ince ayar öncesine göre düşmüş olabilir.
|
| 297 |
+
- Kalibrasyon Türkçe veriyle yapıldı; farklı bir alanda kullanmadan önce kendi verinizle eşikleri kontrol edin.
|
| 298 |
+
- 77'den fazla seçenekli sorularda seçenek açıklamaları kısalır.
|
| 299 |
+
- **Cevaplanabilirlik:** Konusu soruyla aynı olan ama istenen bilgiyi hiç içermeyen belgelerde model hâlâ yüksek olasılıkla
|
| 300 |
+
"evet" diyebilir. Örneğin "Türksat 6A hangi tarihte fırlatıldı?" sorusuna, fırlatma tarihini içermeyen ya da yalnızca başka
|
| 301 |
+
bir olayın tarihini içeren Türksat belgeleri için ~0,92 verir. Bu tür kararları tek başına bu modele bırakmayın.
|
| 302 |
+
|
| 303 |
+
## Lisans
|
| 304 |
+
|
| 305 |
+
Taban model Apache-2.0'dır. Eğitim verilerinin lisansları farklıdır; ayrıntılar [`LICENSE.md`](LICENSE.md)'de.
|
| 306 |
+
Ticari kullanımdan önce kontrol edin.
|
training/data_manifest.json
ADDED
|
@@ -0,0 +1,109 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"key": "864dcc13473ae4c0",
|
| 3 |
+
"created_at": "2026-10-05T16:03:13+00:00",
|
| 4 |
+
"sources": {
|
| 5 |
+
"bev_tr": {
|
| 6 |
+
"repo": "hayriyigit/bev-decision-150K-tr",
|
| 7 |
+
"revision": "70292beffb5bda58603177bdde76c6b10e91b468",
|
| 8 |
+
"format": "bev",
|
| 9 |
+
"original": {
|
| 10 |
+
"repo": "avbiswas/bev-decision-150K",
|
| 11 |
+
"revision": "f83fe8b97094112fa305bfc793be07c8f8742282"
|
| 12 |
+
}
|
| 13 |
+
},
|
| 14 |
+
"jev_bench_tr": {
|
| 15 |
+
"repo": "hayriyigit/jev-bench-tr",
|
| 16 |
+
"revision": "f78242b1abb5df2cb1817c597d01902ed575a718",
|
| 17 |
+
"format": "jev",
|
| 18 |
+
"original": {
|
| 19 |
+
"repo": "Praveenrajus/jev-bench",
|
| 20 |
+
"revision": "b41e6b2f68a429608e1c0d354324d0bebaec7971"
|
| 21 |
+
}
|
| 22 |
+
},
|
| 23 |
+
"massive_tr": {
|
| 24 |
+
"repo": "mteb/amazon_massive_intent",
|
| 25 |
+
"revision": "940fd47a81eaa7f2cc7b129674d945d618ac38c2",
|
| 26 |
+
"format": "classification",
|
| 27 |
+
"original": {
|
| 28 |
+
"repo": "mteb/amazon_massive_intent",
|
| 29 |
+
"revision": "940fd47a81eaa7f2cc7b129674d945d618ac38c2"
|
| 30 |
+
}
|
| 31 |
+
},
|
| 32 |
+
"xnli_tr": {
|
| 33 |
+
"repo": "facebook/xnli",
|
| 34 |
+
"revision": "b8dd5d7af51114dbda02c0e3f6133f332186418e",
|
| 35 |
+
"format": "classification",
|
| 36 |
+
"original": {
|
| 37 |
+
"repo": "facebook/xnli",
|
| 38 |
+
"revision": "b8dd5d7af51114dbda02c0e3f6133f332186418e"
|
| 39 |
+
}
|
| 40 |
+
},
|
| 41 |
+
"beyazperde_tr": {
|
| 42 |
+
"repo": "turkish-nlp-suite/beyazperde-all-movie-reviews",
|
| 43 |
+
"revision": "ba9bf2f104e7dfaac3b520049eecffc0ed01ebb5",
|
| 44 |
+
"format": "classification"
|
| 45 |
+
},
|
| 46 |
+
"squad_tr": {
|
| 47 |
+
"repo": "hayriyigit/squad_v2-tr",
|
| 48 |
+
"revision": "f007719f333819c541f3e22964dbbc78bf493a4b",
|
| 49 |
+
"format": "classification",
|
| 50 |
+
"original": {
|
| 51 |
+
"repo": "rajpurkar/squad_v2",
|
| 52 |
+
"revision": "3ffb306f725f7d2ce8394bc1873b24868140c412"
|
| 53 |
+
}
|
| 54 |
+
}
|
| 55 |
+
},
|
| 56 |
+
"rows": {
|
| 57 |
+
"train": 169533,
|
| 58 |
+
"test": 77016
|
| 59 |
+
},
|
| 60 |
+
"rows_by_source": {
|
| 61 |
+
"train": {
|
| 62 |
+
"bev_tr": 102782,
|
| 63 |
+
"bev_tr_en": 34261,
|
| 64 |
+
"massive_tr": 5000,
|
| 65 |
+
"xnli_tr": 2490,
|
| 66 |
+
"beyazperde_tr": 5000,
|
| 67 |
+
"squad_tr": 20000
|
| 68 |
+
},
|
| 69 |
+
"test": {
|
| 70 |
+
"bev_tr": 24386,
|
| 71 |
+
"bev_tr_en": 3000,
|
| 72 |
+
"jev_bench_tr": 22773,
|
| 73 |
+
"jev_bench_tr_en": 2000,
|
| 74 |
+
"massive_tr": 2974,
|
| 75 |
+
"xnli_tr": 5010,
|
| 76 |
+
"beyazperde_tr": 5000,
|
| 77 |
+
"squad_tr": 11873
|
| 78 |
+
}
|
| 79 |
+
},
|
| 80 |
+
"decisions": {
|
| 81 |
+
"train": {
|
| 82 |
+
"choice": 99140,
|
| 83 |
+
"noul": 140240,
|
| 84 |
+
"score": 68432
|
| 85 |
+
},
|
| 86 |
+
"test": {
|
| 87 |
+
"noul": 39371,
|
| 88 |
+
"choice": 39616,
|
| 89 |
+
"score": 22729
|
| 90 |
+
}
|
| 91 |
+
},
|
| 92 |
+
"train_rows_dropped_overlapping_eval": {
|
| 93 |
+
"bev_tr train x jev_bench_tr test": 337,
|
| 94 |
+
"bev_tr train x beyazperde_tr test": 6,
|
| 95 |
+
"bev_tr train x squad_tr test": 2,
|
| 96 |
+
"bev_tr train x massive_tr test": 10,
|
| 97 |
+
"massive_tr train x bev_tr test": 477,
|
| 98 |
+
"massive_tr train x jev_bench_tr test": 16,
|
| 99 |
+
"massive_tr train x massive_tr test": 16,
|
| 100 |
+
"beyazperde_tr train x beyazperde_tr test": 76
|
| 101 |
+
},
|
| 102 |
+
"train_rows_left_out_by_domain_weight": {
|
| 103 |
+
"bev_tr/Software engineering and code": 4376,
|
| 104 |
+
"bev_tr/Spatial and logical reasoning": 11581,
|
| 105 |
+
"bev_tr/Browser interaction": 3202,
|
| 106 |
+
"bev_tr/Engagement and ranking": 3318
|
| 107 |
+
},
|
| 108 |
+
"skipped": {}
|
| 109 |
+
}
|
training/laya.yaml
ADDED
|
@@ -0,0 +1,211 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
# Laya fine-tuning on the Turkish decision data, end to end:
|
| 2 |
+
# scripts/run_laya.sh configs/laya.yaml
|
| 3 |
+
# Everything goes to `workdir`; after a time limit, a crash or Ctrl+C run the same command again to resume.
|
| 4 |
+
workdir: ./laya_run
|
| 5 |
+
|
| 6 |
+
data:
|
| 7 |
+
sources:
|
| 8 |
+
- name: bev_tr # avbiswas/bev-decision-150K, machine-translated to Turkish
|
| 9 |
+
repo: hayriyigit/bev-decision-150K-tr
|
| 10 |
+
format: bev # bev | jev | laya (state/questions/gold rows) | classification
|
| 11 |
+
train: data/train.parquet # glob(s) inside the repo
|
| 12 |
+
test: data/test.parquet
|
| 13 |
+
domain_weights: # share of a domain's training rows kept (default 1)
|
| 14 |
+
"Software engineering and code": 0.3 # code, test ids and paths stayed English
|
| 15 |
+
"Browser interaction": 0.3 # page elements stayed English
|
| 16 |
+
"Engagement and ranking": 0.2 # headline click-through: labels close to noise
|
| 17 |
+
"Spatial and logical reasoning": 0.3 # templated, 13% of the rows
|
| 18 |
+
original: # the English source, row for row
|
| 19 |
+
repo: avbiswas/bev-decision-150K
|
| 20 |
+
revision: f83fe8b97094112fa305bfc793be07c8f8742282
|
| 21 |
+
train_share: 0.25 # English twins: ~25% of this source's training rows
|
| 22 |
+
test_rows: 3000 # English test rows: a regression check for the other languages
|
| 23 |
+
|
| 24 |
+
- name: jev_bench_tr # a benchmark: evaluation only, never trained on
|
| 25 |
+
repo: hayriyigit/jev-bench-tr # private: needs HF_TOKEN (environment or .env)
|
| 26 |
+
format: jev
|
| 27 |
+
test: data/*/test.jsonl
|
| 28 |
+
original:
|
| 29 |
+
repo: Praveenrajus/jev-bench
|
| 30 |
+
revision: b41e6b2f68a429608e1c0d354324d0bebaec7971
|
| 31 |
+
test_rows: 2000
|
| 32 |
+
|
| 33 |
+
# Native Turkish. The questions use jev-bench-tr's wording for the same tasks.
|
| 34 |
+
- name: massive_tr # MASSIVE tr-TR: localized voice-assistant requests
|
| 35 |
+
repo: mteb/amazon_massive_intent
|
| 36 |
+
format: classification
|
| 37 |
+
train: train/tr.json.gz
|
| 38 |
+
test: test/tr.json.gz
|
| 39 |
+
max_rows: 5000
|
| 40 |
+
text: text
|
| 41 |
+
label: label
|
| 42 |
+
original: # the same utterances in en-US, row for row (decontamination)
|
| 43 |
+
repo: mteb/amazon_massive_intent
|
| 44 |
+
train: train/en.json.gz
|
| 45 |
+
test: test/en.json.gz
|
| 46 |
+
question:
|
| 47 |
+
type: choice
|
| 48 |
+
instructions: "Kullanıcının sesli asistana yönelik isteği hangi niyeti ifade ediyor?"
|
| 49 |
+
criteria:
|
| 50 |
+
alarm_query: "Alarm sorgusu"
|
| 51 |
+
alarm_remove: "Alarmı sil"
|
| 52 |
+
alarm_set: "Alarm kur"
|
| 53 |
+
audio_volume_down: "Ses seviyesini azalt"
|
| 54 |
+
audio_volume_mute: "Sesi kapat"
|
| 55 |
+
audio_volume_other: "Ses seviyesi diğer"
|
| 56 |
+
audio_volume_up: "Ses seviyesini artır"
|
| 57 |
+
calendar_query: "Takvim sorgusu"
|
| 58 |
+
calendar_remove: "Takvimden sil"
|
| 59 |
+
calendar_set: "Takvime ekle"
|
| 60 |
+
cooking_query: "Yemek pişirme sorgusu"
|
| 61 |
+
cooking_recipe: "Yemek tarifi"
|
| 62 |
+
datetime_convert: "Tarih ve saat dönüştür"
|
| 63 |
+
datetime_query: "Tarih ve saat sorgusu"
|
| 64 |
+
email_addcontact: "E-posta kişi ekle"
|
| 65 |
+
email_query: "E-posta sorgusu"
|
| 66 |
+
email_querycontact: "E-posta kişi sorgula"
|
| 67 |
+
email_sendemail: "E-posta gönder"
|
| 68 |
+
general_greet: "Genel selamlaşma"
|
| 69 |
+
general_joke: "Genel şaka"
|
| 70 |
+
general_quirky: "Genel tuhaflık"
|
| 71 |
+
iot_cleaning: "Iot temizlik"
|
| 72 |
+
iot_coffee: "Iot kahve"
|
| 73 |
+
iot_hue_lightchange: "Iot hue ışık değiştir"
|
| 74 |
+
iot_hue_lightdim: "Iot hue ışık kıs"
|
| 75 |
+
iot_hue_lightoff: "Iot hue ışık kapat"
|
| 76 |
+
iot_hue_lighton: "Iot hue ışık aç"
|
| 77 |
+
iot_hue_lightup: "Iot hue ışık parlaklığı artır"
|
| 78 |
+
iot_wemo_off: "Iot wemo kapat"
|
| 79 |
+
iot_wemo_on: "Iot wemo aç"
|
| 80 |
+
lists_createoradd: "Liste oluştur veya ekle"
|
| 81 |
+
lists_query: "Liste sorgusu"
|
| 82 |
+
lists_remove: "Listeden sil"
|
| 83 |
+
music_dislikeness: "Müzik beğenmeme"
|
| 84 |
+
music_likeness: "Müzik beğenisi"
|
| 85 |
+
music_query: "Müzik sorgusu"
|
| 86 |
+
music_settings: "Müzik ayarları"
|
| 87 |
+
news_query: "Haber sorgusu"
|
| 88 |
+
play_audiobook: "Sesli kitap oynat"
|
| 89 |
+
play_game: "Oyun oyna"
|
| 90 |
+
play_music: "Müzik çal"
|
| 91 |
+
play_podcasts: "Podcast oynat"
|
| 92 |
+
play_radio: "Radyo aç"
|
| 93 |
+
qa_currency: "Soru-cevap döviz"
|
| 94 |
+
qa_definition: "Soru-cevap tanım"
|
| 95 |
+
qa_factoid: "Soru-cevap bilgi"
|
| 96 |
+
qa_maths: "Soru-cevap matematik"
|
| 97 |
+
qa_stock: "Soru-cevap borsa"
|
| 98 |
+
recommendation_events: "Öneri etkinlikler"
|
| 99 |
+
recommendation_locations: "Öneri konumlar"
|
| 100 |
+
recommendation_movies: "Öneri filmler"
|
| 101 |
+
social_post: "Sosyal medya paylaşımı"
|
| 102 |
+
social_query: "Sosyal medya sorgusu"
|
| 103 |
+
takeaway_order: "Paket servis siparişi"
|
| 104 |
+
takeaway_query: "Paket servis sorgusu"
|
| 105 |
+
transport_query: "Ulaşım sorgusu"
|
| 106 |
+
transport_taxi: "Ulaşım taksi"
|
| 107 |
+
transport_ticket: "Ulaşım bileti"
|
| 108 |
+
transport_traffic: "Ulaşım trafik"
|
| 109 |
+
weather_query: "Hava durumu sorgusu"
|
| 110 |
+
|
| 111 |
+
- name: xnli_tr # XNLI Turkish: human-translated validation (train) and test sets;
|
| 112 |
+
repo: facebook/xnli # its machine-translated train split is left out
|
| 113 |
+
format: classification
|
| 114 |
+
train: tr/validation-*.parquet
|
| 115 |
+
test: tr/test-*.parquet
|
| 116 |
+
text: {premise: premise, hypothesis: hypothesis}
|
| 117 |
+
label: label
|
| 118 |
+
label_map: {0: entailment, 1: neutral, 2: contradiction}
|
| 119 |
+
original:
|
| 120 |
+
repo: facebook/xnli
|
| 121 |
+
train: en/validation-*.parquet
|
| 122 |
+
test: en/test-*.parquet
|
| 123 |
+
question:
|
| 124 |
+
type: choice
|
| 125 |
+
instructions: "`hypothesis` ile `premise` arasındaki ilişki nedir?"
|
| 126 |
+
criteria:
|
| 127 |
+
entailment: "Hipotez, öncül verildiğinde kesinlikle doğrudur."
|
| 128 |
+
neutral: "Hipotez doğru olabilir; öncül bunu kesinleştirmiyor."
|
| 129 |
+
contradiction: "Hipotez, öncül verildiğinde kesinlikle yanlıştır."
|
| 130 |
+
|
| 131 |
+
- name: beyazperde_tr # native Turkish movie reviews, 0.5-5 stars -> five levels
|
| 132 |
+
repo: turkish-nlp-suite/beyazperde-all-movie-reviews
|
| 133 |
+
format: classification
|
| 134 |
+
train: dataset/train.json
|
| 135 |
+
test: dataset/test.json
|
| 136 |
+
max_rows: 5000
|
| 137 |
+
text: text
|
| 138 |
+
label: star
|
| 139 |
+
label_map: {0.5: 0, 1: 0, 1.5: 1, 2: 1, 2.5: 2, 3: 2, 3.5: 3, 4: 3, 4.5: 4, 5: 4}
|
| 140 |
+
question:
|
| 141 |
+
type: score
|
| 142 |
+
instructions: "Bu film yorumunun duygusu (sentiment) nedir?"
|
| 143 |
+
criteria:
|
| 144 |
+
- "Çok negatif: güçlü bir şekilde olumsuz, düşmanca veya küçümseyici."
|
| 145 |
+
- "Negatif: genel olarak olumsuz, bazı eleştirilerle."
|
| 146 |
+
- "Nötr: karışık veya olgusal, net bir eğilim yok."
|
| 147 |
+
- "Pozitif: genel olarak olumlu, bazı övgülerle."
|
| 148 |
+
- "Çok pozitif: hevesli, güçlü bir şekilde olumlu."
|
| 149 |
+
- name: squad_tr # SQuAD 2.0, machine-translated: can the paragraph answer the question?
|
| 150 |
+
repo: hayriyigit/squad_v2-tr # configs/squad.yaml; a third of train (half of dev) is unanswerable,
|
| 151 |
+
format: classification # written so the paragraph holds a plausible wrong answer
|
| 152 |
+
train: squad_v2/train-*.parquet
|
| 153 |
+
test: squad_v2/validation-*.parquet
|
| 154 |
+
max_rows: 20000 # of 130K: the other sources stay the bulk of training
|
| 155 |
+
text: {question: question, document: context}
|
| 156 |
+
label: answers.text # no answer: unanswerable
|
| 157 |
+
original:
|
| 158 |
+
repo: rajpurkar/squad_v2
|
| 159 |
+
revision: 3ffb306f725f7d2ce8394bc1873b24868140c412
|
| 160 |
+
question:
|
| 161 |
+
type: noul
|
| 162 |
+
instructions: "Verilen `document`, `question` sorusunun TÜM kısımlarını doğru ve eksiksiz şekilde cevaplamak için yeterli bilgiyi açıkça içeriyor mu? Yalnızca belgede açıkça belirtilen veya belgeden doğrudan ve tartışmasız biçimde çıkarılabilen bilgileri kullan. Konusal olarak ilgili olmak yeterli değildir. Soruda belirli bir tarih, sayı, isim, konum, neden, yöntem veya başka bir bilgi isteniyorsa, istenen bilgi belgede doğru olay, varlık ve ilişkiyle bağlantılı olmalıdır. İstenen cevaba benzeyen bir bilginin belgede başka bir olay veya varlık için geçmesi cevap sayılmaz. Sorunun birden fazla kısmı varsa TÜM kısımlar cevaplanabiliyorsa EVET, herhangi bir gerekli kısım eksikse HAYIR döndür. Dış bilgi kullanma."
|
| 163 |
+
# More data: another entry like these; `max_rows: N` samples a source down (train and test).
|
| 164 |
+
decontaminate: true # drop training rows whose text also appears in a test row
|
| 165 |
+
seed: 0
|
| 166 |
+
|
| 167 |
+
model:
|
| 168 |
+
base: convaiinnovations/laya-multilingual # mmBERT, 100+ languages; `convaiinnovations/laya` is English only.
|
| 169 |
+
# A local directory works too (offline servers).
|
| 170 |
+
# Tokens per decision: question + options (head_max_len) and the state. The checkpoint ships with 1024/256;
|
| 171 |
+
# its encoder knows 8192 positions. Measured on this data (training decisions that do not fit, tokens/epoch):
|
| 172 |
+
# 1024/256: 2.7% (Contract evidence 95%, Financial 15%, Response preference 14%, MASSIVE 13%) x1.00
|
| 173 |
+
# 2048/512: 0.5% (Contract evidence 69%) x1.08
|
| 174 |
+
# 4096/512: 0.1% (Contract evidence 15%) x1.11
|
| 175 |
+
# The exported model infers with these values too; a short input costs the same either way.
|
| 176 |
+
max_len: 8192
|
| 177 |
+
head_max_len: 768
|
| 178 |
+
|
| 179 |
+
train:
|
| 180 |
+
epochs: 2 # one or two: validation after each shows whether the second helps
|
| 181 |
+
micro_batch: 8 # sequences per forward pass per GPU: lower it on an out-of-memory error
|
| 182 |
+
grad_accum: 4 # effective batch = micro_batch x grad_accum x GPUs
|
| 183 |
+
lr_encoder: 1.0e-5 # Laya's recipe uses 2.5e-5; lower limits forgetting of other languages
|
| 184 |
+
lr_head: 1.0e-4
|
| 185 |
+
label_smoothing: 0.1 # hard labels -> 0.9 on the answer (score: the rest on neighbouring levels)
|
| 186 |
+
drop_truncated: true # training decisions whose state or options do not fit max_len are noise
|
| 187 |
+
holdout: 2000 # Turkish training decisions held out, whole states: validation + calibration
|
| 188 |
+
freeze_embeddings: false # worth an A/B run: protects other languages, saves ~2.5 GB of optimizer state
|
| 189 |
+
precision: auto # auto: bf16 where supported, else fp16 (CUDA); MPS/CPU train in fp32
|
| 190 |
+
gpus: auto # every visible GPU; more than one runs DDP (torchrun) by itself
|
| 191 |
+
gradient_checkpointing: true
|
| 192 |
+
log_every: 50 # optimizer updates between progress lines (train_log.jsonl too)
|
| 193 |
+
|
| 194 |
+
checkpoint:
|
| 195 |
+
every_minutes: 30 # also at every epoch end
|
| 196 |
+
save_optimizer: true # exact resume; ~4 GB per checkpoint (twice that while one is written)
|
| 197 |
+
hub_repo: hayriyigit/laya-tr-test-v2 # also push it to this repo's `checkpoints` branch (one copy kept); a new
|
| 198 |
+
# machine resumes from there. null: local only. Needs a write HF_TOKEN.
|
| 199 |
+
hub_every_minutes: 60 # and at every epoch end and stop
|
| 200 |
+
|
| 201 |
+
limits:
|
| 202 |
+
max_hours: 0 # wall-clock budget of one invocation, e.g. the job's time limit; 0 = none
|
| 203 |
+
reserve_minutes: 20 # training stops this long before max_hours: checkpoint + export + eval
|
| 204 |
+
|
| 205 |
+
eval:
|
| 206 |
+
batch_size: 32
|
| 207 |
+
max_items: null # sample the test split down (~85k decisions with all sources)
|
| 208 |
+
baseline: true # score the base checkpoint too: a before/after table per source and type
|
| 209 |
+
|
| 210 |
+
export:
|
| 211 |
+
push_to_hub: {repo: hayriyigit/laya-tr-test-v2, private: true} # the final model -> main; null: keep it local
|
training/run_stats.json
ADDED
|
@@ -0,0 +1,759 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"model_config": {
|
| 3 |
+
"encoder": "jhu-clsp/mmBERT-base",
|
| 4 |
+
"head_layers": 2,
|
| 5 |
+
"max_len": 8192,
|
| 6 |
+
"head_max_len": 768,
|
| 7 |
+
"max_prefixes": 6,
|
| 8 |
+
"act_costs": {
|
| 9 |
+
"escalate": 0.5
|
| 10 |
+
},
|
| 11 |
+
"cost_wrong_act": 3.0,
|
| 12 |
+
"amp_dtype": "bf16",
|
| 13 |
+
"model_name": "laya-finetuned-laya_run",
|
| 14 |
+
"temperature": [
|
| 15 |
+
0.8975213766098022,
|
| 16 |
+
1.0775697231292725,
|
| 17 |
+
0.94973224401474
|
| 18 |
+
],
|
| 19 |
+
"training": {
|
| 20 |
+
"updates": 15987,
|
| 21 |
+
"epochs_completed": 4,
|
| 22 |
+
"hours": 4.97,
|
| 23 |
+
"world_size": 1,
|
| 24 |
+
"fine_tuned_from_checkpoint": false
|
| 25 |
+
},
|
| 26 |
+
"fine_tuned": true
|
| 27 |
+
},
|
| 28 |
+
"export": {
|
| 29 |
+
"train_key": "faca7844a506b4f7",
|
| 30 |
+
"temperature": [
|
| 31 |
+
0.8975213766098022,
|
| 32 |
+
1.0775697231292725,
|
| 33 |
+
0.94973224401474
|
| 34 |
+
],
|
| 35 |
+
"base": "convaiinnovations/laya-multilingual",
|
| 36 |
+
"data": {
|
| 37 |
+
"bev_tr": {
|
| 38 |
+
"repo": "hayriyigit/bev-decision-150K-tr",
|
| 39 |
+
"revision": "70292beffb5bda58603177bdde76c6b10e91b468",
|
| 40 |
+
"format": "bev",
|
| 41 |
+
"original": {
|
| 42 |
+
"repo": "avbiswas/bev-decision-150K",
|
| 43 |
+
"revision": "f83fe8b97094112fa305bfc793be07c8f8742282"
|
| 44 |
+
}
|
| 45 |
+
},
|
| 46 |
+
"jev_bench_tr": {
|
| 47 |
+
"repo": "hayriyigit/jev-bench-tr",
|
| 48 |
+
"revision": "f78242b1abb5df2cb1817c597d01902ed575a718",
|
| 49 |
+
"format": "jev",
|
| 50 |
+
"original": {
|
| 51 |
+
"repo": "Praveenrajus/jev-bench",
|
| 52 |
+
"revision": "b41e6b2f68a429608e1c0d354324d0bebaec7971"
|
| 53 |
+
}
|
| 54 |
+
},
|
| 55 |
+
"massive_tr": {
|
| 56 |
+
"repo": "mteb/amazon_massive_intent",
|
| 57 |
+
"revision": "940fd47a81eaa7f2cc7b129674d945d618ac38c2",
|
| 58 |
+
"format": "classification",
|
| 59 |
+
"original": {
|
| 60 |
+
"repo": "mteb/amazon_massive_intent",
|
| 61 |
+
"revision": "940fd47a81eaa7f2cc7b129674d945d618ac38c2"
|
| 62 |
+
}
|
| 63 |
+
},
|
| 64 |
+
"xnli_tr": {
|
| 65 |
+
"repo": "facebook/xnli",
|
| 66 |
+
"revision": "b8dd5d7af51114dbda02c0e3f6133f332186418e",
|
| 67 |
+
"format": "classification",
|
| 68 |
+
"original": {
|
| 69 |
+
"repo": "facebook/xnli",
|
| 70 |
+
"revision": "b8dd5d7af51114dbda02c0e3f6133f332186418e"
|
| 71 |
+
}
|
| 72 |
+
},
|
| 73 |
+
"beyazperde_tr": {
|
| 74 |
+
"repo": "turkish-nlp-suite/beyazperde-all-movie-reviews",
|
| 75 |
+
"revision": "ba9bf2f104e7dfaac3b520049eecffc0ed01ebb5",
|
| 76 |
+
"format": "classification"
|
| 77 |
+
},
|
| 78 |
+
"squad_tr": {
|
| 79 |
+
"repo": "hayriyigit/squad_v2-tr",
|
| 80 |
+
"revision": "f007719f333819c541f3e22964dbbc78bf493a4b",
|
| 81 |
+
"format": "classification",
|
| 82 |
+
"original": {
|
| 83 |
+
"repo": "rajpurkar/squad_v2",
|
| 84 |
+
"revision": "3ffb306f725f7d2ce8394bc1873b24868140c412"
|
| 85 |
+
}
|
| 86 |
+
}
|
| 87 |
+
}
|
| 88 |
+
},
|
| 89 |
+
"checkpoint_progress": {
|
| 90 |
+
"train_key": "faca7844a506b4f7",
|
| 91 |
+
"epoch": 2,
|
| 92 |
+
"batch": 0,
|
| 93 |
+
"update": 19114,
|
| 94 |
+
"done": true,
|
| 95 |
+
"saved_at": "2026-10-05T17:39:10+00:00"
|
| 96 |
+
},
|
| 97 |
+
"data": {
|
| 98 |
+
"train": {
|
| 99 |
+
"by_source": {
|
| 100 |
+
"bev_tr": {
|
| 101 |
+
"rows": 102782,
|
| 102 |
+
"decisions": 206164,
|
| 103 |
+
"types": {
|
| 104 |
+
"choice": 69040,
|
| 105 |
+
"noul": 89653,
|
| 106 |
+
"score": 47471
|
| 107 |
+
},
|
| 108 |
+
"state_chars_median": 306
|
| 109 |
+
},
|
| 110 |
+
"bev_tr_en": {
|
| 111 |
+
"rows": 34261,
|
| 112 |
+
"decisions": 69158,
|
| 113 |
+
"types": {
|
| 114 |
+
"noul": 30587,
|
| 115 |
+
"score": 15961,
|
| 116 |
+
"choice": 22610
|
| 117 |
+
},
|
| 118 |
+
"state_chars_median": 296
|
| 119 |
+
},
|
| 120 |
+
"beyazperde_tr": {
|
| 121 |
+
"rows": 5000,
|
| 122 |
+
"decisions": 5000,
|
| 123 |
+
"types": {
|
| 124 |
+
"score": 5000
|
| 125 |
+
},
|
| 126 |
+
"state_chars_median": 196
|
| 127 |
+
},
|
| 128 |
+
"massive_tr": {
|
| 129 |
+
"rows": 5000,
|
| 130 |
+
"decisions": 5000,
|
| 131 |
+
"types": {
|
| 132 |
+
"choice": 5000
|
| 133 |
+
},
|
| 134 |
+
"state_chars_median": 34
|
| 135 |
+
},
|
| 136 |
+
"squad_tr": {
|
| 137 |
+
"rows": 20000,
|
| 138 |
+
"decisions": 20000,
|
| 139 |
+
"types": {
|
| 140 |
+
"noul": 20000
|
| 141 |
+
},
|
| 142 |
+
"state_chars_median": 789
|
| 143 |
+
},
|
| 144 |
+
"xnli_tr": {
|
| 145 |
+
"rows": 2490,
|
| 146 |
+
"decisions": 2490,
|
| 147 |
+
"types": {
|
| 148 |
+
"choice": 2490
|
| 149 |
+
},
|
| 150 |
+
"state_chars_median": 185
|
| 151 |
+
}
|
| 152 |
+
},
|
| 153 |
+
"by_workflow": {
|
| 154 |
+
"bev_tr/Browser interaction": {
|
| 155 |
+
"rows": 1298,
|
| 156 |
+
"decisions": 1298
|
| 157 |
+
},
|
| 158 |
+
"bev_tr/Civic and safety operations": {
|
| 159 |
+
"rows": 874,
|
| 160 |
+
"decisions": 874
|
| 161 |
+
},
|
| 162 |
+
"bev_tr/Contract evidence": {
|
| 163 |
+
"rows": 1680,
|
| 164 |
+
"decisions": 1680
|
| 165 |
+
},
|
| 166 |
+
"bev_tr/Engagement and ranking": {
|
| 167 |
+
"rows": 842,
|
| 168 |
+
"decisions": 842
|
| 169 |
+
},
|
| 170 |
+
"bev_tr/Financial reporting and banking": {
|
| 171 |
+
"rows": 7563,
|
| 172 |
+
"decisions": 9142
|
| 173 |
+
},
|
| 174 |
+
"bev_tr/Game-state decisions": {
|
| 175 |
+
"rows": 1699,
|
| 176 |
+
"decisions": 11112
|
| 177 |
+
},
|
| 178 |
+
"bev_tr/Reading comprehension": {
|
| 179 |
+
"rows": 2197,
|
| 180 |
+
"decisions": 2197
|
| 181 |
+
},
|
| 182 |
+
"bev_tr/Response preference and quality": {
|
| 183 |
+
"rows": 7308,
|
| 184 |
+
"decisions": 15260
|
| 185 |
+
},
|
| 186 |
+
"bev_tr/Retail, product, and shopping": {
|
| 187 |
+
"rows": 16523,
|
| 188 |
+
"decisions": 33854
|
| 189 |
+
},
|
| 190 |
+
"bev_tr/Scientific and paper understanding": {
|
| 191 |
+
"rows": 8625,
|
| 192 |
+
"decisions": 19053
|
| 193 |
+
},
|
| 194 |
+
"bev_tr/Sensory quality rating": {
|
| 195 |
+
"rows": 1040,
|
| 196 |
+
"decisions": 1040
|
| 197 |
+
},
|
| 198 |
+
"bev_tr/Sentiment, emotion, and moderation": {
|
| 199 |
+
"rows": 20623,
|
| 200 |
+
"decisions": 55779
|
| 201 |
+
},
|
| 202 |
+
"bev_tr/Software engineering and code": {
|
| 203 |
+
"rows": 1799,
|
| 204 |
+
"decisions": 1889
|
| 205 |
+
},
|
| 206 |
+
"bev_tr/Software security": {
|
| 207 |
+
"rows": 5880,
|
| 208 |
+
"decisions": 11760
|
| 209 |
+
},
|
| 210 |
+
"bev_tr/Spam detection": {
|
| 211 |
+
"rows": 957,
|
| 212 |
+
"decisions": 957
|
| 213 |
+
},
|
| 214 |
+
"bev_tr/Spatial and logical reasoning": {
|
| 215 |
+
"rows": 4959,
|
| 216 |
+
"decisions": 13942
|
| 217 |
+
},
|
| 218 |
+
"bev_tr/Support and intent routing": {
|
| 219 |
+
"rows": 10554,
|
| 220 |
+
"decisions": 14778
|
| 221 |
+
},
|
| 222 |
+
"bev_tr/Tool and workflow decisions": {
|
| 223 |
+
"rows": 8361,
|
| 224 |
+
"decisions": 10707
|
| 225 |
+
},
|
| 226 |
+
"bev_tr_en/Browser interaction": {
|
| 227 |
+
"rows": 434,
|
| 228 |
+
"decisions": 434
|
| 229 |
+
},
|
| 230 |
+
"bev_tr_en/Civic and safety operations": {
|
| 231 |
+
"rows": 305,
|
| 232 |
+
"decisions": 305
|
| 233 |
+
},
|
| 234 |
+
"bev_tr_en/Contract evidence": {
|
| 235 |
+
"rows": 556,
|
| 236 |
+
"decisions": 556
|
| 237 |
+
},
|
| 238 |
+
"bev_tr_en/Engagement and ranking": {
|
| 239 |
+
"rows": 291,
|
| 240 |
+
"decisions": 291
|
| 241 |
+
},
|
| 242 |
+
"bev_tr_en/Financial reporting and banking": {
|
| 243 |
+
"rows": 2515,
|
| 244 |
+
"decisions": 3017
|
| 245 |
+
},
|
| 246 |
+
"bev_tr_en/Game-state decisions": {
|
| 247 |
+
"rows": 554,
|
| 248 |
+
"decisions": 3684
|
| 249 |
+
},
|
| 250 |
+
"bev_tr_en/Reading comprehension": {
|
| 251 |
+
"rows": 738,
|
| 252 |
+
"decisions": 738
|
| 253 |
+
},
|
| 254 |
+
"bev_tr_en/Response preference and quality": {
|
| 255 |
+
"rows": 2427,
|
| 256 |
+
"decisions": 5099
|
| 257 |
+
},
|
| 258 |
+
"bev_tr_en/Retail, product, and shopping": {
|
| 259 |
+
"rows": 5499,
|
| 260 |
+
"decisions": 11211
|
| 261 |
+
},
|
| 262 |
+
"bev_tr_en/Scientific and paper understanding": {
|
| 263 |
+
"rows": 2830,
|
| 264 |
+
"decisions": 6346
|
| 265 |
+
},
|
| 266 |
+
"bev_tr_en/Sensory quality rating": {
|
| 267 |
+
"rows": 357,
|
| 268 |
+
"decisions": 357
|
| 269 |
+
},
|
| 270 |
+
"bev_tr_en/Sentiment, emotion, and moderation": {
|
| 271 |
+
"rows": 7008,
|
| 272 |
+
"decisions": 19258
|
| 273 |
+
},
|
| 274 |
+
"bev_tr_en/Software engineering and code": {
|
| 275 |
+
"rows": 575,
|
| 276 |
+
"decisions": 603
|
| 277 |
+
},
|
| 278 |
+
"bev_tr_en/Software security": {
|
| 279 |
+
"rows": 1998,
|
| 280 |
+
"decisions": 3996
|
| 281 |
+
},
|
| 282 |
+
"bev_tr_en/Spam detection": {
|
| 283 |
+
"rows": 341,
|
| 284 |
+
"decisions": 341
|
| 285 |
+
},
|
| 286 |
+
"bev_tr_en/Spatial and logical reasoning": {
|
| 287 |
+
"rows": 1634,
|
| 288 |
+
"decisions": 4611
|
| 289 |
+
},
|
| 290 |
+
"bev_tr_en/Support and intent routing": {
|
| 291 |
+
"rows": 3464,
|
| 292 |
+
"decisions": 4866
|
| 293 |
+
},
|
| 294 |
+
"bev_tr_en/Tool and workflow decisions": {
|
| 295 |
+
"rows": 2735,
|
| 296 |
+
"decisions": 3445
|
| 297 |
+
}
|
| 298 |
+
}
|
| 299 |
+
},
|
| 300 |
+
"test": {
|
| 301 |
+
"by_source": {
|
| 302 |
+
"bev_tr": {
|
| 303 |
+
"rows": 24386,
|
| 304 |
+
"decisions": 46320,
|
| 305 |
+
"types": {
|
| 306 |
+
"noul": 17572,
|
| 307 |
+
"choice": 18808,
|
| 308 |
+
"score": 9940
|
| 309 |
+
},
|
| 310 |
+
"state_chars_median": 374
|
| 311 |
+
},
|
| 312 |
+
"bev_tr_en": {
|
| 313 |
+
"rows": 3000,
|
| 314 |
+
"decisions": 5766,
|
| 315 |
+
"types": {
|
| 316 |
+
"choice": 2386,
|
| 317 |
+
"score": 1263,
|
| 318 |
+
"noul": 2117
|
| 319 |
+
},
|
| 320 |
+
"state_chars_median": 379
|
| 321 |
+
},
|
| 322 |
+
"beyazperde_tr": {
|
| 323 |
+
"rows": 5000,
|
| 324 |
+
"decisions": 5000,
|
| 325 |
+
"types": {
|
| 326 |
+
"score": 5000
|
| 327 |
+
},
|
| 328 |
+
"state_chars_median": 191
|
| 329 |
+
},
|
| 330 |
+
"jev_bench_tr": {
|
| 331 |
+
"rows": 22773,
|
| 332 |
+
"decisions": 22773,
|
| 333 |
+
"types": {
|
| 334 |
+
"choice": 9599,
|
| 335 |
+
"noul": 7174,
|
| 336 |
+
"score": 6000
|
| 337 |
+
},
|
| 338 |
+
"state_chars_median": 172
|
| 339 |
+
},
|
| 340 |
+
"jev_bench_tr_en": {
|
| 341 |
+
"rows": 2000,
|
| 342 |
+
"decisions": 2000,
|
| 343 |
+
"types": {
|
| 344 |
+
"score": 526,
|
| 345 |
+
"choice": 839,
|
| 346 |
+
"noul": 635
|
| 347 |
+
},
|
| 348 |
+
"state_chars_median": 165
|
| 349 |
+
},
|
| 350 |
+
"massive_tr": {
|
| 351 |
+
"rows": 2974,
|
| 352 |
+
"decisions": 2974,
|
| 353 |
+
"types": {
|
| 354 |
+
"choice": 2974
|
| 355 |
+
},
|
| 356 |
+
"state_chars_median": 34
|
| 357 |
+
},
|
| 358 |
+
"squad_tr": {
|
| 359 |
+
"rows": 11873,
|
| 360 |
+
"decisions": 11873,
|
| 361 |
+
"types": {
|
| 362 |
+
"noul": 11873
|
| 363 |
+
},
|
| 364 |
+
"state_chars_median": 813
|
| 365 |
+
},
|
| 366 |
+
"xnli_tr": {
|
| 367 |
+
"rows": 5010,
|
| 368 |
+
"decisions": 5010,
|
| 369 |
+
"types": {
|
| 370 |
+
"choice": 5010
|
| 371 |
+
},
|
| 372 |
+
"state_chars_median": 186
|
| 373 |
+
}
|
| 374 |
+
},
|
| 375 |
+
"by_workflow": {
|
| 376 |
+
"bev_tr/Browser interaction": {
|
| 377 |
+
"rows": 286,
|
| 378 |
+
"decisions": 286
|
| 379 |
+
},
|
| 380 |
+
"bev_tr/Civic and safety operations": {
|
| 381 |
+
"rows": 261,
|
| 382 |
+
"decisions": 261
|
| 383 |
+
},
|
| 384 |
+
"bev_tr/Contract evidence": {
|
| 385 |
+
"rows": 320,
|
| 386 |
+
"decisions": 320
|
| 387 |
+
},
|
| 388 |
+
"bev_tr/Engagement and ranking": {
|
| 389 |
+
"rows": 82,
|
| 390 |
+
"decisions": 82
|
| 391 |
+
},
|
| 392 |
+
"bev_tr/Financial reporting and banking": {
|
| 393 |
+
"rows": 1440,
|
| 394 |
+
"decisions": 1740
|
| 395 |
+
},
|
| 396 |
+
"bev_tr/Game-state decisions": {
|
| 397 |
+
"rows": 301,
|
| 398 |
+
"decisions": 2039
|
| 399 |
+
},
|
| 400 |
+
"bev_tr/Reading comprehension": {
|
| 401 |
+
"rows": 1422,
|
| 402 |
+
"decisions": 1422
|
| 403 |
+
},
|
| 404 |
+
"bev_tr/Response preference and quality": {
|
| 405 |
+
"rows": 1360,
|
| 406 |
+
"decisions": 3176
|
| 407 |
+
},
|
| 408 |
+
"bev_tr/Retail, product, and shopping": {
|
| 409 |
+
"rows": 3360,
|
| 410 |
+
"decisions": 6601
|
| 411 |
+
},
|
| 412 |
+
"bev_tr/Scientific and paper understanding": {
|
| 413 |
+
"rows": 508,
|
| 414 |
+
"decisions": 508
|
| 415 |
+
},
|
| 416 |
+
"bev_tr/Sensory quality rating": {
|
| 417 |
+
"rows": 260,
|
| 418 |
+
"decisions": 260
|
| 419 |
+
},
|
| 420 |
+
"bev_tr/Sentiment, emotion, and moderation": {
|
| 421 |
+
"rows": 4396,
|
| 422 |
+
"decisions": 11526
|
| 423 |
+
},
|
| 424 |
+
"bev_tr/Software engineering and code": {
|
| 425 |
+
"rows": 2011,
|
| 426 |
+
"decisions": 2039
|
| 427 |
+
},
|
| 428 |
+
"bev_tr/Software security": {
|
| 429 |
+
"rows": 1120,
|
| 430 |
+
"decisions": 2240
|
| 431 |
+
},
|
| 432 |
+
"bev_tr/Spam detection": {
|
| 433 |
+
"rows": 271,
|
| 434 |
+
"decisions": 271
|
| 435 |
+
},
|
| 436 |
+
"bev_tr/Spatial and logical reasoning": {
|
| 437 |
+
"rows": 3039,
|
| 438 |
+
"decisions": 8384
|
| 439 |
+
},
|
| 440 |
+
"bev_tr/Support and intent routing": {
|
| 441 |
+
"rows": 2087,
|
| 442 |
+
"decisions": 2858
|
| 443 |
+
},
|
| 444 |
+
"bev_tr/Tool and workflow decisions": {
|
| 445 |
+
"rows": 1862,
|
| 446 |
+
"decisions": 2307
|
| 447 |
+
},
|
| 448 |
+
"bev_tr_en/Browser interaction": {
|
| 449 |
+
"rows": 39,
|
| 450 |
+
"decisions": 39
|
| 451 |
+
},
|
| 452 |
+
"bev_tr_en/Civic and safety operations": {
|
| 453 |
+
"rows": 31,
|
| 454 |
+
"decisions": 31
|
| 455 |
+
},
|
| 456 |
+
"bev_tr_en/Contract evidence": {
|
| 457 |
+
"rows": 49,
|
| 458 |
+
"decisions": 49
|
| 459 |
+
},
|
| 460 |
+
"bev_tr_en/Engagement and ranking": {
|
| 461 |
+
"rows": 6,
|
| 462 |
+
"decisions": 6
|
| 463 |
+
},
|
| 464 |
+
"bev_tr_en/Financial reporting and banking": {
|
| 465 |
+
"rows": 165,
|
| 466 |
+
"decisions": 196
|
| 467 |
+
},
|
| 468 |
+
"bev_tr_en/Game-state decisions": {
|
| 469 |
+
"rows": 44,
|
| 470 |
+
"decisions": 303
|
| 471 |
+
},
|
| 472 |
+
"bev_tr_en/Reading comprehension": {
|
| 473 |
+
"rows": 168,
|
| 474 |
+
"decisions": 168
|
| 475 |
+
},
|
| 476 |
+
"bev_tr_en/Response preference and quality": {
|
| 477 |
+
"rows": 157,
|
| 478 |
+
"decisions": 389
|
| 479 |
+
},
|
| 480 |
+
"bev_tr_en/Retail, product, and shopping": {
|
| 481 |
+
"rows": 452,
|
| 482 |
+
"decisions": 902
|
| 483 |
+
},
|
| 484 |
+
"bev_tr_en/Scientific and paper understanding": {
|
| 485 |
+
"rows": 66,
|
| 486 |
+
"decisions": 66
|
| 487 |
+
},
|
| 488 |
+
"bev_tr_en/Sensory quality rating": {
|
| 489 |
+
"rows": 37,
|
| 490 |
+
"decisions": 37
|
| 491 |
+
},
|
| 492 |
+
"bev_tr_en/Sentiment, emotion, and moderation": {
|
| 493 |
+
"rows": 564,
|
| 494 |
+
"decisions": 1439
|
| 495 |
+
},
|
| 496 |
+
"bev_tr_en/Software engineering and code": {
|
| 497 |
+
"rows": 258,
|
| 498 |
+
"decisions": 260
|
| 499 |
+
},
|
| 500 |
+
"bev_tr_en/Software security": {
|
| 501 |
+
"rows": 117,
|
| 502 |
+
"decisions": 234
|
| 503 |
+
},
|
| 504 |
+
"bev_tr_en/Spam detection": {
|
| 505 |
+
"rows": 23,
|
| 506 |
+
"decisions": 23
|
| 507 |
+
},
|
| 508 |
+
"bev_tr_en/Spatial and logical reasoning": {
|
| 509 |
+
"rows": 383,
|
| 510 |
+
"decisions": 1042
|
| 511 |
+
},
|
| 512 |
+
"bev_tr_en/Support and intent routing": {
|
| 513 |
+
"rows": 230,
|
| 514 |
+
"decisions": 310
|
| 515 |
+
},
|
| 516 |
+
"bev_tr_en/Tool and workflow decisions": {
|
| 517 |
+
"rows": 211,
|
| 518 |
+
"decisions": 272
|
| 519 |
+
},
|
| 520 |
+
"jev_bench_tr/arc_challenge": {
|
| 521 |
+
"rows": 1000,
|
| 522 |
+
"decisions": 1000
|
| 523 |
+
},
|
| 524 |
+
"jev_bench_tr/banking77": {
|
| 525 |
+
"rows": 1000,
|
| 526 |
+
"decisions": 1000
|
| 527 |
+
},
|
| 528 |
+
"jev_bench_tr/boolq": {
|
| 529 |
+
"rows": 1000,
|
| 530 |
+
"decisions": 1000
|
| 531 |
+
},
|
| 532 |
+
"jev_bench_tr/chaosnli": {
|
| 533 |
+
"rows": 1599,
|
| 534 |
+
"decisions": 1599
|
| 535 |
+
},
|
| 536 |
+
"jev_bench_tr/civil_comments": {
|
| 537 |
+
"rows": 2000,
|
| 538 |
+
"decisions": 2000
|
| 539 |
+
},
|
| 540 |
+
"jev_bench_tr/clinc150": {
|
| 541 |
+
"rows": 1000,
|
| 542 |
+
"decisions": 1000
|
| 543 |
+
},
|
| 544 |
+
"jev_bench_tr/fever_evidence": {
|
| 545 |
+
"rows": 1000,
|
| 546 |
+
"decisions": 1000
|
| 547 |
+
},
|
| 548 |
+
"jev_bench_tr/go_emotions": {
|
| 549 |
+
"rows": 1000,
|
| 550 |
+
"decisions": 1000
|
| 551 |
+
},
|
| 552 |
+
"jev_bench_tr/helpsteer2_helpfulness": {
|
| 553 |
+
"rows": 1000,
|
| 554 |
+
"decisions": 1000
|
| 555 |
+
},
|
| 556 |
+
"jev_bench_tr/helpsteer2_verbosity": {
|
| 557 |
+
"rows": 1000,
|
| 558 |
+
"decisions": 1000
|
| 559 |
+
},
|
| 560 |
+
"jev_bench_tr/ledgar": {
|
| 561 |
+
"rows": 1000,
|
| 562 |
+
"decisions": 1000
|
| 563 |
+
},
|
| 564 |
+
"jev_bench_tr/massive": {
|
| 565 |
+
"rows": 1000,
|
| 566 |
+
"decisions": 1000
|
| 567 |
+
},
|
| 568 |
+
"jev_bench_tr/measuring_hate_speech": {
|
| 569 |
+
"rows": 1000,
|
| 570 |
+
"decisions": 1000
|
| 571 |
+
},
|
| 572 |
+
"jev_bench_tr/mmlu": {
|
| 573 |
+
"rows": 1000,
|
| 574 |
+
"decisions": 1000
|
| 575 |
+
},
|
| 576 |
+
"jev_bench_tr/mnli": {
|
| 577 |
+
"rows": 1000,
|
| 578 |
+
"decisions": 1000
|
| 579 |
+
},
|
| 580 |
+
"jev_bench_tr/paws": {
|
| 581 |
+
"rows": 1000,
|
| 582 |
+
"decisions": 1000
|
| 583 |
+
},
|
| 584 |
+
"jev_bench_tr/sms_spam": {
|
| 585 |
+
"rows": 800,
|
| 586 |
+
"decisions": 800
|
| 587 |
+
},
|
| 588 |
+
"jev_bench_tr/sst5": {
|
| 589 |
+
"rows": 1000,
|
| 590 |
+
"decisions": 1000
|
| 591 |
+
},
|
| 592 |
+
"jev_bench_tr/strategyqa_closed": {
|
| 593 |
+
"rows": 687,
|
| 594 |
+
"decisions": 687
|
| 595 |
+
},
|
| 596 |
+
"jev_bench_tr/strategyqa_grounded": {
|
| 597 |
+
"rows": 687,
|
| 598 |
+
"decisions": 687
|
| 599 |
+
},
|
| 600 |
+
"jev_bench_tr/stsb": {
|
| 601 |
+
"rows": 1000,
|
| 602 |
+
"decisions": 1000
|
| 603 |
+
},
|
| 604 |
+
"jev_bench_tr/yelp5": {
|
| 605 |
+
"rows": 1000,
|
| 606 |
+
"decisions": 1000
|
| 607 |
+
},
|
| 608 |
+
"jev_bench_tr_en/arc_challenge": {
|
| 609 |
+
"rows": 91,
|
| 610 |
+
"decisions": 91
|
| 611 |
+
},
|
| 612 |
+
"jev_bench_tr_en/banking77": {
|
| 613 |
+
"rows": 71,
|
| 614 |
+
"decisions": 71
|
| 615 |
+
},
|
| 616 |
+
"jev_bench_tr_en/boolq": {
|
| 617 |
+
"rows": 80,
|
| 618 |
+
"decisions": 80
|
| 619 |
+
},
|
| 620 |
+
"jev_bench_tr_en/chaosnli": {
|
| 621 |
+
"rows": 151,
|
| 622 |
+
"decisions": 151
|
| 623 |
+
},
|
| 624 |
+
"jev_bench_tr_en/civil_comments": {
|
| 625 |
+
"rows": 171,
|
| 626 |
+
"decisions": 171
|
| 627 |
+
},
|
| 628 |
+
"jev_bench_tr_en/clinc150": {
|
| 629 |
+
"rows": 96,
|
| 630 |
+
"decisions": 96
|
| 631 |
+
},
|
| 632 |
+
"jev_bench_tr_en/fever_evidence": {
|
| 633 |
+
"rows": 91,
|
| 634 |
+
"decisions": 91
|
| 635 |
+
},
|
| 636 |
+
"jev_bench_tr_en/go_emotions": {
|
| 637 |
+
"rows": 93,
|
| 638 |
+
"decisions": 93
|
| 639 |
+
},
|
| 640 |
+
"jev_bench_tr_en/helpsteer2_helpfulness": {
|
| 641 |
+
"rows": 89,
|
| 642 |
+
"decisions": 89
|
| 643 |
+
},
|
| 644 |
+
"jev_bench_tr_en/helpsteer2_verbosity": {
|
| 645 |
+
"rows": 79,
|
| 646 |
+
"decisions": 79
|
| 647 |
+
},
|
| 648 |
+
"jev_bench_tr_en/ledgar": {
|
| 649 |
+
"rows": 70,
|
| 650 |
+
"decisions": 70
|
| 651 |
+
},
|
| 652 |
+
"jev_bench_tr_en/massive": {
|
| 653 |
+
"rows": 95,
|
| 654 |
+
"decisions": 95
|
| 655 |
+
},
|
| 656 |
+
"jev_bench_tr_en/measuring_hate_speech": {
|
| 657 |
+
"rows": 96,
|
| 658 |
+
"decisions": 96
|
| 659 |
+
},
|
| 660 |
+
"jev_bench_tr_en/mmlu": {
|
| 661 |
+
"rows": 80,
|
| 662 |
+
"decisions": 80
|
| 663 |
+
},
|
| 664 |
+
"jev_bench_tr_en/mnli": {
|
| 665 |
+
"rows": 92,
|
| 666 |
+
"decisions": 92
|
| 667 |
+
},
|
| 668 |
+
"jev_bench_tr_en/paws": {
|
| 669 |
+
"rows": 80,
|
| 670 |
+
"decisions": 80
|
| 671 |
+
},
|
| 672 |
+
"jev_bench_tr_en/sms_spam": {
|
| 673 |
+
"rows": 82,
|
| 674 |
+
"decisions": 82
|
| 675 |
+
},
|
| 676 |
+
"jev_bench_tr_en/sst5": {
|
| 677 |
+
"rows": 73,
|
| 678 |
+
"decisions": 73
|
| 679 |
+
},
|
| 680 |
+
"jev_bench_tr_en/strategyqa_closed": {
|
| 681 |
+
"rows": 60,
|
| 682 |
+
"decisions": 60
|
| 683 |
+
},
|
| 684 |
+
"jev_bench_tr_en/strategyqa_grounded": {
|
| 685 |
+
"rows": 71,
|
| 686 |
+
"decisions": 71
|
| 687 |
+
},
|
| 688 |
+
"jev_bench_tr_en/stsb": {
|
| 689 |
+
"rows": 90,
|
| 690 |
+
"decisions": 90
|
| 691 |
+
},
|
| 692 |
+
"jev_bench_tr_en/yelp5": {
|
| 693 |
+
"rows": 99,
|
| 694 |
+
"decisions": 99
|
| 695 |
+
}
|
| 696 |
+
}
|
| 697 |
+
}
|
| 698 |
+
},
|
| 699 |
+
"token_lengths": {
|
| 700 |
+
"train": {
|
| 701 |
+
"items": 305812,
|
| 702 |
+
"tokens_p50": 188,
|
| 703 |
+
"tokens_p90": 521,
|
| 704 |
+
"tokens_p99": 1474,
|
| 705 |
+
"tokens_max": 7430,
|
| 706 |
+
"tokens_total": 84154177
|
| 707 |
+
},
|
| 708 |
+
"holdout": {
|
| 709 |
+
"items": 1571,
|
| 710 |
+
"tokens_p50": 205,
|
| 711 |
+
"tokens_p90": 533,
|
| 712 |
+
"tokens_p99": 1133,
|
| 713 |
+
"tokens_max": 6016,
|
| 714 |
+
"tokens_total": 444056
|
| 715 |
+
},
|
| 716 |
+
"test": {
|
| 717 |
+
"items": 101716,
|
| 718 |
+
"tokens_p50": 187,
|
| 719 |
+
"tokens_p90": 657,
|
| 720 |
+
"tokens_p99": 1258,
|
| 721 |
+
"tokens_max": 7616,
|
| 722 |
+
"tokens_total": 29589080
|
| 723 |
+
}
|
| 724 |
+
},
|
| 725 |
+
"timeline": {
|
| 726 |
+
"start": "16:01:56",
|
| 727 |
+
"data_built": "16:03:13",
|
| 728 |
+
"tokenized": "16:05:25",
|
| 729 |
+
"training_start": "16:05:33",
|
| 730 |
+
"epoch1_validation": "16:51:47",
|
| 731 |
+
"training_end": "17:39:10",
|
| 732 |
+
"exported": "17:39:50",
|
| 733 |
+
"eval_finetuned": "17:41:38",
|
| 734 |
+
"eval_base": "17:43:27",
|
| 735 |
+
"pushed": "17:44:00"
|
| 736 |
+
},
|
| 737 |
+
"log_lines": [
|
| 738 |
+
"16:01:56 python 3.11.16, torch 2.14.1+cu130, GPUs: NVIDIA GeForce RTX 5090 (31 GB)",
|
| 739 |
+
"16:05:25 items: train 305,812, held out 1,571 (validation + calibration), test 101,716",
|
| 740 |
+
"16:05:33 training: 305,812 items, 1 process(es) on cuda, precision torch.bfloat16, 38,227 micro-batches of 8 per rank per epoch, 19,114 updates",
|
| 741 |
+
"16:51:47 validation epoch_end=1 val_n=1571 val_accuracy=0.8027 val_brier=0.273 val_nll=0.5038 val_ece=0.0505",
|
| 742 |
+
"17:39:09 validation epoch_end=2 val_n=1571 val_accuracy=0.809 val_brier=0.273 val_nll=0.5066 val_ece=0.0529",
|
| 743 |
+
"17:39:49 calibration on 1,571 held-out items: temperatures (choice, score, noul) = [0.898, 1.078, 0.95]",
|
| 744 |
+
"17:44:00 pushed: https://huggingface.co/hayriyigit/laya-tr-test-v2 (022598fcfd60)"
|
| 745 |
+
],
|
| 746 |
+
"packages": {
|
| 747 |
+
"huggingface-hub": "1.33.0",
|
| 748 |
+
"laya": "0.3.24",
|
| 749 |
+
"numpy": "2.4.6",
|
| 750 |
+
"pyarrow": "25.0.1",
|
| 751 |
+
"safetensors": "0.8.0",
|
| 752 |
+
"torch": "2.14.1",
|
| 753 |
+
"transformers": "5.18.0"
|
| 754 |
+
},
|
| 755 |
+
"hardware": {
|
| 756 |
+
"gpu": "NVIDIA GeForce RTX 5090, 32607 MiB, 595.58.03",
|
| 757 |
+
"python": "3.11.16"
|
| 758 |
+
}
|
| 759 |
+
}
|
training/train_log.jsonl
ADDED
|
@@ -0,0 +1,384 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{"epoch": 1, "update": 50, "of": 19114, "loss": 0.9676, "reward": -0.7282, "lr": 1e-05, "updates_per_s": 3.703, "eta_h": 1.43}
|
| 2 |
+
{"epoch": 1, "update": 100, "of": 19114, "loss": 0.8659, "reward": -0.5494, "lr": 1e-05, "updates_per_s": 3.822, "eta_h": 1.38}
|
| 3 |
+
{"epoch": 1, "update": 150, "of": 19114, "loss": 0.7264, "reward": -0.462, "lr": 1e-05, "updates_per_s": 3.976, "eta_h": 1.32}
|
| 4 |
+
{"epoch": 1, "update": 200, "of": 19114, "loss": 0.779, "reward": -0.5027, "lr": 1e-05, "updates_per_s": 3.78, "eta_h": 1.39}
|
| 5 |
+
{"epoch": 1, "update": 250, "of": 19114, "loss": 0.7802, "reward": -0.4722, "lr": 1e-05, "updates_per_s": 3.791, "eta_h": 1.38}
|
| 6 |
+
{"epoch": 1, "update": 300, "of": 19114, "loss": 0.7498, "reward": -0.4015, "lr": 9.99e-06, "updates_per_s": 3.857, "eta_h": 1.36}
|
| 7 |
+
{"epoch": 1, "update": 350, "of": 19114, "loss": 0.6773, "reward": -0.3672, "lr": 9.99e-06, "updates_per_s": 3.922, "eta_h": 1.33}
|
| 8 |
+
{"epoch": 1, "update": 400, "of": 19114, "loss": 0.7862, "reward": -0.447, "lr": 9.99e-06, "updates_per_s": 3.75, "eta_h": 1.39}
|
| 9 |
+
{"epoch": 1, "update": 450, "of": 19114, "loss": 0.733, "reward": -0.4036, "lr": 9.99e-06, "updates_per_s": 3.726, "eta_h": 1.39}
|
| 10 |
+
{"epoch": 1, "update": 500, "of": 19114, "loss": 0.7412, "reward": -0.3884, "lr": 9.98e-06, "updates_per_s": 3.642, "eta_h": 1.42}
|
| 11 |
+
{"epoch": 1, "update": 550, "of": 19114, "loss": 0.7121, "reward": -0.3799, "lr": 9.98e-06, "updates_per_s": 3.552, "eta_h": 1.45}
|
| 12 |
+
{"epoch": 1, "update": 600, "of": 19114, "loss": 0.748, "reward": -0.4066, "lr": 9.98e-06, "updates_per_s": 3.507, "eta_h": 1.47}
|
| 13 |
+
{"epoch": 1, "update": 650, "of": 19114, "loss": 0.6668, "reward": -0.3694, "lr": 9.97e-06, "updates_per_s": 3.451, "eta_h": 1.49}
|
| 14 |
+
{"epoch": 1, "update": 700, "of": 19114, "loss": 0.7151, "reward": -0.3882, "lr": 9.97e-06, "updates_per_s": 3.459, "eta_h": 1.48}
|
| 15 |
+
{"epoch": 1, "update": 750, "of": 19114, "loss": 0.7507, "reward": -0.4131, "lr": 9.97e-06, "updates_per_s": 3.503, "eta_h": 1.46}
|
| 16 |
+
{"epoch": 1, "update": 800, "of": 19114, "loss": 0.7307, "reward": -0.4029, "lr": 9.96e-06, "updates_per_s": 3.502, "eta_h": 1.45}
|
| 17 |
+
{"epoch": 1, "update": 850, "of": 19114, "loss": 0.6777, "reward": -0.3543, "lr": 9.96e-06, "updates_per_s": 3.477, "eta_h": 1.46}
|
| 18 |
+
{"epoch": 1, "update": 900, "of": 19114, "loss": 0.7035, "reward": -0.3545, "lr": 9.95e-06, "updates_per_s": 3.413, "eta_h": 1.48}
|
| 19 |
+
{"epoch": 1, "update": 950, "of": 19114, "loss": 0.6899, "reward": -0.3354, "lr": 9.95e-06, "updates_per_s": 3.434, "eta_h": 1.47}
|
| 20 |
+
{"epoch": 1, "update": 1000, "of": 19114, "loss": 0.7165, "reward": -0.3411, "lr": 9.94e-06, "updates_per_s": 3.456, "eta_h": 1.46}
|
| 21 |
+
{"epoch": 1, "update": 1050, "of": 19114, "loss": 0.6948, "reward": -0.3657, "lr": 9.93e-06, "updates_per_s": 3.472, "eta_h": 1.45}
|
| 22 |
+
{"epoch": 1, "update": 1100, "of": 19114, "loss": 0.725, "reward": -0.3556, "lr": 9.93e-06, "updates_per_s": 3.487, "eta_h": 1.43}
|
| 23 |
+
{"epoch": 1, "update": 1150, "of": 19114, "loss": 0.6866, "reward": -0.3482, "lr": 9.92e-06, "updates_per_s": 3.49, "eta_h": 1.43}
|
| 24 |
+
{"epoch": 1, "update": 1200, "of": 19114, "loss": 0.6902, "reward": -0.3472, "lr": 9.91e-06, "updates_per_s": 3.503, "eta_h": 1.42}
|
| 25 |
+
{"epoch": 1, "update": 1250, "of": 19114, "loss": 0.7313, "reward": -0.3827, "lr": 9.91e-06, "updates_per_s": 3.462, "eta_h": 1.43}
|
| 26 |
+
{"epoch": 1, "update": 1300, "of": 19114, "loss": 0.6951, "reward": -0.3273, "lr": 9.9e-06, "updates_per_s": 3.464, "eta_h": 1.43}
|
| 27 |
+
{"epoch": 1, "update": 1350, "of": 19114, "loss": 0.6582, "reward": -0.3171, "lr": 9.89e-06, "updates_per_s": 3.471, "eta_h": 1.42}
|
| 28 |
+
{"epoch": 1, "update": 1400, "of": 19114, "loss": 0.6798, "reward": -0.2925, "lr": 9.88e-06, "updates_per_s": 3.48, "eta_h": 1.41}
|
| 29 |
+
{"epoch": 1, "update": 1450, "of": 19114, "loss": 0.6946, "reward": -0.341, "lr": 9.87e-06, "updates_per_s": 3.465, "eta_h": 1.42}
|
| 30 |
+
{"epoch": 1, "update": 1500, "of": 19114, "loss": 0.6881, "reward": -0.3315, "lr": 9.86e-06, "updates_per_s": 3.489, "eta_h": 1.4}
|
| 31 |
+
{"epoch": 1, "update": 1550, "of": 19114, "loss": 0.667, "reward": -0.3254, "lr": 9.85e-06, "updates_per_s": 3.491, "eta_h": 1.4}
|
| 32 |
+
{"epoch": 1, "update": 1600, "of": 19114, "loss": 0.7073, "reward": -0.3556, "lr": 9.85e-06, "updates_per_s": 3.475, "eta_h": 1.4}
|
| 33 |
+
{"epoch": 1, "update": 1650, "of": 19114, "loss": 0.6481, "reward": -0.3196, "lr": 9.84e-06, "updates_per_s": 3.453, "eta_h": 1.4}
|
| 34 |
+
{"epoch": 1, "update": 1700, "of": 19114, "loss": 0.6893, "reward": -0.2924, "lr": 9.83e-06, "updates_per_s": 3.456, "eta_h": 1.4}
|
| 35 |
+
{"epoch": 1, "update": 1750, "of": 19114, "loss": 0.7015, "reward": -0.346, "lr": 9.82e-06, "updates_per_s": 3.453, "eta_h": 1.4}
|
| 36 |
+
{"epoch": 1, "update": 1800, "of": 19114, "loss": 0.66, "reward": -0.309, "lr": 9.8e-06, "updates_per_s": 3.471, "eta_h": 1.39}
|
| 37 |
+
{"epoch": 1, "update": 1850, "of": 19114, "loss": 0.592, "reward": -0.2554, "lr": 9.79e-06, "updates_per_s": 3.458, "eta_h": 1.39}
|
| 38 |
+
{"epoch": 1, "update": 1900, "of": 19114, "loss": 0.6635, "reward": -0.303, "lr": 9.78e-06, "updates_per_s": 3.446, "eta_h": 1.39}
|
| 39 |
+
{"epoch": 1, "update": 1950, "of": 19114, "loss": 0.608, "reward": -0.2729, "lr": 9.77e-06, "updates_per_s": 3.45, "eta_h": 1.38}
|
| 40 |
+
{"epoch": 1, "update": 2000, "of": 19114, "loss": 0.6718, "reward": -0.2987, "lr": 9.76e-06, "updates_per_s": 3.459, "eta_h": 1.37}
|
| 41 |
+
{"epoch": 1, "update": 2050, "of": 19114, "loss": 0.6678, "reward": -0.3191, "lr": 9.75e-06, "updates_per_s": 3.462, "eta_h": 1.37}
|
| 42 |
+
{"epoch": 1, "update": 2100, "of": 19114, "loss": 0.6161, "reward": -0.2603, "lr": 9.73e-06, "updates_per_s": 3.436, "eta_h": 1.38}
|
| 43 |
+
{"epoch": 1, "update": 2150, "of": 19114, "loss": 0.6726, "reward": -0.2775, "lr": 9.72e-06, "updates_per_s": 3.436, "eta_h": 1.37}
|
| 44 |
+
{"epoch": 1, "update": 2200, "of": 19114, "loss": 0.6671, "reward": -0.2788, "lr": 9.71e-06, "updates_per_s": 3.43, "eta_h": 1.37}
|
| 45 |
+
{"epoch": 1, "update": 2250, "of": 19114, "loss": 0.5802, "reward": -0.2316, "lr": 9.7e-06, "updates_per_s": 3.445, "eta_h": 1.36}
|
| 46 |
+
{"epoch": 1, "update": 2300, "of": 19114, "loss": 0.6089, "reward": -0.2749, "lr": 9.68e-06, "updates_per_s": 3.442, "eta_h": 1.36}
|
| 47 |
+
{"epoch": 1, "update": 2350, "of": 19114, "loss": 0.6308, "reward": -0.2401, "lr": 9.67e-06, "updates_per_s": 3.448, "eta_h": 1.35}
|
| 48 |
+
{"epoch": 1, "update": 2400, "of": 19114, "loss": 0.6242, "reward": -0.2926, "lr": 9.65e-06, "updates_per_s": 3.433, "eta_h": 1.35}
|
| 49 |
+
{"epoch": 1, "update": 2450, "of": 19114, "loss": 0.7232, "reward": -0.31, "lr": 9.64e-06, "updates_per_s": 3.409, "eta_h": 1.36}
|
| 50 |
+
{"epoch": 1, "update": 2500, "of": 19114, "loss": 0.5985, "reward": -0.2206, "lr": 9.63e-06, "updates_per_s": 3.413, "eta_h": 1.35}
|
| 51 |
+
{"epoch": 1, "update": 2550, "of": 19114, "loss": 0.6211, "reward": -0.258, "lr": 9.61e-06, "updates_per_s": 3.421, "eta_h": 1.34}
|
| 52 |
+
{"epoch": 1, "update": 2600, "of": 19114, "loss": 0.6457, "reward": -0.2639, "lr": 9.6e-06, "updates_per_s": 3.431, "eta_h": 1.34}
|
| 53 |
+
{"epoch": 1, "update": 2650, "of": 19114, "loss": 0.5989, "reward": -0.2568, "lr": 9.58e-06, "updates_per_s": 3.432, "eta_h": 1.33}
|
| 54 |
+
{"epoch": 1, "update": 2700, "of": 19114, "loss": 0.6808, "reward": -0.2918, "lr": 9.56e-06, "updates_per_s": 3.432, "eta_h": 1.33}
|
| 55 |
+
{"epoch": 1, "update": 2750, "of": 19114, "loss": 0.6144, "reward": -0.2789, "lr": 9.55e-06, "updates_per_s": 3.417, "eta_h": 1.33}
|
| 56 |
+
{"epoch": 1, "update": 2800, "of": 19114, "loss": 0.6046, "reward": -0.2002, "lr": 9.53e-06, "updates_per_s": 3.413, "eta_h": 1.33}
|
| 57 |
+
{"epoch": 1, "update": 2850, "of": 19114, "loss": 0.6271, "reward": -0.2754, "lr": 9.52e-06, "updates_per_s": 3.392, "eta_h": 1.33}
|
| 58 |
+
{"epoch": 1, "update": 2900, "of": 19114, "loss": 0.6387, "reward": -0.2986, "lr": 9.5e-06, "updates_per_s": 3.393, "eta_h": 1.33}
|
| 59 |
+
{"epoch": 1, "update": 2950, "of": 19114, "loss": 0.6033, "reward": -0.2581, "lr": 9.48e-06, "updates_per_s": 3.384, "eta_h": 1.33}
|
| 60 |
+
{"epoch": 1, "update": 3000, "of": 19114, "loss": 0.6385, "reward": -0.2789, "lr": 9.46e-06, "updates_per_s": 3.382, "eta_h": 1.32}
|
| 61 |
+
{"epoch": 1, "update": 3050, "of": 19114, "loss": 0.6042, "reward": -0.2416, "lr": 9.45e-06, "updates_per_s": 3.389, "eta_h": 1.32}
|
| 62 |
+
{"epoch": 1, "update": 3100, "of": 19114, "loss": 0.6209, "reward": -0.3178, "lr": 9.43e-06, "updates_per_s": 3.391, "eta_h": 1.31}
|
| 63 |
+
{"epoch": 1, "update": 3150, "of": 19114, "loss": 0.6097, "reward": -0.254, "lr": 9.41e-06, "updates_per_s": 3.394, "eta_h": 1.31}
|
| 64 |
+
{"epoch": 1, "update": 3200, "of": 19114, "loss": 0.5939, "reward": -0.2349, "lr": 9.39e-06, "updates_per_s": 3.388, "eta_h": 1.3}
|
| 65 |
+
{"epoch": 1, "update": 3250, "of": 19114, "loss": 0.6371, "reward": -0.2833, "lr": 9.37e-06, "updates_per_s": 3.395, "eta_h": 1.3}
|
| 66 |
+
{"epoch": 1, "update": 3300, "of": 19114, "loss": 0.6349, "reward": -0.2234, "lr": 9.35e-06, "updates_per_s": 3.397, "eta_h": 1.29}
|
| 67 |
+
{"epoch": 1, "update": 3350, "of": 19114, "loss": 0.6257, "reward": -0.2581, "lr": 9.33e-06, "updates_per_s": 3.403, "eta_h": 1.29}
|
| 68 |
+
{"epoch": 1, "update": 3400, "of": 19114, "loss": 0.6179, "reward": -0.2864, "lr": 9.32e-06, "updates_per_s": 3.398, "eta_h": 1.28}
|
| 69 |
+
{"epoch": 1, "update": 3450, "of": 19114, "loss": 0.5949, "reward": -0.2151, "lr": 9.3e-06, "updates_per_s": 3.405, "eta_h": 1.28}
|
| 70 |
+
{"epoch": 1, "update": 3500, "of": 19114, "loss": 0.5697, "reward": -0.2078, "lr": 9.28e-06, "updates_per_s": 3.398, "eta_h": 1.28}
|
| 71 |
+
{"epoch": 1, "update": 3550, "of": 19114, "loss": 0.6126, "reward": -0.2112, "lr": 9.26e-06, "updates_per_s": 3.407, "eta_h": 1.27}
|
| 72 |
+
{"epoch": 1, "update": 3600, "of": 19114, "loss": 0.6196, "reward": -0.2453, "lr": 9.23e-06, "updates_per_s": 3.41, "eta_h": 1.26}
|
| 73 |
+
{"epoch": 1, "update": 3650, "of": 19114, "loss": 0.5694, "reward": -0.2399, "lr": 9.21e-06, "updates_per_s": 3.396, "eta_h": 1.26}
|
| 74 |
+
{"epoch": 1, "update": 3700, "of": 19114, "loss": 0.6384, "reward": -0.276, "lr": 9.19e-06, "updates_per_s": 3.399, "eta_h": 1.26}
|
| 75 |
+
{"epoch": 1, "update": 3750, "of": 19114, "loss": 0.5989, "reward": -0.2065, "lr": 9.17e-06, "updates_per_s": 3.396, "eta_h": 1.26}
|
| 76 |
+
{"epoch": 1, "update": 3800, "of": 19114, "loss": 0.5993, "reward": -0.2637, "lr": 9.15e-06, "updates_per_s": 3.396, "eta_h": 1.25}
|
| 77 |
+
{"epoch": 1, "update": 3850, "of": 19114, "loss": 0.5763, "reward": -0.1884, "lr": 9.13e-06, "updates_per_s": 3.398, "eta_h": 1.25}
|
| 78 |
+
{"epoch": 1, "update": 3900, "of": 19114, "loss": 0.6119, "reward": -0.1813, "lr": 9.11e-06, "updates_per_s": 3.401, "eta_h": 1.24}
|
| 79 |
+
{"epoch": 1, "update": 3950, "of": 19114, "loss": 0.5939, "reward": -0.2282, "lr": 9.08e-06, "updates_per_s": 3.401, "eta_h": 1.24}
|
| 80 |
+
{"epoch": 1, "update": 4000, "of": 19114, "loss": 0.5774, "reward": -0.2366, "lr": 9.06e-06, "updates_per_s": 3.399, "eta_h": 1.24}
|
| 81 |
+
{"epoch": 1, "update": 4050, "of": 19114, "loss": 0.6067, "reward": -0.2169, "lr": 9.04e-06, "updates_per_s": 3.404, "eta_h": 1.23}
|
| 82 |
+
{"epoch": 1, "update": 4100, "of": 19114, "loss": 0.5673, "reward": -0.2407, "lr": 9.02e-06, "updates_per_s": 3.41, "eta_h": 1.22}
|
| 83 |
+
{"epoch": 1, "update": 4150, "of": 19114, "loss": 0.6036, "reward": -0.2475, "lr": 8.99e-06, "updates_per_s": 3.42, "eta_h": 1.22}
|
| 84 |
+
{"epoch": 1, "update": 4200, "of": 19114, "loss": 0.5861, "reward": -0.2058, "lr": 8.97e-06, "updates_per_s": 3.428, "eta_h": 1.21}
|
| 85 |
+
{"epoch": 1, "update": 4250, "of": 19114, "loss": 0.5705, "reward": -0.2101, "lr": 8.95e-06, "updates_per_s": 3.434, "eta_h": 1.2}
|
| 86 |
+
{"epoch": 1, "update": 4300, "of": 19114, "loss": 0.6129, "reward": -0.2695, "lr": 8.92e-06, "updates_per_s": 3.429, "eta_h": 1.2}
|
| 87 |
+
{"epoch": 1, "update": 4350, "of": 19114, "loss": 0.5928, "reward": -0.2349, "lr": 8.9e-06, "updates_per_s": 3.426, "eta_h": 1.2}
|
| 88 |
+
{"epoch": 1, "update": 4400, "of": 19114, "loss": 0.621, "reward": -0.2424, "lr": 8.87e-06, "updates_per_s": 3.433, "eta_h": 1.19}
|
| 89 |
+
{"epoch": 1, "update": 4450, "of": 19114, "loss": 0.5972, "reward": -0.2167, "lr": 8.85e-06, "updates_per_s": 3.426, "eta_h": 1.19}
|
| 90 |
+
{"epoch": 1, "update": 4500, "of": 19114, "loss": 0.6053, "reward": -0.2342, "lr": 8.82e-06, "updates_per_s": 3.42, "eta_h": 1.19}
|
| 91 |
+
{"epoch": 1, "update": 4550, "of": 19114, "loss": 0.5865, "reward": -0.2232, "lr": 8.8e-06, "updates_per_s": 3.423, "eta_h": 1.18}
|
| 92 |
+
{"epoch": 1, "update": 4600, "of": 19114, "loss": 0.5574, "reward": -0.1736, "lr": 8.77e-06, "updates_per_s": 3.421, "eta_h": 1.18}
|
| 93 |
+
{"epoch": 1, "update": 4650, "of": 19114, "loss": 0.5367, "reward": -0.1848, "lr": 8.75e-06, "updates_per_s": 3.423, "eta_h": 1.17}
|
| 94 |
+
{"epoch": 1, "update": 4700, "of": 19114, "loss": 0.5776, "reward": -0.197, "lr": 8.72e-06, "updates_per_s": 3.423, "eta_h": 1.17}
|
| 95 |
+
{"epoch": 1, "update": 4750, "of": 19114, "loss": 0.5766, "reward": -0.2082, "lr": 8.7e-06, "updates_per_s": 3.426, "eta_h": 1.16}
|
| 96 |
+
{"epoch": 1, "update": 4800, "of": 19114, "loss": 0.5253, "reward": -0.1799, "lr": 8.67e-06, "updates_per_s": 3.418, "eta_h": 1.16}
|
| 97 |
+
{"epoch": 1, "update": 4850, "of": 19114, "loss": 0.5574, "reward": -0.1865, "lr": 8.64e-06, "updates_per_s": 3.423, "eta_h": 1.16}
|
| 98 |
+
{"epoch": 1, "update": 4900, "of": 19114, "loss": 0.5506, "reward": -0.208, "lr": 8.62e-06, "updates_per_s": 3.425, "eta_h": 1.15}
|
| 99 |
+
{"epoch": 1, "update": 4950, "of": 19114, "loss": 0.5555, "reward": -0.196, "lr": 8.59e-06, "updates_per_s": 3.427, "eta_h": 1.15}
|
| 100 |
+
{"epoch": 1, "update": 5000, "of": 19114, "loss": 0.5671, "reward": -0.1792, "lr": 8.56e-06, "updates_per_s": 3.429, "eta_h": 1.14}
|
| 101 |
+
{"epoch": 1, "update": 5050, "of": 19114, "loss": 0.5231, "reward": -0.1531, "lr": 8.54e-06, "updates_per_s": 3.434, "eta_h": 1.14}
|
| 102 |
+
{"epoch": 1, "update": 5100, "of": 19114, "loss": 0.5953, "reward": -0.2469, "lr": 8.51e-06, "updates_per_s": 3.432, "eta_h": 1.13}
|
| 103 |
+
{"epoch": 1, "update": 5150, "of": 19114, "loss": 0.4801, "reward": -0.1816, "lr": 8.48e-06, "updates_per_s": 3.43, "eta_h": 1.13}
|
| 104 |
+
{"epoch": 1, "update": 5200, "of": 19114, "loss": 0.5941, "reward": -0.2, "lr": 8.45e-06, "updates_per_s": 3.432, "eta_h": 1.13}
|
| 105 |
+
{"epoch": 1, "update": 5250, "of": 19114, "loss": 0.6068, "reward": -0.1891, "lr": 8.43e-06, "updates_per_s": 3.435, "eta_h": 1.12}
|
| 106 |
+
{"epoch": 1, "update": 5300, "of": 19114, "loss": 0.5488, "reward": -0.1877, "lr": 8.4e-06, "updates_per_s": 3.434, "eta_h": 1.12}
|
| 107 |
+
{"epoch": 1, "update": 5350, "of": 19114, "loss": 0.5284, "reward": -0.1544, "lr": 8.37e-06, "updates_per_s": 3.431, "eta_h": 1.11}
|
| 108 |
+
{"epoch": 1, "update": 5400, "of": 19114, "loss": 0.5725, "reward": -0.2085, "lr": 8.34e-06, "updates_per_s": 3.43, "eta_h": 1.11}
|
| 109 |
+
{"epoch": 1, "update": 5450, "of": 19114, "loss": 0.5441, "reward": -0.1881, "lr": 8.31e-06, "updates_per_s": 3.434, "eta_h": 1.11}
|
| 110 |
+
{"epoch": 1, "update": 5500, "of": 19114, "loss": 0.5883, "reward": -0.195, "lr": 8.28e-06, "updates_per_s": 3.428, "eta_h": 1.1}
|
| 111 |
+
{"epoch": 1, "update": 5550, "of": 19114, "loss": 0.5344, "reward": -0.1853, "lr": 8.25e-06, "updates_per_s": 3.424, "eta_h": 1.1}
|
| 112 |
+
{"epoch": 1, "update": 5600, "of": 19114, "loss": 0.5899, "reward": -0.2048, "lr": 8.22e-06, "updates_per_s": 3.427, "eta_h": 1.1}
|
| 113 |
+
{"epoch": 1, "update": 5650, "of": 19114, "loss": 0.5912, "reward": -0.1963, "lr": 8.2e-06, "updates_per_s": 3.425, "eta_h": 1.09}
|
| 114 |
+
{"epoch": 1, "update": 5700, "of": 19114, "loss": 0.5464, "reward": -0.1813, "lr": 8.17e-06, "updates_per_s": 3.428, "eta_h": 1.09}
|
| 115 |
+
{"epoch": 1, "update": 5750, "of": 19114, "loss": 0.6396, "reward": -0.2443, "lr": 8.14e-06, "updates_per_s": 3.425, "eta_h": 1.08}
|
| 116 |
+
{"epoch": 1, "update": 5800, "of": 19114, "loss": 0.5133, "reward": -0.1488, "lr": 8.11e-06, "updates_per_s": 3.429, "eta_h": 1.08}
|
| 117 |
+
{"epoch": 1, "update": 5850, "of": 19114, "loss": 0.5596, "reward": -0.1812, "lr": 8.08e-06, "updates_per_s": 3.425, "eta_h": 1.08}
|
| 118 |
+
{"epoch": 1, "update": 5900, "of": 19114, "loss": 0.5278, "reward": -0.1572, "lr": 8.04e-06, "updates_per_s": 3.426, "eta_h": 1.07}
|
| 119 |
+
{"epoch": 1, "update": 5950, "of": 19114, "loss": 0.5343, "reward": -0.1829, "lr": 8.01e-06, "updates_per_s": 3.426, "eta_h": 1.07}
|
| 120 |
+
{"epoch": 1, "update": 6000, "of": 19114, "loss": 0.5098, "reward": -0.1493, "lr": 7.98e-06, "updates_per_s": 3.426, "eta_h": 1.06}
|
| 121 |
+
{"epoch": 1, "update": 6050, "of": 19114, "loss": 0.5828, "reward": -0.1584, "lr": 7.95e-06, "updates_per_s": 3.425, "eta_h": 1.06}
|
| 122 |
+
{"epoch": 1, "update": 6100, "of": 19114, "loss": 0.525, "reward": -0.1536, "lr": 7.92e-06, "updates_per_s": 3.421, "eta_h": 1.06}
|
| 123 |
+
{"epoch": 1, "update": 6150, "of": 19114, "loss": 0.5693, "reward": -0.1933, "lr": 7.89e-06, "updates_per_s": 3.424, "eta_h": 1.05}
|
| 124 |
+
{"epoch": 1, "update": 6200, "of": 19114, "loss": 0.552, "reward": -0.177, "lr": 7.86e-06, "updates_per_s": 3.416, "eta_h": 1.05}
|
| 125 |
+
{"epoch": 1, "update": 6250, "of": 19114, "loss": 0.5752, "reward": -0.1984, "lr": 7.83e-06, "updates_per_s": 3.417, "eta_h": 1.05}
|
| 126 |
+
{"epoch": 1, "update": 6300, "of": 19114, "loss": 0.5667, "reward": -0.171, "lr": 7.8e-06, "updates_per_s": 3.42, "eta_h": 1.04}
|
| 127 |
+
{"epoch": 1, "update": 6350, "of": 19114, "loss": 0.519, "reward": -0.1678, "lr": 7.76e-06, "updates_per_s": 3.42, "eta_h": 1.04}
|
| 128 |
+
{"epoch": 1, "update": 6400, "of": 19114, "loss": 0.5264, "reward": -0.1737, "lr": 7.73e-06, "updates_per_s": 3.42, "eta_h": 1.03}
|
| 129 |
+
{"epoch": 1, "update": 6450, "of": 19114, "loss": 0.5363, "reward": -0.1668, "lr": 7.7e-06, "updates_per_s": 3.425, "eta_h": 1.03}
|
| 130 |
+
{"epoch": 1, "update": 6500, "of": 19114, "loss": 0.5195, "reward": -0.1593, "lr": 7.67e-06, "updates_per_s": 3.426, "eta_h": 1.02}
|
| 131 |
+
{"epoch": 1, "update": 6550, "of": 19114, "loss": 0.5635, "reward": -0.1461, "lr": 7.63e-06, "updates_per_s": 3.427, "eta_h": 1.02}
|
| 132 |
+
{"epoch": 1, "update": 6600, "of": 19114, "loss": 0.5656, "reward": -0.1851, "lr": 7.6e-06, "updates_per_s": 3.426, "eta_h": 1.01}
|
| 133 |
+
{"epoch": 1, "update": 6650, "of": 19114, "loss": 0.5465, "reward": -0.1995, "lr": 7.57e-06, "updates_per_s": 3.425, "eta_h": 1.01}
|
| 134 |
+
{"epoch": 1, "update": 6700, "of": 19114, "loss": 0.4699, "reward": -0.1336, "lr": 7.54e-06, "updates_per_s": 3.424, "eta_h": 1.01}
|
| 135 |
+
{"epoch": 1, "update": 6750, "of": 19114, "loss": 0.577, "reward": -0.1832, "lr": 7.5e-06, "updates_per_s": 3.427, "eta_h": 1.0}
|
| 136 |
+
{"epoch": 1, "update": 6800, "of": 19114, "loss": 0.5067, "reward": -0.1524, "lr": 7.47e-06, "updates_per_s": 3.43, "eta_h": 1.0}
|
| 137 |
+
{"epoch": 1, "update": 6850, "of": 19114, "loss": 0.5535, "reward": -0.1901, "lr": 7.44e-06, "updates_per_s": 3.432, "eta_h": 0.99}
|
| 138 |
+
{"epoch": 1, "update": 6900, "of": 19114, "loss": 0.4925, "reward": -0.1303, "lr": 7.4e-06, "updates_per_s": 3.437, "eta_h": 0.99}
|
| 139 |
+
{"epoch": 1, "update": 6950, "of": 19114, "loss": 0.5237, "reward": -0.1069, "lr": 7.37e-06, "updates_per_s": 3.44, "eta_h": 0.98}
|
| 140 |
+
{"epoch": 1, "update": 7000, "of": 19114, "loss": 0.475, "reward": -0.1109, "lr": 7.34e-06, "updates_per_s": 3.444, "eta_h": 0.98}
|
| 141 |
+
{"epoch": 1, "update": 7050, "of": 19114, "loss": 0.5778, "reward": -0.2136, "lr": 7.3e-06, "updates_per_s": 3.444, "eta_h": 0.97}
|
| 142 |
+
{"epoch": 1, "update": 7100, "of": 19114, "loss": 0.4967, "reward": -0.1363, "lr": 7.27e-06, "updates_per_s": 3.448, "eta_h": 0.97}
|
| 143 |
+
{"epoch": 1, "update": 7150, "of": 19114, "loss": 0.587, "reward": -0.1813, "lr": 7.23e-06, "updates_per_s": 3.448, "eta_h": 0.96}
|
| 144 |
+
{"epoch": 1, "update": 7200, "of": 19114, "loss": 0.4736, "reward": -0.1243, "lr": 7.2e-06, "updates_per_s": 3.448, "eta_h": 0.96}
|
| 145 |
+
{"epoch": 1, "update": 7250, "of": 19114, "loss": 0.5076, "reward": -0.1452, "lr": 7.17e-06, "updates_per_s": 3.451, "eta_h": 0.95}
|
| 146 |
+
{"epoch": 1, "update": 7300, "of": 19114, "loss": 0.5401, "reward": -0.1715, "lr": 7.13e-06, "updates_per_s": 3.452, "eta_h": 0.95}
|
| 147 |
+
{"epoch": 1, "update": 7350, "of": 19114, "loss": 0.5955, "reward": -0.2178, "lr": 7.1e-06, "updates_per_s": 3.453, "eta_h": 0.95}
|
| 148 |
+
{"epoch": 1, "update": 7400, "of": 19114, "loss": 0.5958, "reward": -0.2236, "lr": 7.06e-06, "updates_per_s": 3.452, "eta_h": 0.94}
|
| 149 |
+
{"epoch": 1, "update": 7450, "of": 19114, "loss": 0.5353, "reward": -0.1656, "lr": 7.03e-06, "updates_per_s": 3.453, "eta_h": 0.94}
|
| 150 |
+
{"epoch": 1, "update": 7500, "of": 19114, "loss": 0.4858, "reward": -0.1196, "lr": 6.99e-06, "updates_per_s": 3.455, "eta_h": 0.93}
|
| 151 |
+
{"epoch": 1, "update": 7550, "of": 19114, "loss": 0.5318, "reward": -0.1446, "lr": 6.96e-06, "updates_per_s": 3.459, "eta_h": 0.93}
|
| 152 |
+
{"epoch": 1, "update": 7600, "of": 19114, "loss": 0.5387, "reward": -0.202, "lr": 6.92e-06, "updates_per_s": 3.456, "eta_h": 0.93}
|
| 153 |
+
{"epoch": 1, "update": 7650, "of": 19114, "loss": 0.5277, "reward": -0.1727, "lr": 6.89e-06, "updates_per_s": 3.458, "eta_h": 0.92}
|
| 154 |
+
{"epoch": 1, "update": 7700, "of": 19114, "loss": 0.5034, "reward": -0.1312, "lr": 6.85e-06, "updates_per_s": 3.458, "eta_h": 0.92}
|
| 155 |
+
{"epoch": 1, "update": 7750, "of": 19114, "loss": 0.502, "reward": -0.1358, "lr": 6.82e-06, "updates_per_s": 3.459, "eta_h": 0.91}
|
| 156 |
+
{"epoch": 1, "update": 7800, "of": 19114, "loss": 0.5348, "reward": -0.1928, "lr": 6.78e-06, "updates_per_s": 3.45, "eta_h": 0.91}
|
| 157 |
+
{"epoch": 1, "update": 7850, "of": 19114, "loss": 0.521, "reward": -0.1694, "lr": 6.75e-06, "updates_per_s": 3.451, "eta_h": 0.91}
|
| 158 |
+
{"epoch": 1, "update": 7900, "of": 19114, "loss": 0.5682, "reward": -0.1756, "lr": 6.71e-06, "updates_per_s": 3.451, "eta_h": 0.9}
|
| 159 |
+
{"epoch": 1, "update": 7950, "of": 19114, "loss": 0.538, "reward": -0.1745, "lr": 6.67e-06, "updates_per_s": 3.455, "eta_h": 0.9}
|
| 160 |
+
{"epoch": 1, "update": 8000, "of": 19114, "loss": 0.5322, "reward": -0.1127, "lr": 6.64e-06, "updates_per_s": 3.459, "eta_h": 0.89}
|
| 161 |
+
{"epoch": 1, "update": 8050, "of": 19114, "loss": 0.556, "reward": -0.1719, "lr": 6.6e-06, "updates_per_s": 3.46, "eta_h": 0.89}
|
| 162 |
+
{"epoch": 1, "update": 8100, "of": 19114, "loss": 0.6277, "reward": -0.1939, "lr": 6.57e-06, "updates_per_s": 3.453, "eta_h": 0.89}
|
| 163 |
+
{"epoch": 1, "update": 8150, "of": 19114, "loss": 0.5259, "reward": -0.1392, "lr": 6.53e-06, "updates_per_s": 3.456, "eta_h": 0.88}
|
| 164 |
+
{"epoch": 1, "update": 8200, "of": 19114, "loss": 0.5939, "reward": -0.1955, "lr": 6.5e-06, "updates_per_s": 3.455, "eta_h": 0.88}
|
| 165 |
+
{"epoch": 1, "update": 8250, "of": 19114, "loss": 0.5003, "reward": -0.1354, "lr": 6.46e-06, "updates_per_s": 3.458, "eta_h": 0.87}
|
| 166 |
+
{"epoch": 1, "update": 8300, "of": 19114, "loss": 0.562, "reward": -0.1574, "lr": 6.42e-06, "updates_per_s": 3.46, "eta_h": 0.87}
|
| 167 |
+
{"epoch": 1, "update": 8350, "of": 19114, "loss": 0.5196, "reward": -0.1205, "lr": 6.39e-06, "updates_per_s": 3.459, "eta_h": 0.86}
|
| 168 |
+
{"epoch": 1, "update": 8400, "of": 19114, "loss": 0.5116, "reward": -0.1389, "lr": 6.35e-06, "updates_per_s": 3.46, "eta_h": 0.86}
|
| 169 |
+
{"epoch": 1, "update": 8450, "of": 19114, "loss": 0.4929, "reward": -0.1736, "lr": 6.31e-06, "updates_per_s": 3.458, "eta_h": 0.86}
|
| 170 |
+
{"epoch": 1, "update": 8500, "of": 19114, "loss": 0.5228, "reward": -0.1208, "lr": 6.28e-06, "updates_per_s": 3.458, "eta_h": 0.85}
|
| 171 |
+
{"epoch": 1, "update": 8550, "of": 19114, "loss": 0.5238, "reward": -0.1638, "lr": 6.24e-06, "updates_per_s": 3.456, "eta_h": 0.85}
|
| 172 |
+
{"epoch": 1, "update": 8600, "of": 19114, "loss": 0.465, "reward": -0.1014, "lr": 6.2e-06, "updates_per_s": 3.457, "eta_h": 0.84}
|
| 173 |
+
{"epoch": 1, "update": 8650, "of": 19114, "loss": 0.5267, "reward": -0.1319, "lr": 6.17e-06, "updates_per_s": 3.458, "eta_h": 0.84}
|
| 174 |
+
{"epoch": 1, "update": 8700, "of": 19114, "loss": 0.5344, "reward": -0.1493, "lr": 6.13e-06, "updates_per_s": 3.453, "eta_h": 0.84}
|
| 175 |
+
{"epoch": 1, "update": 8750, "of": 19114, "loss": 0.5124, "reward": -0.1661, "lr": 6.1e-06, "updates_per_s": 3.451, "eta_h": 0.83}
|
| 176 |
+
{"epoch": 1, "update": 8800, "of": 19114, "loss": 0.4869, "reward": -0.1282, "lr": 6.06e-06, "updates_per_s": 3.45, "eta_h": 0.83}
|
| 177 |
+
{"epoch": 1, "update": 8850, "of": 19114, "loss": 0.51, "reward": -0.1366, "lr": 6.02e-06, "updates_per_s": 3.453, "eta_h": 0.83}
|
| 178 |
+
{"epoch": 1, "update": 8900, "of": 19114, "loss": 0.4986, "reward": -0.1287, "lr": 5.98e-06, "updates_per_s": 3.457, "eta_h": 0.82}
|
| 179 |
+
{"epoch": 1, "update": 8950, "of": 19114, "loss": 0.5442, "reward": -0.1459, "lr": 5.95e-06, "updates_per_s": 3.457, "eta_h": 0.82}
|
| 180 |
+
{"epoch": 1, "update": 9000, "of": 19114, "loss": 0.5285, "reward": -0.1283, "lr": 5.91e-06, "updates_per_s": 3.459, "eta_h": 0.81}
|
| 181 |
+
{"epoch": 1, "update": 9050, "of": 19114, "loss": 0.5477, "reward": -0.172, "lr": 5.87e-06, "updates_per_s": 3.456, "eta_h": 0.81}
|
| 182 |
+
{"epoch": 1, "update": 9100, "of": 19114, "loss": 0.5066, "reward": -0.1264, "lr": 5.84e-06, "updates_per_s": 3.457, "eta_h": 0.8}
|
| 183 |
+
{"epoch": 1, "update": 9150, "of": 19114, "loss": 0.52, "reward": -0.1525, "lr": 5.8e-06, "updates_per_s": 3.457, "eta_h": 0.8}
|
| 184 |
+
{"epoch": 1, "update": 9200, "of": 19114, "loss": 0.5032, "reward": -0.1318, "lr": 5.76e-06, "updates_per_s": 3.455, "eta_h": 0.8}
|
| 185 |
+
{"epoch": 1, "update": 9250, "of": 19114, "loss": 0.4917, "reward": -0.0938, "lr": 5.73e-06, "updates_per_s": 3.457, "eta_h": 0.79}
|
| 186 |
+
{"epoch": 1, "update": 9300, "of": 19114, "loss": 0.5083, "reward": -0.1573, "lr": 5.69e-06, "updates_per_s": 3.46, "eta_h": 0.79}
|
| 187 |
+
{"epoch": 1, "update": 9350, "of": 19114, "loss": 0.4839, "reward": -0.126, "lr": 5.65e-06, "updates_per_s": 3.457, "eta_h": 0.78}
|
| 188 |
+
{"epoch": 1, "update": 9400, "of": 19114, "loss": 0.4729, "reward": -0.106, "lr": 5.62e-06, "updates_per_s": 3.454, "eta_h": 0.78}
|
| 189 |
+
{"epoch": 1, "update": 9450, "of": 19114, "loss": 0.5096, "reward": -0.14, "lr": 5.58e-06, "updates_per_s": 3.45, "eta_h": 0.78}
|
| 190 |
+
{"epoch": 1, "update": 9500, "of": 19114, "loss": 0.4801, "reward": -0.1013, "lr": 5.54e-06, "updates_per_s": 3.447, "eta_h": 0.77}
|
| 191 |
+
{"epoch": 1, "update": 9550, "of": 19114, "loss": 0.4724, "reward": -0.1007, "lr": 5.51e-06, "updates_per_s": 3.449, "eta_h": 0.77}
|
| 192 |
+
{"epoch_end": 1, "val_n": 1571, "val_accuracy": 0.8027, "val_brier": 0.273, "val_nll": 0.5038, "val_ece": 0.0505}
|
| 193 |
+
{"epoch": 2, "update": 9600, "of": 19114, "loss": 0.5465, "reward": -0.0155, "lr": 5.47e-06, "updates_per_s": 3.381, "eta_h": 0.78}
|
| 194 |
+
{"epoch": 2, "update": 9650, "of": 19114, "loss": 0.582, "reward": -0.053, "lr": 5.43e-06, "updates_per_s": 3.382, "eta_h": 0.78}
|
| 195 |
+
{"epoch": 2, "update": 9700, "of": 19114, "loss": 0.5974, "reward": -0.0584, "lr": 5.39e-06, "updates_per_s": 3.378, "eta_h": 0.77}
|
| 196 |
+
{"epoch": 2, "update": 9750, "of": 19114, "loss": 0.6032, "reward": -0.0451, "lr": 5.36e-06, "updates_per_s": 3.377, "eta_h": 0.77}
|
| 197 |
+
{"epoch": 2, "update": 9800, "of": 19114, "loss": 0.5758, "reward": -0.0472, "lr": 5.32e-06, "updates_per_s": 3.381, "eta_h": 0.77}
|
| 198 |
+
{"epoch": 2, "update": 9850, "of": 19114, "loss": 0.5911, "reward": -0.0368, "lr": 5.28e-06, "updates_per_s": 3.384, "eta_h": 0.76}
|
| 199 |
+
{"epoch": 2, "update": 9900, "of": 19114, "loss": 0.5476, "reward": -0.0507, "lr": 5.25e-06, "updates_per_s": 3.378, "eta_h": 0.76}
|
| 200 |
+
{"epoch": 2, "update": 9950, "of": 19114, "loss": 0.4982, "reward": -0.0238, "lr": 5.21e-06, "updates_per_s": 3.379, "eta_h": 0.75}
|
| 201 |
+
{"epoch": 2, "update": 10000, "of": 19114, "loss": 0.5766, "reward": -0.0521, "lr": 5.17e-06, "updates_per_s": 3.378, "eta_h": 0.75}
|
| 202 |
+
{"epoch": 2, "update": 10050, "of": 19114, "loss": 0.5493, "reward": -0.0329, "lr": 5.14e-06, "updates_per_s": 3.376, "eta_h": 0.75}
|
| 203 |
+
{"epoch": 2, "update": 10100, "of": 19114, "loss": 0.636, "reward": -0.0738, "lr": 5.1e-06, "updates_per_s": 3.373, "eta_h": 0.74}
|
| 204 |
+
{"epoch": 2, "update": 10150, "of": 19114, "loss": 0.5632, "reward": -0.0443, "lr": 5.06e-06, "updates_per_s": 3.373, "eta_h": 0.74}
|
| 205 |
+
{"epoch": 2, "update": 10200, "of": 19114, "loss": 0.5705, "reward": -0.0508, "lr": 5.03e-06, "updates_per_s": 3.374, "eta_h": 0.73}
|
| 206 |
+
{"epoch": 2, "update": 10250, "of": 19114, "loss": 0.5515, "reward": -0.0428, "lr": 4.99e-06, "updates_per_s": 3.375, "eta_h": 0.73}
|
| 207 |
+
{"epoch": 2, "update": 10300, "of": 19114, "loss": 0.563, "reward": -0.0422, "lr": 4.95e-06, "updates_per_s": 3.375, "eta_h": 0.73}
|
| 208 |
+
{"epoch": 2, "update": 10350, "of": 19114, "loss": 0.5597, "reward": -0.0637, "lr": 4.92e-06, "updates_per_s": 3.374, "eta_h": 0.72}
|
| 209 |
+
{"epoch": 2, "update": 10400, "of": 19114, "loss": 0.5243, "reward": -0.0277, "lr": 4.88e-06, "updates_per_s": 3.373, "eta_h": 0.72}
|
| 210 |
+
{"epoch": 2, "update": 10450, "of": 19114, "loss": 0.6082, "reward": -0.061, "lr": 4.84e-06, "updates_per_s": 3.372, "eta_h": 0.71}
|
| 211 |
+
{"epoch": 2, "update": 10500, "of": 19114, "loss": 0.5289, "reward": -0.0448, "lr": 4.81e-06, "updates_per_s": 3.372, "eta_h": 0.71}
|
| 212 |
+
{"epoch": 2, "update": 10550, "of": 19114, "loss": 0.5814, "reward": -0.074, "lr": 4.77e-06, "updates_per_s": 3.37, "eta_h": 0.71}
|
| 213 |
+
{"epoch": 2, "update": 10600, "of": 19114, "loss": 0.531, "reward": 0.001, "lr": 4.73e-06, "updates_per_s": 3.373, "eta_h": 0.7}
|
| 214 |
+
{"epoch": 2, "update": 10650, "of": 19114, "loss": 0.539, "reward": -0.021, "lr": 4.7e-06, "updates_per_s": 3.374, "eta_h": 0.7}
|
| 215 |
+
{"epoch": 2, "update": 10700, "of": 19114, "loss": 0.5796, "reward": -0.0383, "lr": 4.66e-06, "updates_per_s": 3.376, "eta_h": 0.69}
|
| 216 |
+
{"epoch": 2, "update": 10750, "of": 19114, "loss": 0.5226, "reward": -0.022, "lr": 4.62e-06, "updates_per_s": 3.378, "eta_h": 0.69}
|
| 217 |
+
{"epoch": 2, "update": 10800, "of": 19114, "loss": 0.5274, "reward": -0.0432, "lr": 4.59e-06, "updates_per_s": 3.38, "eta_h": 0.68}
|
| 218 |
+
{"epoch": 2, "update": 10850, "of": 19114, "loss": 0.5228, "reward": -0.0153, "lr": 4.55e-06, "updates_per_s": 3.384, "eta_h": 0.68}
|
| 219 |
+
{"epoch": 2, "update": 10900, "of": 19114, "loss": 0.5779, "reward": -0.0576, "lr": 4.51e-06, "updates_per_s": 3.384, "eta_h": 0.67}
|
| 220 |
+
{"epoch": 2, "update": 10950, "of": 19114, "loss": 0.5509, "reward": -0.0509, "lr": 4.48e-06, "updates_per_s": 3.386, "eta_h": 0.67}
|
| 221 |
+
{"epoch": 2, "update": 11000, "of": 19114, "loss": 0.5499, "reward": -0.0386, "lr": 4.44e-06, "updates_per_s": 3.388, "eta_h": 0.67}
|
| 222 |
+
{"epoch": 2, "update": 11050, "of": 19114, "loss": 0.5752, "reward": -0.0416, "lr": 4.41e-06, "updates_per_s": 3.388, "eta_h": 0.66}
|
| 223 |
+
{"epoch": 2, "update": 11100, "of": 19114, "loss": 0.6068, "reward": -0.0401, "lr": 4.37e-06, "updates_per_s": 3.388, "eta_h": 0.66}
|
| 224 |
+
{"epoch": 2, "update": 11150, "of": 19114, "loss": 0.5268, "reward": -0.0113, "lr": 4.34e-06, "updates_per_s": 3.386, "eta_h": 0.65}
|
| 225 |
+
{"epoch": 2, "update": 11200, "of": 19114, "loss": 0.5504, "reward": -0.0254, "lr": 4.3e-06, "updates_per_s": 3.385, "eta_h": 0.65}
|
| 226 |
+
{"epoch": 2, "update": 11250, "of": 19114, "loss": 0.5881, "reward": -0.0372, "lr": 4.26e-06, "updates_per_s": 3.388, "eta_h": 0.64}
|
| 227 |
+
{"epoch": 2, "update": 11300, "of": 19114, "loss": 0.5705, "reward": -0.0407, "lr": 4.23e-06, "updates_per_s": 3.39, "eta_h": 0.64}
|
| 228 |
+
{"epoch": 2, "update": 11350, "of": 19114, "loss": 0.5725, "reward": -0.0461, "lr": 4.19e-06, "updates_per_s": 3.391, "eta_h": 0.64}
|
| 229 |
+
{"epoch": 2, "update": 11400, "of": 19114, "loss": 0.5763, "reward": -0.034, "lr": 4.16e-06, "updates_per_s": 3.394, "eta_h": 0.63}
|
| 230 |
+
{"epoch": 2, "update": 11450, "of": 19114, "loss": 0.5517, "reward": -0.0443, "lr": 4.12e-06, "updates_per_s": 3.392, "eta_h": 0.63}
|
| 231 |
+
{"epoch": 2, "update": 11500, "of": 19114, "loss": 0.5552, "reward": -0.0194, "lr": 4.09e-06, "updates_per_s": 3.393, "eta_h": 0.62}
|
| 232 |
+
{"epoch": 2, "update": 11550, "of": 19114, "loss": 0.5255, "reward": -0.0392, "lr": 4.05e-06, "updates_per_s": 3.394, "eta_h": 0.62}
|
| 233 |
+
{"epoch": 2, "update": 11600, "of": 19114, "loss": 0.5199, "reward": -0.0339, "lr": 4.02e-06, "updates_per_s": 3.394, "eta_h": 0.62}
|
| 234 |
+
{"epoch": 2, "update": 11650, "of": 19114, "loss": 0.5679, "reward": -0.0718, "lr": 3.98e-06, "updates_per_s": 3.393, "eta_h": 0.61}
|
| 235 |
+
{"epoch": 2, "update": 11700, "of": 19114, "loss": 0.5723, "reward": -0.0698, "lr": 3.95e-06, "updates_per_s": 3.396, "eta_h": 0.61}
|
| 236 |
+
{"epoch": 2, "update": 11750, "of": 19114, "loss": 0.5956, "reward": -0.0756, "lr": 3.91e-06, "updates_per_s": 3.396, "eta_h": 0.6}
|
| 237 |
+
{"epoch": 2, "update": 11800, "of": 19114, "loss": 0.5972, "reward": -0.0776, "lr": 3.88e-06, "updates_per_s": 3.393, "eta_h": 0.6}
|
| 238 |
+
{"epoch": 2, "update": 11850, "of": 19114, "loss": 0.5323, "reward": -0.0362, "lr": 3.84e-06, "updates_per_s": 3.394, "eta_h": 0.59}
|
| 239 |
+
{"epoch": 2, "update": 11900, "of": 19114, "loss": 0.5383, "reward": -0.0207, "lr": 3.81e-06, "updates_per_s": 3.396, "eta_h": 0.59}
|
| 240 |
+
{"epoch": 2, "update": 11950, "of": 19114, "loss": 0.5622, "reward": -0.0383, "lr": 3.78e-06, "updates_per_s": 3.393, "eta_h": 0.59}
|
| 241 |
+
{"epoch": 2, "update": 12000, "of": 19114, "loss": 0.5403, "reward": -0.0284, "lr": 3.74e-06, "updates_per_s": 3.392, "eta_h": 0.58}
|
| 242 |
+
{"epoch": 2, "update": 12050, "of": 19114, "loss": 0.5439, "reward": -0.0532, "lr": 3.71e-06, "updates_per_s": 3.39, "eta_h": 0.58}
|
| 243 |
+
{"epoch": 2, "update": 12100, "of": 19114, "loss": 0.5623, "reward": -0.0564, "lr": 3.67e-06, "updates_per_s": 3.388, "eta_h": 0.58}
|
| 244 |
+
{"epoch": 2, "update": 12150, "of": 19114, "loss": 0.6047, "reward": -0.0301, "lr": 3.64e-06, "updates_per_s": 3.388, "eta_h": 0.57}
|
| 245 |
+
{"epoch": 2, "update": 12200, "of": 19114, "loss": 0.5104, "reward": -0.036, "lr": 3.61e-06, "updates_per_s": 3.389, "eta_h": 0.57}
|
| 246 |
+
{"epoch": 2, "update": 12250, "of": 19114, "loss": 0.5518, "reward": -0.0103, "lr": 3.57e-06, "updates_per_s": 3.388, "eta_h": 0.56}
|
| 247 |
+
{"epoch": 2, "update": 12300, "of": 19114, "loss": 0.5601, "reward": -0.0344, "lr": 3.54e-06, "updates_per_s": 3.391, "eta_h": 0.56}
|
| 248 |
+
{"epoch": 2, "update": 12350, "of": 19114, "loss": 0.5691, "reward": -0.0361, "lr": 3.51e-06, "updates_per_s": 3.386, "eta_h": 0.55}
|
| 249 |
+
{"epoch": 2, "update": 12400, "of": 19114, "loss": 0.5619, "reward": -0.0339, "lr": 3.47e-06, "updates_per_s": 3.383, "eta_h": 0.55}
|
| 250 |
+
{"epoch": 2, "update": 12450, "of": 19114, "loss": 0.5797, "reward": -0.0665, "lr": 3.44e-06, "updates_per_s": 3.385, "eta_h": 0.55}
|
| 251 |
+
{"epoch": 2, "update": 12500, "of": 19114, "loss": 0.5333, "reward": -0.0503, "lr": 3.41e-06, "updates_per_s": 3.382, "eta_h": 0.54}
|
| 252 |
+
{"epoch": 2, "update": 12550, "of": 19114, "loss": 0.5093, "reward": -0.0332, "lr": 3.37e-06, "updates_per_s": 3.383, "eta_h": 0.54}
|
| 253 |
+
{"epoch": 2, "update": 12600, "of": 19114, "loss": 0.5297, "reward": 0.0076, "lr": 3.34e-06, "updates_per_s": 3.383, "eta_h": 0.53}
|
| 254 |
+
{"epoch": 2, "update": 12650, "of": 19114, "loss": 0.5468, "reward": -0.0331, "lr": 3.31e-06, "updates_per_s": 3.379, "eta_h": 0.53}
|
| 255 |
+
{"epoch": 2, "update": 12700, "of": 19114, "loss": 0.5582, "reward": -0.0384, "lr": 3.28e-06, "updates_per_s": 3.38, "eta_h": 0.53}
|
| 256 |
+
{"epoch": 2, "update": 12750, "of": 19114, "loss": 0.5362, "reward": -0.0659, "lr": 3.25e-06, "updates_per_s": 3.377, "eta_h": 0.52}
|
| 257 |
+
{"epoch": 2, "update": 12800, "of": 19114, "loss": 0.551, "reward": -0.0474, "lr": 3.21e-06, "updates_per_s": 3.379, "eta_h": 0.52}
|
| 258 |
+
{"epoch": 2, "update": 12850, "of": 19114, "loss": 0.5132, "reward": -0.0201, "lr": 3.18e-06, "updates_per_s": 3.377, "eta_h": 0.52}
|
| 259 |
+
{"epoch": 2, "update": 12900, "of": 19114, "loss": 0.4692, "reward": -0.0256, "lr": 3.15e-06, "updates_per_s": 3.374, "eta_h": 0.51}
|
| 260 |
+
{"epoch": 2, "update": 12950, "of": 19114, "loss": 0.5337, "reward": 0.0022, "lr": 3.12e-06, "updates_per_s": 3.375, "eta_h": 0.51}
|
| 261 |
+
{"epoch": 2, "update": 13000, "of": 19114, "loss": 0.541, "reward": -0.0493, "lr": 3.09e-06, "updates_per_s": 3.373, "eta_h": 0.5}
|
| 262 |
+
{"epoch": 2, "update": 13050, "of": 19114, "loss": 0.5865, "reward": -0.0471, "lr": 3.06e-06, "updates_per_s": 3.37, "eta_h": 0.5}
|
| 263 |
+
{"epoch": 2, "update": 13100, "of": 19114, "loss": 0.5663, "reward": -0.0208, "lr": 3.03e-06, "updates_per_s": 3.37, "eta_h": 0.5}
|
| 264 |
+
{"epoch": 2, "update": 13150, "of": 19114, "loss": 0.6086, "reward": -0.0851, "lr": 2.99e-06, "updates_per_s": 3.369, "eta_h": 0.49}
|
| 265 |
+
{"epoch": 2, "update": 13200, "of": 19114, "loss": 0.5441, "reward": -0.0458, "lr": 2.96e-06, "updates_per_s": 3.372, "eta_h": 0.49}
|
| 266 |
+
{"epoch": 2, "update": 13250, "of": 19114, "loss": 0.5128, "reward": -0.029, "lr": 2.93e-06, "updates_per_s": 3.372, "eta_h": 0.48}
|
| 267 |
+
{"epoch": 2, "update": 13300, "of": 19114, "loss": 0.5556, "reward": -0.0419, "lr": 2.9e-06, "updates_per_s": 3.374, "eta_h": 0.48}
|
| 268 |
+
{"epoch": 2, "update": 13350, "of": 19114, "loss": 0.5459, "reward": -0.0356, "lr": 2.87e-06, "updates_per_s": 3.373, "eta_h": 0.47}
|
| 269 |
+
{"epoch": 2, "update": 13400, "of": 19114, "loss": 0.5516, "reward": -0.0292, "lr": 2.84e-06, "updates_per_s": 3.374, "eta_h": 0.47}
|
| 270 |
+
{"epoch": 2, "update": 13450, "of": 19114, "loss": 0.5226, "reward": -0.0104, "lr": 2.81e-06, "updates_per_s": 3.377, "eta_h": 0.47}
|
| 271 |
+
{"epoch": 2, "update": 13500, "of": 19114, "loss": 0.56, "reward": -0.0074, "lr": 2.78e-06, "updates_per_s": 3.378, "eta_h": 0.46}
|
| 272 |
+
{"epoch": 2, "update": 13550, "of": 19114, "loss": 0.5346, "reward": -0.02, "lr": 2.75e-06, "updates_per_s": 3.377, "eta_h": 0.46}
|
| 273 |
+
{"epoch": 2, "update": 13600, "of": 19114, "loss": 0.5347, "reward": -0.0344, "lr": 2.72e-06, "updates_per_s": 3.379, "eta_h": 0.45}
|
| 274 |
+
{"epoch": 2, "update": 13650, "of": 19114, "loss": 0.5214, "reward": -0.0275, "lr": 2.7e-06, "updates_per_s": 3.379, "eta_h": 0.45}
|
| 275 |
+
{"epoch": 2, "update": 13700, "of": 19114, "loss": 0.5627, "reward": -0.0529, "lr": 2.67e-06, "updates_per_s": 3.38, "eta_h": 0.44}
|
| 276 |
+
{"epoch": 2, "update": 13750, "of": 19114, "loss": 0.5628, "reward": -0.0194, "lr": 2.64e-06, "updates_per_s": 3.382, "eta_h": 0.44}
|
| 277 |
+
{"epoch": 2, "update": 13800, "of": 19114, "loss": 0.5396, "reward": -0.0382, "lr": 2.61e-06, "updates_per_s": 3.382, "eta_h": 0.44}
|
| 278 |
+
{"epoch": 2, "update": 13850, "of": 19114, "loss": 0.573, "reward": -0.051, "lr": 2.58e-06, "updates_per_s": 3.384, "eta_h": 0.43}
|
| 279 |
+
{"epoch": 2, "update": 13900, "of": 19114, "loss": 0.5152, "reward": -0.0199, "lr": 2.55e-06, "updates_per_s": 3.381, "eta_h": 0.43}
|
| 280 |
+
{"epoch": 2, "update": 13950, "of": 19114, "loss": 0.5452, "reward": -0.0274, "lr": 2.53e-06, "updates_per_s": 3.383, "eta_h": 0.42}
|
| 281 |
+
{"epoch": 2, "update": 14000, "of": 19114, "loss": 0.5712, "reward": -0.0825, "lr": 2.5e-06, "updates_per_s": 3.384, "eta_h": 0.42}
|
| 282 |
+
{"epoch": 2, "update": 14050, "of": 19114, "loss": 0.5167, "reward": -0.0236, "lr": 2.47e-06, "updates_per_s": 3.385, "eta_h": 0.42}
|
| 283 |
+
{"epoch": 2, "update": 14100, "of": 19114, "loss": 0.5281, "reward": -0.0368, "lr": 2.44e-06, "updates_per_s": 3.385, "eta_h": 0.41}
|
| 284 |
+
{"epoch": 2, "update": 14150, "of": 19114, "loss": 0.551, "reward": -0.0443, "lr": 2.42e-06, "updates_per_s": 3.385, "eta_h": 0.41}
|
| 285 |
+
{"epoch": 2, "update": 14200, "of": 19114, "loss": 0.5738, "reward": -0.0309, "lr": 2.39e-06, "updates_per_s": 3.385, "eta_h": 0.4}
|
| 286 |
+
{"epoch": 2, "update": 14250, "of": 19114, "loss": 0.5781, "reward": -0.067, "lr": 2.36e-06, "updates_per_s": 3.384, "eta_h": 0.4}
|
| 287 |
+
{"epoch": 2, "update": 14300, "of": 19114, "loss": 0.5552, "reward": -0.0464, "lr": 2.34e-06, "updates_per_s": 3.381, "eta_h": 0.4}
|
| 288 |
+
{"epoch": 2, "update": 14350, "of": 19114, "loss": 0.5349, "reward": -0.0562, "lr": 2.31e-06, "updates_per_s": 3.379, "eta_h": 0.39}
|
| 289 |
+
{"epoch": 2, "update": 14400, "of": 19114, "loss": 0.5637, "reward": -0.0562, "lr": 2.28e-06, "updates_per_s": 3.378, "eta_h": 0.39}
|
| 290 |
+
{"epoch": 2, "update": 14450, "of": 19114, "loss": 0.5479, "reward": -0.0055, "lr": 2.26e-06, "updates_per_s": 3.379, "eta_h": 0.38}
|
| 291 |
+
{"epoch": 2, "update": 14500, "of": 19114, "loss": 0.5235, "reward": -0.0301, "lr": 2.23e-06, "updates_per_s": 3.381, "eta_h": 0.38}
|
| 292 |
+
{"epoch": 2, "update": 14550, "of": 19114, "loss": 0.5014, "reward": 0.001, "lr": 2.21e-06, "updates_per_s": 3.382, "eta_h": 0.37}
|
| 293 |
+
{"epoch": 2, "update": 14600, "of": 19114, "loss": 0.5165, "reward": -0.0051, "lr": 2.18e-06, "updates_per_s": 3.385, "eta_h": 0.37}
|
| 294 |
+
{"epoch": 2, "update": 14650, "of": 19114, "loss": 0.4872, "reward": -0.0288, "lr": 2.16e-06, "updates_per_s": 3.386, "eta_h": 0.37}
|
| 295 |
+
{"epoch": 2, "update": 14700, "of": 19114, "loss": 0.4837, "reward": 0.0167, "lr": 2.13e-06, "updates_per_s": 3.385, "eta_h": 0.36}
|
| 296 |
+
{"epoch": 2, "update": 14750, "of": 19114, "loss": 0.4972, "reward": -0.0095, "lr": 2.11e-06, "updates_per_s": 3.386, "eta_h": 0.36}
|
| 297 |
+
{"epoch": 2, "update": 14800, "of": 19114, "loss": 0.5448, "reward": -0.0429, "lr": 2.08e-06, "updates_per_s": 3.386, "eta_h": 0.35}
|
| 298 |
+
{"epoch": 2, "update": 14850, "of": 19114, "loss": 0.488, "reward": -0.0018, "lr": 2.06e-06, "updates_per_s": 3.387, "eta_h": 0.35}
|
| 299 |
+
{"epoch": 2, "update": 14900, "of": 19114, "loss": 0.5332, "reward": -0.042, "lr": 2.04e-06, "updates_per_s": 3.385, "eta_h": 0.35}
|
| 300 |
+
{"epoch": 2, "update": 14950, "of": 19114, "loss": 0.5557, "reward": -0.0505, "lr": 2.01e-06, "updates_per_s": 3.384, "eta_h": 0.34}
|
| 301 |
+
{"epoch": 2, "update": 15000, "of": 19114, "loss": 0.507, "reward": -0.0047, "lr": 1.99e-06, "updates_per_s": 3.382, "eta_h": 0.34}
|
| 302 |
+
{"epoch": 2, "update": 15050, "of": 19114, "loss": 0.5094, "reward": -0.012, "lr": 1.97e-06, "updates_per_s": 3.383, "eta_h": 0.33}
|
| 303 |
+
{"epoch": 2, "update": 15100, "of": 19114, "loss": 0.5306, "reward": -0.0592, "lr": 1.94e-06, "updates_per_s": 3.384, "eta_h": 0.33}
|
| 304 |
+
{"epoch": 2, "update": 15150, "of": 19114, "loss": 0.5039, "reward": 0.0147, "lr": 1.92e-06, "updates_per_s": 3.384, "eta_h": 0.33}
|
| 305 |
+
{"epoch": 2, "update": 15200, "of": 19114, "loss": 0.5239, "reward": -0.0395, "lr": 1.9e-06, "updates_per_s": 3.386, "eta_h": 0.32}
|
| 306 |
+
{"epoch": 2, "update": 15250, "of": 19114, "loss": 0.5373, "reward": -0.0281, "lr": 1.88e-06, "updates_per_s": 3.387, "eta_h": 0.32}
|
| 307 |
+
{"epoch": 2, "update": 15300, "of": 19114, "loss": 0.5113, "reward": -0.0165, "lr": 1.86e-06, "updates_per_s": 3.386, "eta_h": 0.31}
|
| 308 |
+
{"epoch": 2, "update": 15350, "of": 19114, "loss": 0.5117, "reward": -0.0096, "lr": 1.83e-06, "updates_per_s": 3.387, "eta_h": 0.31}
|
| 309 |
+
{"epoch": 2, "update": 15400, "of": 19114, "loss": 0.5676, "reward": -0.0406, "lr": 1.81e-06, "updates_per_s": 3.384, "eta_h": 0.3}
|
| 310 |
+
{"epoch": 2, "update": 15450, "of": 19114, "loss": 0.5114, "reward": 0.018, "lr": 1.79e-06, "updates_per_s": 3.383, "eta_h": 0.3}
|
| 311 |
+
{"epoch": 2, "update": 15500, "of": 19114, "loss": 0.5533, "reward": -0.0254, "lr": 1.77e-06, "updates_per_s": 3.383, "eta_h": 0.3}
|
| 312 |
+
{"epoch": 2, "update": 15550, "of": 19114, "loss": 0.5064, "reward": 0.0133, "lr": 1.75e-06, "updates_per_s": 3.384, "eta_h": 0.29}
|
| 313 |
+
{"epoch": 2, "update": 15600, "of": 19114, "loss": 0.532, "reward": -0.015, "lr": 1.73e-06, "updates_per_s": 3.383, "eta_h": 0.29}
|
| 314 |
+
{"epoch": 2, "update": 15650, "of": 19114, "loss": 0.509, "reward": 0.0276, "lr": 1.71e-06, "updates_per_s": 3.383, "eta_h": 0.28}
|
| 315 |
+
{"epoch": 2, "update": 15700, "of": 19114, "loss": 0.5305, "reward": -0.01, "lr": 1.69e-06, "updates_per_s": 3.383, "eta_h": 0.28}
|
| 316 |
+
{"epoch": 2, "update": 15750, "of": 19114, "loss": 0.5594, "reward": -0.0173, "lr": 1.67e-06, "updates_per_s": 3.385, "eta_h": 0.28}
|
| 317 |
+
{"epoch": 2, "update": 15800, "of": 19114, "loss": 0.4929, "reward": -0.0002, "lr": 1.65e-06, "updates_per_s": 3.385, "eta_h": 0.27}
|
| 318 |
+
{"epoch": 2, "update": 15850, "of": 19114, "loss": 0.5706, "reward": -0.063, "lr": 1.63e-06, "updates_per_s": 3.385, "eta_h": 0.27}
|
| 319 |
+
{"epoch": 2, "update": 15900, "of": 19114, "loss": 0.5933, "reward": -0.0575, "lr": 1.61e-06, "updates_per_s": 3.382, "eta_h": 0.26}
|
| 320 |
+
{"epoch": 2, "update": 15950, "of": 19114, "loss": 0.5291, "reward": -0.0193, "lr": 1.59e-06, "updates_per_s": 3.381, "eta_h": 0.26}
|
| 321 |
+
{"epoch": 2, "update": 16000, "of": 19114, "loss": 0.5159, "reward": -0.0096, "lr": 1.58e-06, "updates_per_s": 3.379, "eta_h": 0.26}
|
| 322 |
+
{"epoch": 2, "update": 16050, "of": 19114, "loss": 0.58, "reward": -0.0716, "lr": 1.56e-06, "updates_per_s": 3.379, "eta_h": 0.25}
|
| 323 |
+
{"epoch": 2, "update": 16100, "of": 19114, "loss": 0.4772, "reward": 0.0265, "lr": 1.54e-06, "updates_per_s": 3.377, "eta_h": 0.25}
|
| 324 |
+
{"epoch": 2, "update": 16150, "of": 19114, "loss": 0.5123, "reward": -0.0376, "lr": 1.52e-06, "updates_per_s": 3.379, "eta_h": 0.24}
|
| 325 |
+
{"epoch": 2, "update": 16200, "of": 19114, "loss": 0.522, "reward": -0.0105, "lr": 1.51e-06, "updates_per_s": 3.381, "eta_h": 0.24}
|
| 326 |
+
{"epoch": 2, "update": 16250, "of": 19114, "loss": 0.4901, "reward": -0.0031, "lr": 1.49e-06, "updates_per_s": 3.383, "eta_h": 0.24}
|
| 327 |
+
{"epoch": 2, "update": 16300, "of": 19114, "loss": 0.4937, "reward": -0.0017, "lr": 1.47e-06, "updates_per_s": 3.383, "eta_h": 0.23}
|
| 328 |
+
{"epoch": 2, "update": 16350, "of": 19114, "loss": 0.5456, "reward": -0.0386, "lr": 1.46e-06, "updates_per_s": 3.383, "eta_h": 0.23}
|
| 329 |
+
{"epoch": 2, "update": 16400, "of": 19114, "loss": 0.5895, "reward": -0.0809, "lr": 1.44e-06, "updates_per_s": 3.381, "eta_h": 0.22}
|
| 330 |
+
{"epoch": 2, "update": 16450, "of": 19114, "loss": 0.5427, "reward": -0.0139, "lr": 1.42e-06, "updates_per_s": 3.382, "eta_h": 0.22}
|
| 331 |
+
{"epoch": 2, "update": 16500, "of": 19114, "loss": 0.5146, "reward": -0.0102, "lr": 1.41e-06, "updates_per_s": 3.383, "eta_h": 0.21}
|
| 332 |
+
{"epoch": 2, "update": 16550, "of": 19114, "loss": 0.5504, "reward": 0.0044, "lr": 1.39e-06, "updates_per_s": 3.383, "eta_h": 0.21}
|
| 333 |
+
{"epoch": 2, "update": 16600, "of": 19114, "loss": 0.519, "reward": -0.006, "lr": 1.38e-06, "updates_per_s": 3.383, "eta_h": 0.21}
|
| 334 |
+
{"epoch": 2, "update": 16650, "of": 19114, "loss": 0.5146, "reward": -0.0003, "lr": 1.36e-06, "updates_per_s": 3.384, "eta_h": 0.2}
|
| 335 |
+
{"epoch": 2, "update": 16700, "of": 19114, "loss": 0.5007, "reward": 0.0041, "lr": 1.35e-06, "updates_per_s": 3.384, "eta_h": 0.2}
|
| 336 |
+
{"epoch": 2, "update": 16750, "of": 19114, "loss": 0.5343, "reward": -0.0255, "lr": 1.34e-06, "updates_per_s": 3.384, "eta_h": 0.19}
|
| 337 |
+
{"epoch": 2, "update": 16800, "of": 19114, "loss": 0.518, "reward": 0.0135, "lr": 1.32e-06, "updates_per_s": 3.386, "eta_h": 0.19}
|
| 338 |
+
{"epoch": 2, "update": 16850, "of": 19114, "loss": 0.5603, "reward": -0.0277, "lr": 1.31e-06, "updates_per_s": 3.387, "eta_h": 0.19}
|
| 339 |
+
{"epoch": 2, "update": 16900, "of": 19114, "loss": 0.5155, "reward": -0.0344, "lr": 1.29e-06, "updates_per_s": 3.387, "eta_h": 0.18}
|
| 340 |
+
{"epoch": 2, "update": 16950, "of": 19114, "loss": 0.4803, "reward": 0.0056, "lr": 1.28e-06, "updates_per_s": 3.389, "eta_h": 0.18}
|
| 341 |
+
{"epoch": 2, "update": 17000, "of": 19114, "loss": 0.5313, "reward": 0.008, "lr": 1.27e-06, "updates_per_s": 3.39, "eta_h": 0.17}
|
| 342 |
+
{"epoch": 2, "update": 17050, "of": 19114, "loss": 0.4766, "reward": 0.0082, "lr": 1.26e-06, "updates_per_s": 3.39, "eta_h": 0.17}
|
| 343 |
+
{"epoch": 2, "update": 17100, "of": 19114, "loss": 0.4634, "reward": 0.0173, "lr": 1.24e-06, "updates_per_s": 3.39, "eta_h": 0.17}
|
| 344 |
+
{"epoch": 2, "update": 17150, "of": 19114, "loss": 0.5572, "reward": -0.0462, "lr": 1.23e-06, "updates_per_s": 3.39, "eta_h": 0.16}
|
| 345 |
+
{"epoch": 2, "update": 17200, "of": 19114, "loss": 0.5642, "reward": -0.0464, "lr": 1.22e-06, "updates_per_s": 3.391, "eta_h": 0.16}
|
| 346 |
+
{"epoch": 2, "update": 17250, "of": 19114, "loss": 0.5345, "reward": -0.0071, "lr": 1.21e-06, "updates_per_s": 3.393, "eta_h": 0.15}
|
| 347 |
+
{"epoch": 2, "update": 17300, "of": 19114, "loss": 0.5374, "reward": -0.0347, "lr": 1.2e-06, "updates_per_s": 3.394, "eta_h": 0.15}
|
| 348 |
+
{"epoch": 2, "update": 17350, "of": 19114, "loss": 0.507, "reward": 0.0118, "lr": 1.19e-06, "updates_per_s": 3.395, "eta_h": 0.14}
|
| 349 |
+
{"epoch": 2, "update": 17400, "of": 19114, "loss": 0.5344, "reward": -0.0179, "lr": 1.18e-06, "updates_per_s": 3.396, "eta_h": 0.14}
|
| 350 |
+
{"epoch": 2, "update": 17450, "of": 19114, "loss": 0.5529, "reward": -0.0398, "lr": 1.17e-06, "updates_per_s": 3.393, "eta_h": 0.14}
|
| 351 |
+
{"epoch": 2, "update": 17500, "of": 19114, "loss": 0.4937, "reward": 0.0104, "lr": 1.16e-06, "updates_per_s": 3.395, "eta_h": 0.13}
|
| 352 |
+
{"epoch": 2, "update": 17550, "of": 19114, "loss": 0.5361, "reward": -0.0134, "lr": 1.15e-06, "updates_per_s": 3.396, "eta_h": 0.13}
|
| 353 |
+
{"epoch": 2, "update": 17600, "of": 19114, "loss": 0.5008, "reward": -0.0017, "lr": 1.14e-06, "updates_per_s": 3.396, "eta_h": 0.12}
|
| 354 |
+
{"epoch": 2, "update": 17650, "of": 19114, "loss": 0.5205, "reward": 0.0025, "lr": 1.13e-06, "updates_per_s": 3.395, "eta_h": 0.12}
|
| 355 |
+
{"epoch": 2, "update": 17700, "of": 19114, "loss": 0.5072, "reward": -0.0359, "lr": 1.12e-06, "updates_per_s": 3.393, "eta_h": 0.12}
|
| 356 |
+
{"epoch": 2, "update": 17750, "of": 19114, "loss": 0.4774, "reward": 0.0175, "lr": 1.11e-06, "updates_per_s": 3.394, "eta_h": 0.11}
|
| 357 |
+
{"epoch": 2, "update": 17800, "of": 19114, "loss": 0.4755, "reward": 0.0106, "lr": 1.1e-06, "updates_per_s": 3.394, "eta_h": 0.11}
|
| 358 |
+
{"epoch": 2, "update": 17850, "of": 19114, "loss": 0.5309, "reward": -0.0086, "lr": 1.1e-06, "updates_per_s": 3.395, "eta_h": 0.1}
|
| 359 |
+
{"epoch": 2, "update": 17900, "of": 19114, "loss": 0.4903, "reward": -0.0097, "lr": 1.09e-06, "updates_per_s": 3.396, "eta_h": 0.1}
|
| 360 |
+
{"epoch": 2, "update": 17950, "of": 19114, "loss": 0.5024, "reward": -0.0018, "lr": 1.08e-06, "updates_per_s": 3.397, "eta_h": 0.1}
|
| 361 |
+
{"epoch": 2, "update": 18000, "of": 19114, "loss": 0.5062, "reward": -0.0251, "lr": 1.08e-06, "updates_per_s": 3.397, "eta_h": 0.09}
|
| 362 |
+
{"epoch": 2, "update": 18050, "of": 19114, "loss": 0.4979, "reward": -0.0101, "lr": 1.07e-06, "updates_per_s": 3.398, "eta_h": 0.09}
|
| 363 |
+
{"epoch": 2, "update": 18100, "of": 19114, "loss": 0.5462, "reward": -0.05, "lr": 1.06e-06, "updates_per_s": 3.396, "eta_h": 0.08}
|
| 364 |
+
{"epoch": 2, "update": 18150, "of": 19114, "loss": 0.5155, "reward": 0.0065, "lr": 1.06e-06, "updates_per_s": 3.397, "eta_h": 0.08}
|
| 365 |
+
{"epoch": 2, "update": 18200, "of": 19114, "loss": 0.5107, "reward": -0.0104, "lr": 1.05e-06, "updates_per_s": 3.399, "eta_h": 0.07}
|
| 366 |
+
{"epoch": 2, "update": 18250, "of": 19114, "loss": 0.5273, "reward": -0.0154, "lr": 1.05e-06, "updates_per_s": 3.401, "eta_h": 0.07}
|
| 367 |
+
{"epoch": 2, "update": 18300, "of": 19114, "loss": 0.548, "reward": -0.0419, "lr": 1.04e-06, "updates_per_s": 3.401, "eta_h": 0.07}
|
| 368 |
+
{"epoch": 2, "update": 18350, "of": 19114, "loss": 0.4799, "reward": 0.0011, "lr": 1.04e-06, "updates_per_s": 3.402, "eta_h": 0.06}
|
| 369 |
+
{"epoch": 2, "update": 18400, "of": 19114, "loss": 0.5642, "reward": -0.0233, "lr": 1.03e-06, "updates_per_s": 3.402, "eta_h": 0.06}
|
| 370 |
+
{"epoch": 2, "update": 18450, "of": 19114, "loss": 0.4866, "reward": 0.0211, "lr": 1.03e-06, "updates_per_s": 3.404, "eta_h": 0.05}
|
| 371 |
+
{"epoch": 2, "update": 18500, "of": 19114, "loss": 0.5326, "reward": -0.0807, "lr": 1.02e-06, "updates_per_s": 3.405, "eta_h": 0.05}
|
| 372 |
+
{"epoch": 2, "update": 18550, "of": 19114, "loss": 0.5473, "reward": -0.0502, "lr": 1.02e-06, "updates_per_s": 3.403, "eta_h": 0.05}
|
| 373 |
+
{"epoch": 2, "update": 18600, "of": 19114, "loss": 0.5236, "reward": -0.0228, "lr": 1.02e-06, "updates_per_s": 3.404, "eta_h": 0.04}
|
| 374 |
+
{"epoch": 2, "update": 18650, "of": 19114, "loss": 0.5435, "reward": -0.0027, "lr": 1.01e-06, "updates_per_s": 3.404, "eta_h": 0.04}
|
| 375 |
+
{"epoch": 2, "update": 18700, "of": 19114, "loss": 0.4677, "reward": -0.0034, "lr": 1.01e-06, "updates_per_s": 3.402, "eta_h": 0.03}
|
| 376 |
+
{"epoch": 2, "update": 18750, "of": 19114, "loss": 0.5611, "reward": -0.0467, "lr": 1.01e-06, "updates_per_s": 3.403, "eta_h": 0.03}
|
| 377 |
+
{"epoch": 2, "update": 18800, "of": 19114, "loss": 0.5041, "reward": -0.0031, "lr": 1.01e-06, "updates_per_s": 3.403, "eta_h": 0.03}
|
| 378 |
+
{"epoch": 2, "update": 18850, "of": 19114, "loss": 0.5128, "reward": -0.0238, "lr": 1e-06, "updates_per_s": 3.402, "eta_h": 0.02}
|
| 379 |
+
{"epoch": 2, "update": 18900, "of": 19114, "loss": 0.5425, "reward": -0.0224, "lr": 1e-06, "updates_per_s": 3.404, "eta_h": 0.02}
|
| 380 |
+
{"epoch": 2, "update": 18950, "of": 19114, "loss": 0.4924, "reward": 0.0184, "lr": 1e-06, "updates_per_s": 3.405, "eta_h": 0.01}
|
| 381 |
+
{"epoch": 2, "update": 19000, "of": 19114, "loss": 0.5171, "reward": -0.0246, "lr": 1e-06, "updates_per_s": 3.403, "eta_h": 0.01}
|
| 382 |
+
{"epoch": 2, "update": 19050, "of": 19114, "loss": 0.545, "reward": -0.0529, "lr": 1e-06, "updates_per_s": 3.403, "eta_h": 0.01}
|
| 383 |
+
{"epoch": 2, "update": 19100, "of": 19114, "loss": 0.508, "reward": 0.0026, "lr": 1e-06, "updates_per_s": 3.405, "eta_h": 0.0}
|
| 384 |
+
{"epoch_end": 2, "val_n": 1571, "val_accuracy": 0.809, "val_brier": 0.273, "val_nll": 0.5066, "val_ece": 0.0529}
|