End of training

README.md

@@ -3,6 +3,8 @@ license: mit
 base_model: FacebookAI/roberta-large
 tags:
 - generated_from_trainer
+datasets:
+- PiC/phrase_retrieval
 model-index:
 - name: finetuned
   results: []
@@ -13,7 +15,7 @@ should probably proofread and complete it, then remove this comment. -->
 
 # finetuned
 
-This model is a fine-tuned version of [FacebookAI/roberta-large](https://huggingface.co/FacebookAI/roberta-large) on an unknown dataset.
+This model is a fine-tuned version of [FacebookAI/roberta-large](https://huggingface.co/FacebookAI/roberta-large) on the PiC/phrase_retrieval PR-pass dataset.
 
 ## Model description
 

all_results.json

@@ -8,7 +8,7 @@
     "test_f1": 96.49995382395379,
     "total_flos": 3.764792404688486e+16,
     "train_loss": 0.30838469322403306,
-    "train_runtime": 8673.8349,
+    "train_runtime": 8673.587,
     "train_samples": 20269,
     "train_samples_per_second": 4.674,
     "train_steps_per_second": 0.584

eval_nbest_predictions.json

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:a389dafafd941f907aaff40b26f3c79e5d9a1b3d70f243caed39f4ce7b182277
-size 18817712
+oid sha256:0d6369c966495c330436c353bb278de22b8d94ed261b4a58607ff4e2957d3fae
+size 18781715

eval_predictions.json

@@ -347,7 +347,7 @@
     "2897-2": "fast transport",
     "10064-2": "small source",
     "13837-1": "absolutely no trace",
-    "12317-2": "specific reaction",
+    "12317-2": "different reaction",
     "4381-1": "party tickets",
     "11493-1": "scientific understanding",
     "794-2": "desired direction",
@@ -421,7 +421,7 @@
     "7640-2": "melting point",
     "420-2": "great question",
     "14632-1": "sound stage",
-    "3864-1": "much tension",
+    "3864-1": "optimum tension",
     "11418-2": "one presentation",
     "3470-1": "big threat",
     "1988-1": "particular neighborhood",
@@ -550,7 +550,7 @@
     "9776-2": "one or more individuals",
     "666-2": "high mark",
     "5432-2": "key driver",
-    "170-1": "traditional show",
+    "170-1": "traditional network",
     "9929-1": "path-breaking work",
     "14554-2": "Computing Center",
     "391-1": "smaller population",
@@ -558,7 +558,7 @@
     "5462-2": "old couple",
     "4782-1": "standard time",
     "4449-1": "major use",
-    "4914-2": "one goal",
+    "4914-2": "least one goal",
     "13779-2": "institutional role",
     "6015-1": "one interview",
     "13542-2": "major policy changes",
@@ -771,11 +771,11 @@
     "3175-2": "greater audience",
     "11419-2": "professional opinion",
     "13236-2": "current course",
-    "6265-1": "right-hand part",
+    "6265-1": "right hand part",
     "14754-1": "adjacent room",
     "3910-2": "greater resistance",
     "7606-2": "consistent defense",
-    "14579-2": "group's profile",
+    "14579-2": "music charts",
     "4311-1": "whole purpose",
     "6271-2": "Another chapter",
     "14433-2": "less service",
@@ -803,7 +803,7 @@
     "2428-2": "natural rate",
     "240-2": "give way",
     "14297-1": "significant pressure",
-    "10788-1": "transportation costs",
+    "10788-1": "distribution costs",
     "3623-1": "growing movement",
     "9056-2": "earliest time",
     "14891-2": "uneducated boy",
@@ -954,7 +954,7 @@
     "3155-2": "key purpose",
     "12071-2": "rising tension",
     "13284-1": "occupied land",
-    "14187-1": "portable storage",
+    "14187-1": "portable storage device",
     "12189-2": "large line",
     "13311-2": "good integrity",
     "4497-2": "show use",
@@ -1088,7 +1088,7 @@
     "2401-2": "light parts",
     "7100-2": "second largest amount",
     "9863-2": "early delivery",
-    "13683-2": "another Master",
+    "13683-2": "another Master's",
     "736-2": "higher peak",
     "1990-1": "less confidence",
     "6558-1": "right path",
@@ -1182,13 +1182,13 @@
     "1417-2": "UK practice",
     "5980-2": "classic example",
     "6705-2": "specific judge",
-    "13012-1": "small return",
+    "13012-1": "zero return",
     "11081-2": "real-world setting",
     "14454-2": "actual address",
     "6019-1": "much violence",
     "14436-2": "low block",
     "11070-2": "correct depth",
-    "401-2": "Unique Group",
+    "401-2": "Unique group",
     "14721-2": "sliding glass door",
     "10826-2": "total coverage",
     "11120-1": "momentary loss",
@@ -1221,7 +1221,7 @@
     "14987-2": "rather unusual situation",
     "6394-2": "natural desire",
     "13032-2": "good term",
-    "14366-2": "propagandistic claim",
+    "14366-2": "Suggested interpretation",
     "5426-2": "New military",
     "1257-2": "higher tendency",
     "3522-2": "personal purposes",
@@ -1264,7 +1264,7 @@
     "10993-2": "sufficient room",
     "2535-1": "abundant rain",
     "8519-1": "real facts",
-    "3088-2": "safety practices",
+    "3088-2": "safety standards",
     "2115-1": "even a majority",
     "5754-1": "challenging content",
     "6417-2": "ever-increasing problem",
@@ -1553,7 +1553,7 @@
     "1545-2": "subsequent month",
     "6984-1": "emancipation policy",
     "2299-2": "detailed set",
-    "7333-1": "Radar systems",
+    "7333-1": "controlled way",
     "3871-2": "much stronger sense",
     "13854-1": "spectacular success",
     "7355-1": "mental attitude",
@@ -1580,7 +1580,7 @@
     "1253-1": "great dash",
     "10529-2": "reduced amount",
     "13023-1": "board member",
-    "12729-1": "thin fabric",
+    "12729-1": "light fabrics",
     "9865-1": "temporal course",
     "7394-2": "particular area",
     "11071-1": "secondary concern",
@@ -1588,7 +1588,7 @@
     "13067-1": "unlimited area",
     "2600-2": "world's land",
     "6724-2": "simplistic nature",
-    "8058-1": "clear guidelines",
+    "8058-1": "specific areas",
     "10955-1": "banking facility",
     "4736-2": "external pressure",
     "14211-2": "stronger line",
@@ -1631,7 +1631,7 @@
     "7763-1": "similar order",
     "9466-1": "less ground",
     "10045-1": "whose interest",
-    "14091-2": "one additional business day",
+    "14091-2": "overnight delivery",
     "6991-1": "bus or train fare",
     "83-2": "right place",
     "4467-1": "Official practice",
@@ -1694,7 +1694,7 @@
     "5063-1": "proper use",
     "679-1": "undefined point",
     "3357-2": "hand positions",
-    "5178-1": "nominal power",
+    "5178-1": "minimum number",
     "2234-1": "greater importance",
     "14871-1": "formerly Place",
     "8920-2": "Red Lips",
@@ -1874,7 +1874,7 @@
     "13978-1": "direct economic benefit",
     "6826-2": "developing crisis",
     "8410-1": "domestic product",
-    "12391-1": "Trace amounts",
+    "12391-1": "Small but commercially important quantities",
     "5623-2": "almost the same way",
     "14505-2": "conflict situation",
     "3103-1": "tough day",
@@ -1889,7 +1889,7 @@
     "4281-1": "first such system",
     "4210-1": "largest series",
     "12945-1": "military centres",
-    "14198-1": "Part One",
+    "14198-1": "first part",
     "12265-1": "full engagement",
     "9827-2": "people's view",
     "3186-2": "fairly small part",
@@ -2053,7 +2053,7 @@
     "7793-1": "notable occurrence",
     "2936-2": "true scene",
     "3017-2": "horse racing track",
-    "14204-1": "crystal formation",
+    "14204-1": "rock origins",
     "7960-1": "body defects",
     "1927-1": "good piece",
     "12981-1": "typical approach",
@@ -2167,7 +2167,7 @@
     "6634-1": "field activity",
     "5251-1": "chief use",
     "5782-2": "former production",
-    "6580-1": "poor roast",
+    "6580-1": "light roast",
     "14572-2": "subsequent reaction",
     "1356-2": "traffic violation",
     "1209-1": "one merchant",
@@ -2221,7 +2221,7 @@
     "7648-2": "1993 edition",
     "3407-1": "unrefined form",
     "11497-2": "full competition",
-    "14459-1": "dogmatic approach",
+    "14459-1": "conservative approach",
     "6840-2": "government financing",
     "1715-2": "arduous process",
     "3784-1": "whole lives",
@@ -2354,7 +2354,7 @@
     "193-1": "perfect person",
     "11485-2": "great captain",
     "1783-2": "awkward position",
-    "9968-1": "price discrepancy",
+    "9968-1": "price discrepancies",
     "8573-2": "direct participants",
     "10296-1": "specified amount",
     "2194-1": "clear side",
@@ -2429,7 +2429,7 @@
     "1098-1": "whole race",
     "9695-2": "whole reason",
     "12614-1": "proper use",
-    "7630-1": "treacherous threat",
+    "7630-1": "bottle neck",
     "9387-2": "widely recognized name",
     "8778-1": "human plasma",
     "6274-2": "sound profiles",
@@ -2451,7 +2451,7 @@
     "9644-2": "true position",
     "1063-1": "major method",
     "14797-1": "power companies",
-    "4351-2": "greater odds",
+    "4351-2": "greater chance",
     "7260-2": "reasonable conclusion",
     "11746-2": "combat weapon",
     "346-1": "lot more people",
@@ -2768,14 +2768,14 @@
     "8500-1": "fundamental design",
     "6022-2": "great threat",
     "11013-1": "single component",
-    "8262-1": "one jockey",
+    "8262-1": "DJ audition",
     "12965-1": "black and gold border",
     "5291-2": "one league",
     "12843-2": "low-income neighborhood",
     "9450-1": "bulk feed",
     "5663-2": "peripheral way",
     "9245-2": "previous process",
-    "10814-2": "clear enunciation. Farris sings a bold tenor harmony remarkable for its projection; his powerful delivery",
+    "10814-2": "clear enunciation",
     "14932-2": "strong reason",
     "10356-2": "explicit definition",
     "14225-1": "every occasion",

finetune_logs.txt

@@ -220,3 +220,20 @@
 {'loss': 0.0894, 'grad_norm': 0.05864416062831879, 'learning_rate': 4.025256511444357e-07, 'epoch': 1.97}
 {'eval_exact_match': 95.2, 'eval_f1': 96.51480158730159, 'epoch': 1.97}
 {'train_runtime': 8673.587, 'train_samples_per_second': 4.674, 'train_steps_per_second': 0.584, 'train_loss': 0.30838469322403306, 'epoch': 2.0}
+***** train metrics *****
+  epoch                    =        2.0
+  total_flos               = 35062361GF
+  train_loss               =     0.3084
+  train_runtime            = 2:24:33.58
+  train_samples            =      20269
+  train_samples_per_second =      4.674
+  train_steps_per_second   =      0.584
+***** eval metrics *****
+  epoch            =     2.0
+  eval_exact_match = 95.3333
+  eval_f1          = 96.5898
+  eval_samples     =    3013
+***** predict metrics *****
+  predict_samples  =  5034
+  test_exact_match = 95.34
+  test_f1          =  96.5

train_results.json

@@ -2,7 +2,7 @@
     "epoch": 2.0,
     "total_flos": 3.764792404688486e+16,
     "train_loss": 0.30838469322403306,
-    "train_runtime": 8673.8349,
+    "train_runtime": 8673.587,
     "train_samples": 20269,
     "train_samples_per_second": 4.674,
     "train_steps_per_second": 0.584

trainer_state.json

@@ -663,7 +663,7 @@
       "step": 5068,
       "total_flos": 3.764792404688486e+16,
       "train_loss": 0.30838469322403306,
-      "train_runtime": 8673.8349,
+      "train_runtime": 8673.587,
       "train_samples_per_second": 4.674,
       "train_steps_per_second": 0.584
     }