Upload 3 files

bigram.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:65710f0fda0de4d195b1e1410d43d05f6a125a9722b94920a198b169997c511b
+size 12444963

bigram_log_20251224_161222.csv

@@ -0,0 +1,11 @@
+--- CONFIGURATION ---
+hyper_params,"{'batch_size': 256, 'block_size': 128, 'max_iters': 64, 'eval_interval': 40, 'learning_rate': 0.001, 'device': 'cuda', 'eval_iters': 40, 'n_embd': 128, 'comment_n_embd': '\n1. Characters have no ""meaning"" on their own. In a word-level model, \nthe word ""Apple"" might have an embedding that represents ""fruit,"" ""red,"" or ""sweet."" \nIn your character-level model, the character \'a\' means nothing by itself. \nIt could be part of ""apple,"" ""angry,"" or ""at.""\nThe n_embd (64 in your code) acts as a ""Working Memory"" or ""Space for Context."" \nIt gives the model 64 different ""channels"" to store information about that character in its specific position.\nChannel 1: Is this character at the start of a word?\nChannel 2: Is this a Sinhala character or an English one?\nChannel 3: Is this character part of a vowel sound?\nChannel 4-64: Complex patterns the model discovers during training.\n2. The Role of n_embd as a ""Vector Space""\nWhen you tokenized your text, you turned the character \'ම\' into the integer 45. \nComputers cannot do ""math"" on the number 45 to understand language. They need a Vector.\nThe n_embd parameter defines the size of this vector. \nInstead of the model seeing ""45"", it sees a row of 64 numbers.\n3. How it enables Translation: In your Sinhala-to-English task, \nthe n_embd space is where the ""translation"" actually happens.\nImagine the 64-dimensional space as a giant map. \nDuring training, the model learns to move the vector for the Sinhala character ම \nand the English character I into regions of the map that represent ""First Person Subject.""\n ', 'n_head': 8, 'n_layer': 4, 'dropout': 0, 'temperature': 0.1, 'use_hp_values_in_code': True, 'vocab_size': 151}"
+max_iterations,64
+model_name,bigram
+
+--- TRAINING LOGS ---
+iteration,train_loss,val_loss
+-1,-1.0,2.2445454597473145
+0,2.2465827465057373,2.238518238067627
+40,2.1807796955108643,2.173189163208008
+63,2.1360888481140137,2.1259498596191406

bigram_log_20251224_171302.csv

@@ -0,0 +1,41 @@
+--- CONFIGURATION ---
+hyper_params,"{'batch_size': 256, 'block_size': 128, 'max_iters': 4096, 'eval_interval': 128, 'learning_rate': 0.001, 'device': 'cuda', 'eval_iters': 64, 'n_embd': 128, 'n_head': 8, 'n_layer': 4, 'dropout': 0, 'temperature': 0.1, 'use_hp_values_in_code': True, 'vocab_size': 151}"
+max_iterations,4096
+model_name,bigram
+
+--- TRAINING LOGS ---
+iteration,train_loss,val_loss
+-1,-1.0,2.1259498596191406
+0,2.133580446243286,2.124985694885254
+128,1.9416025876998901,1.932050108909607
+256,1.8115105628967285,1.8087146282196045
+384,1.7188208103179932,1.7210934162139893
+512,1.6488759517669678,1.658238410949707
+640,1.595409631729126,1.605808138847351
+768,1.5545389652252197,1.5685614347457886
+896,1.5171124935150146,1.5404362678527832
+1024,1.491353988647461,1.516745924949646
+1152,1.4669709205627441,1.5016977787017822
+1280,1.4447822570800781,1.477726936340332
+1408,1.4293299913406372,1.468532919883728
+1536,1.4098976850509644,1.451092004776001
+1664,1.3957042694091797,1.4430755376815796
+1792,1.3807258605957031,1.4307889938354492
+1920,1.366390347480774,1.4238680601119995
+2048,1.3562545776367188,1.415414571762085
+2176,1.346989631652832,1.408126711845398
+2304,1.34041166305542,1.4004251956939697
+2432,1.3243310451507568,1.3966668844223022
+2560,1.3223880529403687,1.3857933282852173
+2688,1.3111332654953003,1.3863270282745361
+2816,1.3031277656555176,1.3804112672805786
+2944,1.2927014827728271,1.375433325767517
+3072,1.2913367748260498,1.3715697526931763
+3200,1.282301664352417,1.3655846118927002
+3328,1.2794148921966553,1.361537218093872
+3456,1.2717729806900024,1.3585717678070068
+3584,1.2672683000564575,1.357511281967163
+3712,1.258945345878601,1.350811243057251
+3840,1.2539383172988892,1.3506314754486084
+3968,1.2512305974960327,1.350011944770813
+4095,1.2486156225204468,1.351029872894287