v3 — added self-correction + optimized hyperparams

README.md

@@ -1,9 +1,9 @@
-# NL2SQL Fine-Tuned Model v2
+# NL2SQL Fine-Tuned Model v3
 This is an updated version with self-correction and improved hyperparameters.
 
 ## Config
 {
-  "version": "v2",
+  "version": "v3",
   "base_model": "./hf_models/meta-llama--Llama-3.2-3B-Instruct",
   "merged_repo": "bgglee/NL2SQL_finetuned",
   "method": "LoRA (merged)",

finetune_config.json

@@ -1,5 +1,5 @@
 {
-  "version": "v2",
+  "version": "v3",
   "base_model": "./hf_models/meta-llama--Llama-3.2-3B-Instruct",
   "merged_repo": "bgglee/NL2SQL_finetuned",
   "method": "LoRA (merged)",

model-00001-of-00002.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:327be157f17b9c75f03b065000ad4decd2e8627dfb2873a44c582969a63377a5
+size 4965799096

model-00002-of-00002.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:27a56c7c360ac1ce923510880488bcb82e88b359bfdd32b56e6ff73e62638e98
+oid sha256:a25a2554bc877ae6efdd628c84985539c5b1542870310551a44427a6abd9448b
 size 1459729952

training_loss.csv

@@ -1,37 +1,13 @@
 step,loss
-10,6.7013
-20,1.932
-30,0.0372
-40,0.0363
-50,0.0271
-60,0.0308
-70,0.0333
-80,0.0342
-90,0.0275
-100,0.0304
-110,0.0202
-120,0.0277
-130,0.0359
-140,0.0321
-150,0.0316
-160,0.0203
-170,0.0272
-180,0.0296
-190,0.0302
-200,0.0214
-210,0.0271
-220,0.0251
-230,0.0244
-240,0.0238
-250,0.0267
-260,0.0239
-270,0.0199
-280,0.037
-290,0.0213
-300,0.0224
-310,0.0223
-320,0.0314
-330,0.0214
-340,0.0271
-350,0.0246
-360,0.0287
+10,0.0329
+20,0.0293
+30,0.0239
+40,0.0236
+50,0.0187
+60,0.0224
+70,0.0248
+80,0.0261
+90,0.0212
+100,0.0233
+110,0.0159
+120,0.0228

wrong_ids.json

@@ -0,0 +1 @@
+[]