QLoRA applied #5

README.md

@@ -31,14 +31,14 @@ More information needed
 ### Training hyperparameters
 
 The following hyperparameters were used during training:
-- learning_rate: 0.0002
+- learning_rate: 0.0005
 - train_batch_size: 2
 - eval_batch_size: 8
 - seed: 42
 - optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
 - lr_scheduler_type: constant
 - lr_scheduler_warmup_ratio: 0.03
-- training_steps: 600
+- training_steps: 1500
 
 ### Training results
 

adapter_config.json

@@ -16,10 +16,10 @@
   "rank_pattern": {},
   "revision": null,
   "target_modules": [
-    "dense_h_to_4h",
     "query_key_value",
-    "dense",
-    "dense_4h_to_h"
+    "dense_h_to_4h",
+    "dense_4h_to_h",
+    "dense"
   ],
   "task_type": "QUESTION_ANSWERING"
 }

adapter_model.bin

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:a68ff053219497b38ec722aeaf22be5eb0f7f29d498b64b7a4caf94520bc744e
+oid sha256:ae2669dbfb30749bf3f9cde20df51438b5d89c38758becc5430ef5e8588e7bfa
 size 261226517

tokenizer.json

@@ -1,6 +1,11 @@
 {
   "version": "1.0",
-  "truncation": null,
+  "truncation": {
+    "direction": "Right",
+    "max_length": 384,
+    "strategy": "OnlySecond",
+    "stride": 128
+  },
   "padding": null,
   "added_tokens": [
     {

training_args.bin

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:7e12d324eceb8abe722ac5da81f53ce4305b679f8478bc9f3494dda75d394776
+oid sha256:381acb08ed9e3f6ae286c467a62d06c77d0649db8f6b24284ff79c641fce2aa9
 size 4091