Version 2 with 512 tokens

adapter_config.json

@@ -15,4 +15,4 @@
     "v_proj"
   ],
   "task_type": "CAUSAL_LM"
-}
+}

adapter_model.bin

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:b1d7bd29fe048e3b5924ace47cabaa1253def8bd2dd1130e01c5d5d4bd76f7d8
+oid sha256:1d4481b6cd326970d83ae7c575b7a67295c2b4a2eabfbbbf83273f9778e65fb6
 size 16822989

training_config.json

@@ -0,0 +1,35 @@
+{
+    "trainer": {
+        "evaluation_strategy": "steps",
+        "per_device_train_batch_size": 16,
+        "per_device_eval_batch_size": 16,
+        "gradient_accumulation_steps": 8,
+        "eval_steps": 150,
+        "save_steps": 150,
+        "logging_steps": 5,
+        "learning_rate": 0.0003,
+        "num_train_epochs": 3,
+        "lr_scheduler_type": "cosine",
+        "warmup_steps": 100,
+        "fp16": true,
+        "bf16": false,
+        "torch_compile": false,
+        "optim": "adamw_torch"
+    },
+    "lora": {
+        "r": 8,
+        "lora_alpha": 16,
+        "lora_dropout": 0.05,
+        "bias": "none",
+        "target_modules": ["q_proj", "v_proj"],
+        "task_type": "CAUSAL_LM"
+    },
+    "load_in_8bit": true,
+    "only_target_loss": false,
+    "model_name": "models/llama-7b-hf",
+    "model_type": "causal",
+    "template_category": "causal_newlines",
+    "max_source_tokens_count": 256,
+    "max_target_tokens_count": 512
+}
+