Lora models finetuned with 50 data records.

README.md

@@ -1,3 +1,10 @@
 ---
 license: other
 ---
+Finetuned LoRA model with command
+$llmtune finetune --model llama-30b-4bit --weights llama-30b-4bit.pt  --dataset data50.json --adapter alpaca-adapter-folder-30b-4bit
+
+Using first 50 records of Alpaca dataset from original dataset.json
+The training loss is almost flat, and result is clearly better than original llama-30b-4bit model.
+
+License shall follow typical Alpaca family models situation.

adapter_config.json

@@ -0,0 +1,19 @@
+{
+  "base_model_name_or_path": "",
+  "bias": "none",
+  "enable_lora": null,
+  "fan_in_fan_out": false,
+  "inference_mode": true,
+  "init_lora_weights": true,
+  "lora_alpha": 16,
+  "lora_dropout": 0.05,
+  "merge_weights": false,
+  "modules_to_save": null,
+  "peft_type": "LORA",
+  "r": 8,
+  "target_modules": [
+    "q_proj",
+    "v_proj"
+  ],
+  "task_type": "CAUSAL_LM"
+}

adapter_model.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:4508481cdce9e275cf54674256e0cb8c1036537b18fd391514cd188e3677b612
+size 51204365