Add last checkpoint of the failed training format.

This first test used the original format from the convert tool, but it was later found out this caused broken context. It would work as expected from the initial prompt but the moment you asked it a question about anything in the past it would say something random.

checkpoint-9728-failed/adapter_config.json

@@ -0,0 +1,17 @@
+{
+  "base_model_name_or_path": "models/llama-13b",
+  "bias": "none",
+  "fan_in_fan_out": false,
+  "inference_mode": true,
+  "init_lora_weights": true,
+  "lora_alpha": 128,
+  "lora_dropout": 0,
+  "modules_to_save": null,
+  "peft_type": "LORA",
+  "r": 64,
+  "target_modules": [
+    "q_proj",
+    "v_proj"
+  ],
+  "task_type": "CAUSAL_LM"
+}

checkpoint-9728-failed/adapter_model.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:6fdbccce795dcac8803f9e5d01dc538261095572e7be3a416ea27f1413ee022c
+size 209772877

checkpoint-9728-failed/training_parameters.json

@@ -0,0 +1 @@
+{"lora_name": "VicUnLocked", "always_override": false, "save_steps": 500.0, "micro_batch_size": 4, "batch_size": 128, "epochs": 3.0, "learning_rate": "3e-4", "lr_scheduler_type": "constant_with_warmup", "lora_rank": 64, "lora_alpha": 128, "lora_dropout": 0, "cutoff_len": 2048, "dataset": "vicuna-all", "eval_dataset": "None", "format": "vicuna-format", "eval_steps": 100.0, "raw_text_file": "None", "overlap_len": 128, "newline_favor_len": 128, "higher_rank_limit": false, "warmup_steps": 100.0, "optimizer": "adamw_torch"}