init commit

configs/instruction_expansion_batch.json

@@ -0,0 +1,24 @@
+{
+  "job_type": "instruction_expansion_batch",
+  "dataset": {
+    "input_path": "./train.json",
+    "output_path": "./train_extended.json",
+    "template" : "./chat_template/chat_template_kd.jinja",
+    "num_in_context_samples": 3,
+    "num_output_samples": 10
+  },
+  "models": {
+    "teacher": "teacher/Qwen/Qwen2.5-7B-Instruct/"
+  },
+  "inference":{
+    "prompt" : "Assume you are a data synthesis expert. Given a few instructions as in-context examples, you should generate a new instruction similar to the examples to support the training of large language models. You should place your answer enclosed within <answer></answer> tags. The examples are as follows:",
+    "enable_chunked_prefill": true,
+    "seed": 777,
+    "gpu_memory_utilization": 0.9,
+    "temperature": 0.8,
+    "trust_remote_code": true,
+    "enforce_eager": false,
+    "max_model_len": 4096,
+    "max_new_tokens": 512
+  }
+}