Upload merge_and_push.py

merge_and_push.py

@@ -0,0 +1,56 @@
+"""
+Merge LoRA adapters into the base model and push to Hugging Face Hub.
+Run this after train.py completes successfully.
+"""
+
+import torch
+from transformers import AutoModelForCausalLM, AutoTokenizer
+from peft import PeftModel
+
+# ============================================================================
+# CONFIGURATION
+# ============================================================================
+
+BASE_MODEL = "Qwen/Qwen2.5-7B-Instruct"
+ADAPTER_PATH = "./qwen2.5-7b-telecom-intent-lora"
+MERGED_OUTPUT = "./qwen2.5-7b-telecom-intent-merged"
+HUB_MODEL_ID = "nraptisss/qwen2.5-7b-telecom-intent-merged"
+
+
+def merge_and_push():
+    """Load adapters, merge, save locally, and push to hub."""
+    print(f"Loading base model: {BASE_MODEL}")
+    base_model = AutoModelForCausalLM.from_pretrained(
+        BASE_MODEL,
+        torch_dtype=torch.float16,
+        device_map="auto",
+        trust_remote_code=True,
+    )
+
+    tokenizer = AutoTokenizer.from_pretrained(
+        BASE_MODEL,
+        trust_remote_code=True,
+    )
+    if tokenizer.pad_token is None:
+        tokenizer.pad_token = tokenizer.eos_token
+
+    print(f"Loading adapters from: {ADAPTER_PATH}")
+    model = PeftModel.from_pretrained(base_model, ADAPTER_PATH)
+
+    print("Merging adapters into base model...")
+    model = model.merge_and_unload()
+
+    print(f"Saving merged model to: {MERGED_OUTPUT}")
+    model.save_pretrained(MERGED_OUTPUT)
+    tokenizer.save_pretrained(MERGED_OUTPUT)
+
+    print(f"Pushing merged model to hub: {HUB_MODEL_ID}")
+    model.push_to_hub(HUB_MODEL_ID, private=False)
+    tokenizer.push_to_hub(HUB_MODEL_ID, private=False)
+
+    print("Done! Model available at:")
+    print(f"  https://huggingface.co/{HUB_MODEL_ID}")
+
+
+if __name__ == "__main__":
+    merge_and_push()