update convert script for fp16

Author: ydli-ai

scripts/convert_fp32_to_fp16.bin

@@ -0,0 +1,70 @@
+"""
+  This script provides an exmaple to wrap TencentPretrain for generation.
+  Given the beginning of a text, language model generates the rest.
+"""
+import sys
+import os
+import argparse
+import torch
+import torch.nn.functional as F
+
+tencentpretrain_dir = os.path.abspath(os.path.join(os.path.dirname(__file__), ".."))
+sys.path.append(tencentpretrain_dir)
+
+from tencentpretrain.embeddings import *
+from tencentpretrain.encoders import *
+from tencentpretrain.targets import *
+from tencentpretrain.utils.constants import *
+from tencentpretrain.utils import *
+from tencentpretrain.utils.config import load_hyperparam
+from tencentpretrain.model_loader import *
+from tencentpretrain.opts import model_opts, tokenizer_opts
+
+
+class GenerateLm(torch.nn.Module):
+    def __init__(self, args):
+        super(GenerateLm, self).__init__()
+        self.embedding = Embedding(args)
+        for embedding_name in args.embedding:
+            tmp_emb = str2embedding[embedding_name](args, len(args.tokenizer.vocab))
+            self.embedding.update(tmp_emb, embedding_name)
+        self.encoder = str2encoder[args.encoder](args)
+        self.target = Target()
+        self.target.update(LmTarget(args, len(args.tokenizer.vocab)), "lm")
+
+    def forward(self, src, seg):
+        emb = self.embedding(src, seg)
+        output = self.encoder(emb, seg)
+        output = self.target.lm.output_layer(output)
+        return output
+
+
+if __name__ == '__main__':
+    parser = argparse.ArgumentParser(formatter_class=argparse.ArgumentDefaultsHelpFormatter)
+
+    model_opts(parser)
+
+    parser.add_argument("--load_model_path", default=None, type=str,
+                        help="Path of the input model.")
+    parser.add_argument("--config_path", type=str, required=True,
+                        help="Path of the config file.")
+    parser.add_argument("--output_model_path", type=str)
+
+    tokenizer_opts(parser)
+
+    args = parser.parse_args()
+
+    args.target = "lm"
+    args.batch_size = 1
+
+    args = load_hyperparam(args)
+
+    args.tokenizer = str2tokenizer[args.tokenizer](args)
+
+    model = GenerateLm(args)
+    model = load_model(model, args.load_model_path)
+
+    model.half()
+
+    torch.save(model.state_dict(), args.output_model_path)
+