fix #1176 #1177

Former-commit-id: 01d8cb1ca76e1d599fa5e970b2fcaa3ceedbfffd

src/llmtuner/tuner/core/loader.py

@@ -71,6 +71,7 @@ def load_model_and_tokenizer(
     tokenizer = AutoTokenizer.from_pretrained(
         model_args.model_name_or_path,
         use_fast=model_args.use_fast_tokenizer,
+        split_special_tokens=model_args.split_special_tokens,
         padding_side="right", # training with left-padded tensors in fp16 precision may cause overflow
         **config_kwargs
     )
@@ -184,6 +185,7 @@ def load_model_and_tokenizer(
     model = AutoModelForCausalLM.from_pretrained(
         model_to_load,
         config=config,
+        torch_dtype=getattr(config, "torch_dtype"),
         low_cpu_mem_usage=(not is_deepspeed_zero3_enabled()),
         **config_kwargs
     )