Neutralzz commited on
Commit
2d2e0f7
1 Parent(s): 8d03773

Update README.md

Browse files
Files changed (1) hide show
  1. README.md +7 -1
README.md CHANGED
@@ -21,7 +21,7 @@ python3 embedding_convert.py \
21
  --meta_llama_pth_file /path_to_LLaMA/llama-7b/consolidated.00.pth
22
  ```
23
 
24
- 模型权重还原后,可通过以下代码调试运行:
25
  ```python
26
  import torch
27
  from transformers import AutoTokenizer, AutoModelForCausalLM
@@ -42,3 +42,9 @@ output_ids = output_ids[0][len(input_ids[0]):]
42
  outputs = tokenizer.decode(output_ids, skip_special_tokens=True).strip()
43
  print(outputs)
44
  ```
 
 
 
 
 
 
 
21
  --meta_llama_pth_file /path_to_LLaMA/llama-7b/consolidated.00.pth
22
  ```
23
 
24
+ BiLLa-7B-SFT模型权重还原后,可通过以下代码调试运行:
25
  ```python
26
  import torch
27
  from transformers import AutoTokenizer, AutoModelForCausalLM
 
42
  outputs = tokenizer.decode(output_ids, skip_special_tokens=True).strip()
43
  print(outputs)
44
  ```
45
+
46
+ BiLLa-7B-SFT的模型输入需按以下格式构造(注意`Assistant:`后必须有一个空格):
47
+ ```
48
+ Human: [Your question]
49
+ Assistant:
50
+ ```