finetune and vision demo update

README.md

@@ -217,7 +217,7 @@ with torch.no_grad():
 
 如果你想更进一步了解 GLM-4-9B 系列开源模型，本开源仓库通过以下内容为开发者提供基础的 GLM-4-9B的使用和开发代码
 
-+ [base](basic_demo/README.md): 在这里包含了
++ [basic_demo](basic_demo/README.md): 在这里包含了
     + 使用 transformers 和 vLLM 后端的交互代码
     + OpenAI API 后端交互代码
     + Batch 推理代码

README_en.md

@@ -226,7 +226,7 @@ Note: GLM-4V-9B does not support calling using vLLM method yet.
 If you want to learn more about the GLM-4-9B series open source models, this open source repository provides developers
 with basic GLM-4-9B usage and development code through the following content
 
-+ [base](basic_demo/README.md): Contains
++ [basic_demo](basic_demo/README.md): Contains
 + Interaction code using transformers and vLLM backend
 + OpenAI API backend interaction code
 + Batch reasoning code

basic_demo/trans_cli_demo.py

@@ -43,8 +43,7 @@ tokenizer = AutoTokenizer.from_pretrained(
 model = AutoModel.from_pretrained(
     MODEL_PATH,
     trust_remote_code=True,
-    device_map="auto",
-    torch_dtype=torch.bfloat16).eval()
+    device_map="auto").eval()
 
 
 class StopOnTokens(StoppingCriteria):

basic_demo/trans_cli_vision_demo.py

@@ -17,7 +17,7 @@ from transformers import (
     AutoTokenizer,
     StoppingCriteria,
     StoppingCriteriaList,
-    TextIteratorStreamer, AutoModel
+    TextIteratorStreamer, AutoModel, BitsAndBytesConfig
 )
 
 from PIL import Image
@@ -33,8 +33,17 @@ model = AutoModel.from_pretrained(
     MODEL_PATH,
     trust_remote_code=True,
     device_map="auto",
-    torch_dtype=torch.bfloat16).eval()
+    torch_dtype=torch.bfloat16
+).eval()
 
+## For INT4 inference
+# model = AutoModel.from_pretrained(
+#     MODEL_PATH,
+#     trust_remote_code=True,
+#     quantization_config=BitsAndBytesConfig(load_in_4bit=True),
+#     torch_dtype=torch.bfloat16,
+#     low_cpu_mem_usage=True
+# ).eval()
 
 class StopOnTokens(StoppingCriteria):
     def __call__(self, input_ids: torch.LongTensor, scores: torch.FloatTensor, **kwargs) -> bool:
@@ -83,7 +92,7 @@ if __name__ == "__main__":
             tokenize=True,
             return_tensors="pt",
             return_dict=True
-        ).to(model.device)
+        ).to(next(model.parameters()).device)
         streamer = TextIteratorStreamer(
             tokenizer=tokenizer,
             timeout=60,

finetune_demo/finetune.py

@@ -239,7 +239,7 @@ def process_batch(
         loss_masks = [False, False]
         for message in conv:
             message = process_message(message)
-            loss_mask_val = False if message['role'] in ('system', 'user') else True
+            loss_mask_val = False if message['role'] in ('system', 'user', 'observation') else True
             new_input_ids = tokenizer.apply_chat_template([message], tokenize=True, return_dict=False)[0][2:]
             new_loss_masks = [loss_mask_val] * len(new_input_ids)
             input_ids += new_input_ids