vrr/kvcache-ai-ktransformers
2
0
Fork 0
mirror of https://github.com/kvcache-ai/ktransformers.git synced 2025-09-06 20:49:55 +00:00
Code Issues Projects Releases Packages Wiki Activity Actions 2

🚑️:roll back transformer.py and find that it's multiple chat hsitory have minor accurate error

Browse source
This commit is contained in:
liam 2024-11-01 11:01:30 +08:00
parent a148da2cfe
commit c2b4dc805c
1 changed files with 16 additions and 6 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

22
ktransformers/server/backend/interfaces/transformers.py
View file

@ -172,13 +172,23 @@ class TransformersInterface(BackendInterfaceBase):
new_messages[-1]["content"] += m["content"]
else:
new_messages.append(m)
# if (self.last_request_id is not None) and self.last_request_id == thread_id:
# input_ids = self.tokenizer.encode(self.tokenizer.eos_token+self.tokenizer.apply_chat_template([new_messages[-1]], return_tensors="pt",tokenize=False, add_generation_prompt=True), add_special_tokens = False, return_tensors="pt").to(self.args.device)
# else:
# input_ids = self.tokenizer.apply_chat_template(
# new_messages, return_tensors="pt", add_generation_prompt=True
# ).to(self.args.device)
input_ids = self.tokenizer.apply_chat_template(new_messages,return_tensors='pt',add_generation_prompt=True).to(self.args.device)
if (self.last_request_id is not None) and self.last_request_id == thread_id:
input_ids = self.tokenizer.encode(self.tokenizer.eos_token+self.tokenizer.apply_chat_template([new_messages[-1]], return_tensors="pt",tokenize=False, add_generation_prompt=True), add_special_tokens = False, return_tensors="pt").to(self.args.device)
else:
input_ids = self.tokenizer.apply_chat_template(
new_messages, return_tensors="pt", add_generation_prompt=True
).to(self.args.device)
x = self.generated_ids[:,:self.seq_length]
y = input_ids[:,:self.seq_length]
# We can only hope that the input_ids are the same
unequal_mask = torch.ne(x,y)
unequal_positions = torch.nonzero(unequal_mask)
num_unequal_elements = unequal_mask.sum().item()
logger.warning(f'num_unequal_elements: {num_unequal_elements}')
input_ids = input_ids[:,self.seq_length:]
logger.debug(f"get input ids of shape {input_ids.shape}")
return input_ids