vrr/kvcache-ai-ktransformers
2
0
Fork 0
mirror of https://github.com/kvcache-ai/ktransformers.git synced 2025-09-10 06:14:58 +00:00
Code Issues Projects Releases Packages Wiki Activity Actions

⚡ support force thinking

Browse source
This commit is contained in:
liam 2025-02-12 12:43:53 +08:00
parent 6f3a39be08
commit 4385e85096
3 changed files with 7 additions and 5 deletions
Download patch file Download diff file Expand all files Collapse all files

2
ktransformers/local_chat.py
View file

@ -160,7 +160,7 @@ def local_chat(
messages, add_generation_prompt=True, return_tensors="pt"
)
if force_think:
token_thinks = torch.tensor([tokenizer.encode("<think>\\n",add_special_tokens=False)])
token_thinks = torch.tensor([tokenizer.encode("<think>\\n",add_special_tokens=False)],device=input_tensor.device)
input_tensor = torch.cat(
[input_tensor, token_thinks], dim=1
)

1
ktransformers/server/args.py
View file

@ -122,4 +122,5 @@ class ArgumentParser:
self.cfg.server_ip = args.host
self.cfg.server_port = args.port
self.cfg.backend_type = args.type
self.cfg.user_force_think = args.force_think
return args

9
ktransformers/server/backend/interfaces/transformers.py
View file

@ -325,7 +325,7 @@ class TransformersInterface(BackendInterfaceBase):
else:
raise ValueError("local_messages should be List or str")
if Config().user_force_think:
token_thinks = torch.tensor([self.tokenizer.encode("<think>\\n",add_special_tokens=False)])
token_thinks = torch.tensor([self.tokenizer.encode("<think>\\n",add_special_tokens=False)],device=input_ids.device)
input_ids = torch.cat(
[input_ids, token_thinks], dim=1
)
@ -334,8 +334,9 @@ class TransformersInterface(BackendInterfaceBase):
self.profiler.create_and_start_timer("prefill")
if Config().user_force_think:
print("<think>\n")
yield "<think>\n"
t = "<think>\n"
print(t,end="",flush=True)
yield t
for t in self.prefill(input_ids, self.check_is_new(thread_id)):
if t is not None:
print(t, end="",flush=True)
@ -346,7 +347,7 @@ class TransformersInterface(BackendInterfaceBase):
for t in self.generate():
if t is not None:
print(t, end="",flush=True)
yield t
yield t
print("")
self.profiler.pause_timer("decode")
self.report_last_time_performance()