EngineX-Hygon/sglang
5
0
Fork 0
Code Issues Pull Requests Actions 7 Projects Releases Wiki Activity

Fix openai speculative execution (#456)

Browse Source
This commit is contained in:
Ying Sheng
2024-05-20 17:01:13 -07:00
committed by GitHub
parent ec380dfd30
commit 3e684be7a3
7 changed files with 243 additions and 128 deletions
Download Patch File Download Diff File Expand all files Collapse all files

33
examples/usage/openai_speculative.py
View File

@@ -5,7 +5,7 @@ python3 openai_speculative.py
from sglang import function, gen, set_default_backend, OpenAI
@function(api_num_spec_tokens=512)
@function(api_num_spec_tokens=64)
def gen_character_spec(s):
s += "Construct a character within the following format:\n"
s += "Name: Steve Jobs.\nBirthday: February 24, 1955.\nJob: Apple CEO.\n"
@@ -14,6 +14,15 @@ def gen_character_spec(s):
s += "\nJob:" + gen("job", stop="\n") + "\n"
@function
def gen_character_no_spec(s):
s += "Construct a character within the following format:\n"
s += "Name: Steve Jobs.\nBirthday: February 24, 1955.\nJob: Apple CEO.\n"
s += "\nPlease generate new Name, Birthday and Job.\n"
s += "Name:" + gen("name", stop="\n") + "\nBirthday:" + gen("birthday", stop="\n")
s += "\nJob:" + gen("job", stop="\n") + "\n"
@function(api_num_spec_tokens=64)
def gen_character_spec_no_few_shot(s):
# s += "Construct a character with name, birthday, and job:\n"
@@ -22,17 +31,19 @@ def gen_character_spec_no_few_shot(s):
s += "\nJob:" + gen("job", stop="\n") + "\n"
set_default_backend(OpenAI("gpt-3.5-turbo-instruct"))
if __name__ == "__main__":
backend = OpenAI("gpt-3.5-turbo-instruct")
set_default_backend(backend)
state = gen_character_spec.run()
for function in [gen_character_spec, gen_character_no_spec, gen_character_spec_no_few_shot]:
backend.token_usage.reset()
print("...name:", state["name"])
print("...birthday:", state["birthday"])
print("...job:", state["job"])
print(f"function: {function.func.__name__}")
state = gen_character_spec_no_few_shot.run()
print("\n...name:", state["name"])
print("...birthday:", state["birthday"])
print("...job:", state["job"])
state = function.run()
print("...name:", state["name"])
print("...birthday:", state["birthday"])
print("...job:", state["job"])
print(backend.token_usage)
print()