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

Fix logit processor bugs (#427)

Browse Source
This commit is contained in:
Lianmin Zheng
2024-05-12 04:54:07 -07:00
committed by GitHub
parent 7023f413c6
commit aee4f523cf
26 changed files with 166 additions and 257 deletions
Download Patch File Download Diff File Expand all files Collapse all files

0
test/killall_python.sh → test/killall_sglang.sh
View File

BIN
test/lang/example_image.png Normal file
View File

Binary file not shown.
Side by Side

After

Width:  |  Height:  |  Size: 56 KiB

13
test/lang/test_openai_backend.py
View File

@@ -28,7 +28,7 @@ class TestOpenAIBackend(unittest.TestCase):
if cls.backend is None:
cls.backend = OpenAI("gpt-3.5-turbo-instruct")
cls.chat_backend = OpenAI("gpt-3.5-turbo")
cls.chat_vision_backend = OpenAI("gpt-4-vision-preview")
cls.chat_vision_backend = OpenAI("gpt-4-turbo")
def test_few_shot_qa(self):
set_default_backend(self.backend)
@@ -88,14 +88,3 @@ if __name__ == "__main__":
# t = TestOpenAIBackend()
# t.setUp()
# t.test_few_shot_qa()
# t.test_mt_bench()
# t.test_select()
# t.test_decode_int()
# t.test_decode_json()
# t.test_expert_answer()
# t.test_tool_use()
# t.test_react()
# t.test_parallel_decoding()
# t.test_parallel_encoding()
# t.test_image_qa()
# t.test_stream()

68
test/lang/test_openai_spec.py
View File

@@ -1,68 +0,0 @@
from sglang import OpenAI, function, gen, set_default_backend
@function()
def gen_character_default(s):
s += "Construct a character within the following format:\n"
s += "Name: Steve Jobs.\nBirthday: February 24, 1955.\nJob: Apple CEO.\nWelcome.\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") + "\nWelcome.\n"
@function(api_num_spec_tokens=512)
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.\nWelcome.\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") + "\nWelcome.\n"
@function(api_num_spec_tokens=512)
def gen_character_no_stop(s):
s += "Construct a character within the following format:\n"
s += "Name: Steve Jobs.\nBirthday: February 24, 1955.\nJob: Apple CEO.\nWelcome.\n"
s += "\nPlease generate new Name, Birthday and Job.\n"
s += "Name:" + gen("name") + "\nBirthday:" + gen("birthday")
s += "\nJob:" + gen("job") + "\nWelcome.\n"
@function(api_num_spec_tokens=512)
def gen_character_multi_stop(s):
s += "Construct a character within the following format:\n"
s += (
"Name: Steve Jobs.###Birthday: February 24, 1955.###Job: Apple CEO.\nWelcome.\n"
)
s += "\nPlease generate new Name, Birthday and Job.\n"
s += "Name:" + gen("name", stop=["\n", "###"])
s += "###Birthday:" + gen("birthday", stop=["\n", "###"])
s += "###Job:" + gen("job", stop=["\n", "###"]) + "\nWelcome.\n"
set_default_backend(OpenAI("gpt-3.5-turbo-instruct"))
state = gen_character_default.run()
print(state.text())
print("=" * 60)
state = gen_character_no_stop.run()
print("name###", state["name"])
print("birthday###:", state["birthday"])
print("job###", state["job"])
print("=" * 60)
state = gen_character_multi_stop.run()
print(state.text())
print("=" * 60)
state = gen_character_spec.run()
print(state.text())
print("name###", state["name"])
print("birthday###", state["birthday"])
print("job###", state["job"])

1
test/srt/example_image.png Symbolic link
View File

@@ -0,0 +1 @@
../lang/example_image.png

2
test/srt/test_httpserver_decode.py
View File

@@ -3,7 +3,6 @@ Usage:
python3 -m sglang.launch_server --model-path TinyLlama/TinyLlama-1.1B-Chat-v0.4 --port 30000
python3 test_httpserver_decode.py
Output:
The capital of France is Paris.\nThe capital of the United States is Washington, D.C.\nThe capital of Canada is Ottawa.\nThe capital of Japan is Tokyo
"""
@@ -23,6 +22,7 @@ def test_decode(url, return_logprob, top_logprobs_num, return_text):
"temperature": 0,
"max_new_tokens": 32,
},
"stream": False,
"return_logprob": return_logprob,
"top_logprobs_num": top_logprobs_num,
"return_text_in_logprobs": return_text,

1
test/srt/test_httpserver_decode_stream.py
View File

@@ -26,6 +26,7 @@ def test_decode_stream(url, return_logprob, top_logprobs_num):
"return_logprob": return_logprob,
"top_logprobs_num": top_logprobs_num,
"return_text_in_logprobs": True,
"logprob_start_len": 0,
},
stream=True,
)

4
test/srt/test_httpserver_llava.py
View File

@@ -34,7 +34,7 @@ async def test_concurrent(args):
url + "/generate",
{
"text": "A chat between a curious human and an artificial intelligence assistant. The assistant gives helpful, detailed, and polite answers to the human's questions. USER: <image>\nDescribe this picture ASSISTANT:",
"image_data": "test_image.png",
"image_data": "example_image.png",
"sampling_params": {
"temperature": 0,
"max_new_tokens": 16,
@@ -55,7 +55,7 @@ def test_streaming(args):
url + "/generate",
json={
"text": "A chat between a curious human and an artificial intelligence assistant. The assistant gives helpful, detailed, and polite answers to the human's questions. USER: <image>\nDescribe this picture ASSISTANT:",
"image_data": "test_image.png",
"image_data": "example_image.png",
"sampling_params": {
"temperature": 0,
"max_new_tokens": 128,

2
test/srt/test_httpserver_reuse.py
View File

@@ -6,10 +6,10 @@ The capital of France is Paris.\nThe capital of the United States is Washington,
"""
import argparse
import time
import requests
if __name__ == "__main__":
parser = argparse.ArgumentParser()
parser.add_argument("--host", type=str, default="http://127.0.0.1")

2
test/srt/test_openai_server.py
View File

@@ -163,7 +163,7 @@ def test_regex(args):
regex = (
r"""\{\n"""
+ r""" "name": "[\w]+",\n"""
+ r""" "population": "[\w\d\s]+"\n"""
+ r""" "population": [\w\d\s]+\n"""
+ r"""\}"""
)