github
/
InternLM
mirror of https://github.com/InternLM/InternLM
1
0
Fork 0
Code Issues Projects Releases Wiki Activity

update

pull/751/head
zhulin1 2024-07-01 11:57:44 +08:00
parent f36805270c
commit a99d681d63
1 changed files with 38 additions and 26 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

52
tests/test_hf_model.py
View File

@ -160,6 +160,7 @@ class TestMath:
assert_model(response) assert_model(response)
assert '2' in response assert '2' in response
class TestReward: class TestReward:
"""Test cases for base model.""" """Test cases for base model."""
@ -181,46 +182,57 @@ class TestReward:
tokenizer = AutoTokenizer.from_pretrained(model_name, tokenizer = AutoTokenizer.from_pretrained(model_name,
trust_remote_code=True, trust_remote_code=True,
use_fast=usefast) use_fast=usefast)
model = AutoModel.from_pretrained(model_name, device_map="cuda", model = AutoModel.from_pretrained(
torch_dtype=torch.float16, model_name,
trust_remote_code=True,) device_map='cuda',
torch_dtype=torch.float16,
trust_remote_code=True,
)
tokenizer = AutoTokenizer.from_pretrained(model_name, tokenizer = AutoTokenizer.from_pretrained(model_name,
trust_remote_code=True) trust_remote_code=True)
chat_1 = [ chat_1 = [{
{"role": "user", "content": "Hello! What's your name?"}, 'role': 'user',
{"role": "assistant", "content": "My name is InternLM2! A helpful AI assistant. What can I do for you?"} 'content': "Hello! What's your name?"
] }, {
chat_2 = [ 'role':
{"role": "user", "content": "Hello! What's your name?"}, 'assistant',
{"role": "assistant", "content": "I have no idea."} 'content':
] 'I am InternLM2! A helpful AI assistant. What can I do for you?'
}]
chat_2 = [{
'role': 'user',
'content': "Hello! What's your name?"
}, {
'role': 'assistant',
'content': 'I have no idea.'
}]
# get reward score for a single chat # get reward score for a single chat
score1 = model.get_score(tokenizer, chat_1) score1 = model.get_score(tokenizer, chat_1)
score2 = model.get_score(tokenizer, chat_2) score2 = model.get_score(tokenizer, chat_2)
print("score1: ", score1) print('score1: ', score1)
print("score2: ", score2) print('score2: ', score2)
assert score1 > 0.5 && score1 < 1 && score2 < 0 assert score1 > 0.5 & score1 < 1 & score2 < 0
# batch inference, get multiple scores at once # batch inference, get multiple scores at once
scores = model.get_scores(tokenizer, [chat_1, chat_2]) scores = model.get_scores(tokenizer, [chat_1, chat_2])
print("scores: ", scores) print('scores: ', scores)
assert scores[0] > 0.5 && scores[0] < 1 && scores[1] < 0 assert scores[0] > 0.5 & scores[0] < 1 & scores[1] < 0
# compare whether chat_1 is better than chat_2 # compare whether chat_1 is better than chat_2
compare_res = model.compare(tokenizer, chat_1, chat_2) compare_res = model.compare(tokenizer, chat_1, chat_2)
print("compare_res: ", compare_res) print('compare_res: ', compare_res)
assert compare_res assert compare_res
# >>> compare_res: True # >>> compare_res: True
# rank multiple chats, it will return the ranking index of each chat # rank multiple chats, it will return the ranking index of each chat
# the chat with the highest score will have ranking index as 0 # the chat with the highest score will have ranking index as 0
rank_res = model.rank(tokenizer, [chat_1, chat_2]) rank_res = model.rank(tokenizer, [chat_1, chat_2])
print("rank_res: ", rank_res) # lower index means higher score print('rank_res: ', rank_res) # lower index means higher score
# >>> rank_res: [0, 1] # >>> rank_res: [0, 1]
assert rank_res[0] == 0 && rank_res[1] == 1 assert rank_res[0] == 0 & rank_res[1] == 1
)
class TestMMModel: class TestMMModel:
"""Test cases for base model.""" """Test cases for base model."""