423A35C7/LLaMA-Factory
1
0
Fork 0
mirror of https://github.com/hiyouga/LLaMA-Factory.git synced 2025-12-14 10:56:56 +08:00
Code Issues Packages Projects Releases Wiki Activity

fix #4402 #4617

Browse Source 
Deprecate reserved_label_len arg


Former-commit-id: 4b6568984c0be4b31e7aa91b7c0d52b7f7b12b0b
This commit is contained in:
hiyouga
2024-07-01 01:19:27 +08:00
parent b670fb57db
commit 67d2eb6b2a
13 changed files with 329 additions and 223 deletions
Download Patch File Download Diff File Expand all files Collapse all files

55
tests/data/test_template.py
View File

@@ -21,15 +21,60 @@ from llamafactory.data import get_template_and_fix_tokenizer
TINY_LLAMA = os.environ.get("TINY_LLAMA", "llamafactory/tiny-random-Llama-3")
MESSAGES = [
{"role": "user", "content": "How are you"},
{"role": "assistant", "content": "I am fine!"},
{"role": "user", "content": "你好"},
{"role": "assistant", "content": "很高兴认识你！"},
]
def test_encode_oneturn():
tokenizer = AutoTokenizer.from_pretrained(TINY_LLAMA)
template = get_template_and_fix_tokenizer(tokenizer, name="llama3")
prompt_ids, answer_ids = template.encode_oneturn(tokenizer, MESSAGES)
assert tokenizer.decode(prompt_ids) == (
"<|begin_of_text|><|start_header_id|>user<|end_header_id|>\n\nHow are you<|eot_id|>"
"<|start_header_id|>assistant<|end_header_id|>\n\nI am fine!<|eot_id|>"
"<|start_header_id|>user<|end_header_id|>\n\n你好<|eot_id|>"
"<|start_header_id|>assistant<|end_header_id|>\n\n"
)
assert tokenizer.decode(answer_ids) == "很高兴认识你！<|eot_id|>"
def test_encode_multiturn():
tokenizer = AutoTokenizer.from_pretrained(TINY_LLAMA)
template = get_template_and_fix_tokenizer(tokenizer, name="llama3")
encoded_pairs = template.encode_multiturn(tokenizer, MESSAGES)
assert tokenizer.decode(encoded_pairs[0][0]) == (
"<|begin_of_text|><|start_header_id|>user<|end_header_id|>\n\nHow are you<|eot_id|>"
"<|start_header_id|>assistant<|end_header_id|>\n\n"
)
assert tokenizer.decode(encoded_pairs[0][1]) == "I am fine!<|eot_id|>"
assert tokenizer.decode(encoded_pairs[1][0]) == (
"<|start_header_id|>user<|end_header_id|>\n\n你好<|eot_id|>"
"<|start_header_id|>assistant<|end_header_id|>\n\n"
)
assert tokenizer.decode(encoded_pairs[1][1]) == "很高兴认识你！<|eot_id|>"
def test_jinja_template():
tokenizer = AutoTokenizer.from_pretrained(TINY_LLAMA)
ref_tokenizer = AutoTokenizer.from_pretrained(TINY_LLAMA)
get_template_and_fix_tokenizer(tokenizer, name="llama3")
assert tokenizer.chat_template != ref_tokenizer.chat_template
assert tokenizer.apply_chat_template(MESSAGES) == ref_tokenizer.apply_chat_template(MESSAGES)
messages = [
{"role": "user", "content": "hi!"},
{"role": "assistant", "content": "hello there"},
]
assert tokenizer.apply_chat_template(messages) == ref_tokenizer.apply_chat_template(messages)
def test_qwen_template():
tokenizer = AutoTokenizer.from_pretrained("Qwen/Qwen2-7B-Instruct")
template = get_template_and_fix_tokenizer(tokenizer, name="qwen")
prompt_ids, answer_ids = template.encode_oneturn(tokenizer, MESSAGES)
assert tokenizer.decode(prompt_ids) == (
"<|im_start|>system\nYou are a helpful assistant.<|im_end|>\n"
"<|im_start|>user\nHow are you<|im_end|>\n"
"<|im_start|>assistant\nI am fine!<|im_end|>\n"
"<|im_start|>user\n你好<|im_end|>\n"
"<|im_start|>assistant\n"
)
assert tokenizer.decode(answer_ids) == "很高兴认识你！<|im_end|>"