ros/LLaMA-Factory
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

merge some func

Browse Source 
Former-commit-id: 3085107c44715e4b2ca96d73b20d90c172b95219
This commit is contained in:
BUAADreamer
2024-04-25 22:35:17 +08:00
parent 514ffafc12
commit 759bee48d2
3 changed files with 10 additions and 205 deletions
Download Patch File Download Diff File Expand all files Collapse all files

51
src/llmtuner/data/aligner.py
View File

@@ -36,12 +36,7 @@ def convert_alpaca(examples: Dict[str, List[Any]], dataset_attr: "DatasetAttr")
{"role": Role.ASSISTANT.value, "content": content} for content in examples[dataset_attr.response][i]
]
elif dataset_attr.response and isinstance(examples[dataset_attr.response][i], str):
response = [
{
"role": Role.ASSISTANT.value,
"content": examples[dataset_attr.response][i],
}
]
response = [{"role": Role.ASSISTANT.value, "content": examples[dataset_attr.response][i]}]
else:
response = []
@@ -54,47 +49,6 @@ def convert_alpaca(examples: Dict[str, List[Any]], dataset_attr: "DatasetAttr")
def convert_sharegpt(examples: Dict[str, List[Any]], dataset_attr: "DatasetAttr") -> Dict[str, List[Any]]:
outputs = {"prompt": [], "response": [], "system": [], "tools": []}
tag_mapping = {
dataset_attr.user_tag: Role.USER.value,
dataset_attr.assistant_tag: Role.ASSISTANT.value,
dataset_attr.observation_tag: Role.OBSERVATION.value,
dataset_attr.function_tag: Role.FUNCTION.value,
dataset_attr.system_tag: Role.SYSTEM.value,
}
odd_tags = (dataset_attr.user_tag, dataset_attr.observation_tag)
even_tags = (dataset_attr.assistant_tag, dataset_attr.function_tag)
accept_tags = (odd_tags, even_tags)
for i, messages in enumerate(examples[dataset_attr.messages]):
if dataset_attr.system_tag and messages[0][dataset_attr.role_tag] == dataset_attr.system_tag:
system = messages[0][dataset_attr.content_tag]
messages = messages[1:]
else:
system = examples[dataset_attr.system][i] if dataset_attr.system else ""
messages = messages[: len(messages) // 2 * 2] # should be multiples of 2
if len(messages) == 0:
continue
aligned_messages = []
for turn_idx, message in enumerate(messages):
if message[dataset_attr.role_tag] not in accept_tags[turn_idx % 2]:
raise ValueError("Invalid role tag in {}.".format(messages))
aligned_messages.append(
{"role": tag_mapping[message[dataset_attr.role_tag]], "content": message[dataset_attr.content_tag]}
)
outputs["prompt"].append(aligned_messages[:-1])
outputs["response"].append(aligned_messages[-1:])
outputs["system"].append(system)
outputs["tools"].append(examples[dataset_attr.tools][i] if dataset_attr.tools else "")
outputs["images"].append([])
return outputs
def convert_llava(examples: Dict[str, List[Any]], dataset_attr: "DatasetAttr") -> Dict[str, List[Any]]:
outputs = {"prompt": [], "response": [], "system": [], "tools": [], "images": []}
tag_mapping = {
dataset_attr.user_tag: Role.USER.value,
@@ -130,7 +84,6 @@ def convert_llava(examples: Dict[str, List[Any]], dataset_attr: "DatasetAttr") -
outputs["response"].append(aligned_messages[-1:])
outputs["system"].append(system)
outputs["tools"].append(examples[dataset_attr.tools][i] if dataset_attr.tools else "")
print(examples[dataset_attr.images][i])
outputs["images"].append(examples[dataset_attr.images][i] if dataset_attr.images else [])
return outputs
@@ -148,8 +101,6 @@ def align_dataset(
"""
if dataset_attr.formatting == "alpaca":
convert_func = partial(convert_alpaca, dataset_attr=dataset_attr)
elif dataset_attr.formatting == "llava":
convert_func = partial(convert_llava, dataset_attr=dataset_attr)
else:
convert_func = partial(convert_sharegpt, dataset_attr=dataset_attr)

65
src/llmtuner/data/preprocess.py
View File

@@ -50,6 +50,7 @@ def preprocess_supervised_dataset(
tokenizer: "PreTrainedTokenizer",
template: "Template",
data_args: "DataArguments",
processor: "AutoProcessor" = None,
) -> Dict[str, List[List[int]]]:
# build inputs with format `<bos> X Y <eos>` and labels with format `<ignore> ... <ignore> Y <eos>`
# for multiturn examples, we only mask the prompt part in each prompt-response pair.
@@ -88,7 +89,9 @@ def preprocess_supervised_dataset(
model_inputs["input_ids"].append(input_ids)
model_inputs["attention_mask"].append([1] * len(input_ids))
model_inputs["labels"].append(labels)
if processor is not None and "images" in examples:
pixel_values = processor.image_processor(examples["images"][0], return_tensors="pt")["pixel_values"][0]
model_inputs["pixel_values"].append(pixel_values)
return model_inputs
@@ -138,55 +141,6 @@ def preprocess_packed_supervised_dataset(
return model_inputs
def preprocess_multimodal_supervised_dataset(
examples: Dict[str, List[Any]],
processor: "AutoProcessor",
template: "Template",
data_args: "DataArguments",
) -> Dict[str, List[List[int]]]:
# build inputs with format `<bos> X Y <eos>` and labels with format `<ignore> ... <ignore> Y <eos>`
# for multiturn examples, we only mask the prompt part in each prompt-response pair.
tokenizer = processor.tokenizer
model_inputs = {"input_ids": [], "attention_mask": [], "labels": [], "pixel_values": []}
for i in range(len(examples["prompt"])):
if len(examples["prompt"][i]) % 2 != 1 or len(examples["response"][i]) != 1:
continue
messages = examples["prompt"][i] + examples["response"][i]
input_ids, labels = [], []
for turn_idx, (source_ids, target_ids) in enumerate(
template.encode_multiturn(
tokenizer,
messages,
examples["system"][i],
examples["tools"][i],
data_args.cutoff_len,
data_args.reserved_label_len,
)
):
if data_args.train_on_prompt:
source_mask = source_ids
elif turn_idx != 0 and template.efficient_eos:
source_mask = [tokenizer.eos_token_id] + [IGNORE_INDEX] * (len(source_ids) - 1)
else:
source_mask = [IGNORE_INDEX] * len(source_ids)
input_ids += source_ids + target_ids
labels += source_mask + target_ids
if template.efficient_eos:
input_ids += [tokenizer.eos_token_id]
labels += [tokenizer.eos_token_id]
model_inputs["input_ids"].append(input_ids)
model_inputs["attention_mask"].append([1] * len(input_ids))
model_inputs["labels"].append(labels)
pixel_values = processor.image_processor(examples["images"][0], return_tensors="pt")["pixel_values"][0]
model_inputs["pixel_values"].append(pixel_values)
return model_inputs
def preprocess_unsupervised_dataset(
examples: Dict[str, List[Any]],
tokenizer: "PreTrainedTokenizer",
@@ -307,15 +261,14 @@ def get_preprocess_and_print_func(
preprocess_func = partial(
preprocess_packed_supervised_dataset, tokenizer=tokenizer, template=template, data_args=data_args
)
elif processor is not None:
preprocess_func = partial(
preprocess_multimodal_supervised_dataset, processor=processor, template=template, data_args=data_args
)
else:
preprocess_func = partial(
preprocess_supervised_dataset, tokenizer=tokenizer, template=template, data_args=data_args
preprocess_supervised_dataset,
tokenizer=tokenizer,
template=template,
data_args=data_args,
processor=processor,
)
print_function = partial(print_supervised_dataset_example, tokenizer=tokenizer)
elif stage == "rm":
preprocess_func = partial(