tocmo0nlord/axolotl
1
0
Fork 0
Code Issues Pull Requests Actions 2 Packages Projects Releases Wiki Activity

Fix mypy typing

Browse Source
This commit is contained in:
NanoCode012
2023-05-29 18:13:39 +09:00
parent f1232b35ba
commit e9650d3ae4
8 changed files with 190 additions and 33 deletions
Download Patch File Download Diff File Expand all files Collapse all files

6
src/axolotl/prompt_strategies/pygmalion.py
View File

@@ -3,7 +3,7 @@
import copy
import logging
from collections import defaultdict
from typing import Generator
from typing import Generator, List, Tuple
from axolotl.prompt_tokenizers import (
PromptTokenizingStrategy,
@@ -19,7 +19,7 @@ class PygmalionPromptTokenizingStrategy(PromptTokenizingStrategy):
Tokenizing strategy for Pygmalion.
"""
bot_prefix_token_ids = []
bot_prefix_token_ids: List[int] = []
def __init__(self, prompter, tokenizer, *args, **kwargs):
super().__init__(prompter, tokenizer, *args, **kwargs)
@@ -88,7 +88,7 @@ class PygmalionPrompter:
def build_prompt(
self, source, *args, **kwargs # pylint: disable=unused-argument
) -> Generator[str, None, None]:
) -> Generator[Tuple[str, str], None, None]:
for msg in source:
yield msg["role"], msg["value"]

12
src/axolotl/prompt_tokenizers.py
View File

@@ -226,20 +226,16 @@ class CompletionPromptTokenizingStrategy(InstructionPromptTokenizingStrategy):
Tokenizing strategy for Completion prompts.
"""
def parse_instruction_fields(self, prompt) -> str:
return prompt["text"]
def tokenize_prompt(self, prompt):
instruction = self.parse_instruction_fields(prompt)
full_prompt = self._build_full_prompt(instruction, None, None)
full_prompt = self._build_full_prompt(prompt["text"], None, None)
tokenized_full_prompt = self._tokenize(full_prompt)
return tokenized_full_prompt
def _build_full_prompt(
self, instruction, input, response
): # pylint: disable=unused-argument, redefined-builtin
return next(iter(self.prompter.build_prompt(instruction)))
): # pylint: disable=redefined-builtin
return next(iter(self.prompter.build_prompt(instruction, input, response)))
class ReflectionPromptTokenizingStrategy(PromptTokenizingStrategy):
@@ -419,7 +415,7 @@ def tokenize_prompt_default() -> Tuple[Dict[str, List[int]], int]:
Returns the default values for the tokenize prompt function
"""
result = {
result: Dict[str, List[int]] = {
"input_ids": [],
"attention_mask": [],
"labels": [],

14
src/axolotl/prompters.py
View File

@@ -3,7 +3,7 @@
import dataclasses
import logging
from enum import auto, Enum
from typing import List, Union, Generator
from typing import List, Optional, Union, Generator
IGNORE_TOKEN_ID = -100
@@ -24,7 +24,7 @@ class AlpacaPrompter:
system_prompt = "Below is an instruction that describes a task, paired with an input that provides further context. Write a response that appropriately completes the request.\n\n"
system_no_input_prompt = "Below is an instruction that describes a task. Write a response that appropriately completes the request.\n\n"
prompt_style = None
prompt_style: Optional[PromptStyle] = None
def __init__(self, prompt_style=PromptStyle.INSTRUCT.value):
self.prompt_style = prompt_style if prompt_style else PromptStyle.INSTRUCT.value
@@ -231,18 +231,18 @@ class Conversation:
offset: int
sep_style: SeparatorStyle = SeparatorStyle.SINGLE
sep: str = "###"
sep2: str = None
sep2: Optional[str] = None
def get_prompt(self) -> Generator[str, None, None]:
seps = [self.sep, self.sep2]
preamble = self.system + seps[0]
# seps = [self.sep, self.sep2]
preamble = self.system + self.sep
yield preamble
for _, (role, message) in enumerate(self.messages):
if message:
yield (role + ":", " " + message)
yield role + ":" + " " + message
else:
logging.warning(f"role with empty message: {role}")
yield (role + ":",)
yield role + ":"
def copy(self):
return Conversation(

20
src/axolotl/utils/data.py
View File

@@ -3,7 +3,7 @@
import logging
from hashlib import md5
from pathlib import Path
from typing import Tuple, Union
from typing import List, Tuple, Union
from datasets import (
load_from_disk,
@@ -95,40 +95,36 @@ def load_tokenized_prepared_datasets(
# prefer local dataset, even if hub exists
if Path(d.path).exists():
ds: Dataset = load_dataset(
ds = load_dataset(
"json", data_files=d.path, streaming=False, split=None
)
elif ds_from_hub:
if d.data_files:
ds: Dataset = load_dataset(
ds = load_dataset(
d.path,
streaming=False,
data_files=d.data_files,
use_auth_token=use_auth_token,
)
else:
ds: Dataset = load_dataset(
ds = load_dataset(
d.path, streaming=False, use_auth_token=use_auth_token
)
else:
fp = hf_hub_download(
repo_id=d.path, repo_type="dataset", filename=d.data_files
)
ds: Dataset = load_dataset(
"json", data_files=fp, streaming=False, split=None
)
ds = load_dataset("json", data_files=fp, streaming=False, split=None)
if not ds:
raise ValueError("unhandled dataset load")
# support for using a subset of the data
if d.shards:
if "train" in ds:
ds: DatasetDict = ds.shuffle(seed=42)["train"].shard(
ds = ds.shuffle(seed=42)["train"].shard(
num_shards=d.shards, index=0
)
else:
ds: Dataset = ds.shuffle(seed=42).shard(
num_shards=d.shards, index=0
)
ds = ds.shuffle(seed=42).shard(num_shards=d.shards, index=0)
d_type = d.type
d_type_split = d_type.split(":")
d_base_type = d_type_split[0]
@@ -232,7 +228,7 @@ def load_tokenized_prepared_datasets(
logging.error(f"unhandled prompt tokenization strategy: {d.type}")
logging.info("tokenizing, merging, and shuffling master dataset")
samples = []
samples: List[int] = []
for d in datasets:
samples = samples + list(d)
dataset = Dataset.from_list(samples).shuffle(seed=42)

2
src/axolotl/utils/models.py
View File

@@ -81,7 +81,7 @@ def load_model(
adapter="lora",
inference=False,
):
# type: (str, str, str, str, DictDefault, Optional[str], bool) -> Tuple[PreTrainedModel, PreTrainedTokenizer, Optional[PeftConfig]]
# type: (str, str, str, str, DictDefault, Optional[str], bool) -> Tuple[PreTrainedModel, Optional[PeftConfig]]
"""
Load a model from a base model and a model type.
"""

3
src/axolotl/utils/trainer.py
View File

@@ -5,6 +5,7 @@ import math
import os
import sys
from pathlib import Path
from typing import Optional
import bitsandbytes as bnb
import torch.cuda
@@ -28,7 +29,7 @@ class OneCycleLRSchedulerTrainer(Trainer):
self.lr_scheduler = None
def create_scheduler(
self, num_training_steps: int, optimizer: torch.optim.Optimizer = None
self, num_training_steps: int, optimizer: Optional[torch.optim.Optimizer] = None
):
optimizer = self.optimizer if optimizer is None else optimizer
num_warmup_steps = self.args.get_warmup_steps(num_training_steps)