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

[misc] upgrade format to py39 (#7256)

Browse Source
This commit is contained in:
hoshi-hiyouga
2025-03-12 00:08:41 +08:00
committed by GitHub
parent 5995800bce
commit 264538cb26
113 changed files with 984 additions and 1407 deletions
Download Patch File Download Diff File Expand all files Collapse all files

2
scripts/api_example/test_toolcall.py
View File

@@ -14,7 +14,7 @@
import json
import os
from typing import Sequence
from collections.abc import Sequence
from openai import OpenAI
from transformers.utils.versions import require_version

12
scripts/convert_ckpt/llamafy_baichuan2.py
View File

@@ -15,7 +15,7 @@
import json
import os
from collections import OrderedDict
from typing import Any, Dict
from typing import Any
import fire
import torch
@@ -29,13 +29,13 @@ CONFIG_NAME = "config.json"
def save_weight(input_dir: str, output_dir: str, shard_size: str, save_safetensors: bool):
baichuan2_state_dict: Dict[str, torch.Tensor] = OrderedDict()
baichuan2_state_dict: dict[str, torch.Tensor] = OrderedDict()
for filepath in tqdm(os.listdir(input_dir), desc="Load weights"):
if os.path.isfile(os.path.join(input_dir, filepath)) and filepath.endswith(".bin"):
shard_weight = torch.load(os.path.join(input_dir, filepath), map_location="cpu")
baichuan2_state_dict.update(shard_weight)
llama_state_dict: Dict[str, torch.Tensor] = OrderedDict()
llama_state_dict: dict[str, torch.Tensor] = OrderedDict()
for key, value in tqdm(baichuan2_state_dict.items(), desc="Convert format"):
if "W_pack" in key:
proj_size = value.size(0) // 3
@@ -75,7 +75,7 @@ def save_weight(input_dir: str, output_dir: str, shard_size: str, save_safetenso
def save_config(input_dir: str, output_dir: str):
with open(os.path.join(input_dir, CONFIG_NAME), encoding="utf-8") as f:
llama2_config_dict: Dict[str, Any] = json.load(f)
llama2_config_dict: dict[str, Any] = json.load(f)
llama2_config_dict["architectures"] = ["LlamaForCausalLM"]
llama2_config_dict.pop("auto_map", None)
@@ -94,8 +94,8 @@ def llamafy_baichuan2(
shard_size: str = "2GB",
save_safetensors: bool = True,
):
r"""
Converts the Baichuan2-7B model in the same format as LLaMA2-7B.
r"""Convert the Baichuan2-7B model in the same format as LLaMA2-7B.
Usage: python llamafy_baichuan2.py --input_dir input --output_dir output
Converted model: https://huggingface.co/hiyouga/Baichuan2-7B-Base-LLaMAfied
"""

14
scripts/convert_ckpt/llamafy_qwen.py
View File

@@ -15,7 +15,7 @@
import json
import os
from collections import OrderedDict
from typing import Any, Dict
from typing import Any
import fire
import torch
@@ -37,14 +37,14 @@ CONFIG_NAME = "config.json"
def save_weight(input_dir: str, output_dir: str, shard_size: str, save_safetensors: bool) -> str:
qwen_state_dict: Dict[str, torch.Tensor] = OrderedDict()
qwen_state_dict: dict[str, torch.Tensor] = OrderedDict()
for filepath in tqdm(os.listdir(input_dir), desc="Load weights"):
if os.path.isfile(os.path.join(input_dir, filepath)) and filepath.endswith(".safetensors"):
with safe_open(os.path.join(input_dir, filepath), framework="pt", device="cpu") as f:
for key in f.keys():
qwen_state_dict[key] = f.get_tensor(key)
llama_state_dict: Dict[str, torch.Tensor] = OrderedDict()
llama_state_dict: dict[str, torch.Tensor] = OrderedDict()
torch_dtype = None
for key, value in tqdm(qwen_state_dict.items(), desc="Convert format"):
if torch_dtype is None:
@@ -112,9 +112,9 @@ def save_weight(input_dir: str, output_dir: str, shard_size: str, save_safetenso
def save_config(input_dir: str, output_dir: str, torch_dtype: str):
with open(os.path.join(input_dir, CONFIG_NAME), encoding="utf-8") as f:
qwen_config_dict: Dict[str, Any] = json.load(f)
qwen_config_dict: dict[str, Any] = json.load(f)
llama2_config_dict: Dict[str, Any] = OrderedDict()
llama2_config_dict: dict[str, Any] = OrderedDict()
llama2_config_dict["architectures"] = ["LlamaForCausalLM"]
llama2_config_dict["hidden_act"] = "silu"
llama2_config_dict["hidden_size"] = qwen_config_dict["hidden_size"]
@@ -147,8 +147,8 @@ def llamafy_qwen(
shard_size: str = "2GB",
save_safetensors: bool = False,
):
r"""
Converts the Qwen models in the same format as LLaMA2.
r"""Convert the Qwen models in the same format as LLaMA2.
Usage: python llamafy_qwen.py --input_dir input --output_dir output
Converted model: https://huggingface.co/hiyouga/Qwen-14B-Chat-LLaMAfied
"""

10
scripts/llama_pro.py
View File

@@ -18,7 +18,7 @@
import json
import os
from collections import OrderedDict
from typing import TYPE_CHECKING, Dict
from typing import TYPE_CHECKING
import fire
import torch
@@ -44,11 +44,11 @@ def block_expansion(
shard_size: str = "5GB",
save_safetensors: bool = True,
):
r"""
Performs block expansion for LLaMA, Mistral, Qwen2 or Yi models.
r"""Perform block expansion for LLaMA, Mistral, Qwen2 or Yi models.
Usage: python llama_pro.py --model_name_or_path meta-llama/Llama-2-7b-hf --output_dir llama2_pro --num_expand 8
"""
config: "PretrainedConfig" = AutoConfig.from_pretrained(model_name_or_path, trust_remote_code=True)
config: PretrainedConfig = AutoConfig.from_pretrained(model_name_or_path, trust_remote_code=True)
num_layers = getattr(config, "num_hidden_layers")
if num_layers % num_expand != 0:
raise ValueError(f"`num_layers` {num_layers} should be divisible by `num_expand` {num_expand}.")
@@ -70,7 +70,7 @@ def block_expansion(
split = num_layers // num_expand
layer_cnt = 0
state_dict = model.state_dict()
output_state_dict: Dict[str, "torch.Tensor"] = OrderedDict()
output_state_dict: dict[str, torch.Tensor] = OrderedDict()
for i in range(num_layers):
for key, value in state_dict.items():
if f".{i:d}." in key:

6
scripts/loftq_init.py
View File

@@ -38,8 +38,8 @@ def quantize_loftq(
lora_target: tuple = ("q_proj", "v_proj"),
save_safetensors: bool = True,
):
r"""
Initializes LoRA weights with LoRA-fine-tuning-aware Quantization (LoftQ)
r"""Initialize LoRA weights with LoRA-fine-tuning-aware Quantization (LoftQ).
Usage: python loftq_init.py --model_name_or_path path_to_model --output_dir output_dir
"""
if isinstance(lora_target, str):
@@ -72,7 +72,7 @@ def quantize_loftq(
print(f"Adapter weights saved in {loftq_dir}")
# Save base model
base_model: "PreTrainedModel" = peft_model.unload()
base_model: PreTrainedModel = peft_model.unload()
base_model.save_pretrained(output_dir, safe_serialization=save_safetensors)
tokenizer.save_pretrained(output_dir)
print(f"Model weights saved in {output_dir}")

6
scripts/pissa_init.py
View File

@@ -37,8 +37,8 @@ def quantize_pissa(
lora_target: tuple = ("q_proj", "v_proj"),
save_safetensors: bool = True,
):
r"""
Initializes LoRA weights with Principal Singular values and Singular vectors Adaptation (PiSSA)
r"""Initialize LoRA weights with Principal Singular values and Singular vectors Adaptation (PiSSA).
Usage: python pissa_init.py --model_name_or_path path_to_model --output_dir output_dir
"""
if isinstance(lora_target, str):
@@ -67,7 +67,7 @@ def quantize_pissa(
print(f"Adapter weights saved in {pissa_dir}")
# Save base model
base_model: "PreTrainedModel" = peft_model.unload()
base_model: PreTrainedModel = peft_model.unload()
base_model.save_pretrained(output_dir, safe_serialization=save_safetensors)
tokenizer.save_pretrained(output_dir)
print(f"Model weights saved in {output_dir}")

4
scripts/stat_utils/cal_flops.py
View File

@@ -29,8 +29,8 @@ def calculate_flops(
seq_length: int = 512,
flash_attn: str = "auto",
):
r"""
Calculates the flops of pre-trained models.
r"""Calculate the flops of pre-trained models.
Usage: python cal_flops.py --model_name_or_path path_to_model --batch_size 1 --seq_length 512
"""
with get_accelerator().device(0):

9
scripts/stat_utils/cal_lr.py
View File

@@ -45,8 +45,8 @@ def calculate_lr(
is_mistral_or_gemma: bool = False, # mistral and gemma models opt for a smaller learning rate,
packing: bool = False,
):
r"""
Calculates the optimal learning rate for 7B/13B models using LLaMA's hyper-parameters.
r"""Calculate the optimal learning rate for 7B/13B models using LLaMA's hyper-parameters.
Usage:
python cal_lr.py --model_name_or_path path_to_model --dataset alpaca_en_demo --cutoff_len 1024 --batch_size 16
"""
@@ -89,9 +89,8 @@ def calculate_lr(
lr = BASE_LR * math.sqrt(token_batch_size / BASE_BS) # lr ~ sqrt(batch_size)
lr = lr / 6.0 if is_mistral_or_gemma else lr
print(
"Optimal learning rate is {:.2e} for valid ratio% {:.2f} and effective token batch size {:.2f}".format(
lr, valid_ratio * 100, token_batch_size
)
f"Optimal learning rate is {lr:.2e} for valid ratio% {valid_ratio * 100:.2f} "
f"and effective token batch size {token_batch_size:.2f}"
)

12
scripts/stat_utils/cal_mfu.py
View File

@@ -34,9 +34,7 @@ def compute_model_flops(
include_recompute: bool = False,
include_flashattn: bool = False,
) -> int:
r"""
Calculates the FLOPs of model per forward/backward pass.
"""
r"""Calculate the FLOPs of model per forward/backward pass."""
config = AutoConfig.from_pretrained(model_name_or_path)
hidden_size = getattr(config, "hidden_size", None)
vocab_size = getattr(config, "vocab_size", None)
@@ -86,9 +84,7 @@ def compute_model_flops(
def compute_device_flops(world_size: int) -> float:
r"""
Calculates the FLOPs of the device capability per second.
"""
r"""Calculate the FLOPs of the device capability per second."""
device_name = torch.cuda.get_device_name()
if "H100" in device_name or "H800" in device_name:
return 989 * 1e12 * world_size
@@ -114,8 +110,8 @@ def calculate_mfu(
liger_kernel: bool = False,
unsloth_gc: bool = False,
) -> float:
r"""
Calculates MFU for given model and hyper-params.
r"""Calculate MFU for given model and hyper-params.
Usage: python cal_mfu.py --model_name_or_path path_to_model --batch_size 1 --seq_length 1024
"""
args = {

25
scripts/stat_utils/cal_ppl.py
View File

@@ -13,8 +13,9 @@
# limitations under the License.
import json
from collections.abc import Sequence
from dataclasses import dataclass
from typing import Any, Dict, Literal, Optional, Sequence
from typing import Any, Literal, Optional
import fire
import torch
@@ -30,16 +31,12 @@ from llamafactory.model import load_model, load_tokenizer
@dataclass
class PairwiseDataCollatorWithPadding(MultiModalDataCollatorForSeq2Seq):
r"""
Data collator for pairwise data.
"""
r"""Data collator for pairwise data."""
train_on_prompt: bool = False
def __call__(self, features: Sequence[Dict[str, Any]]) -> Dict[str, torch.Tensor]:
r"""
Pads batched data to the longest sequence in the batch.
"""
def __call__(self, features: Sequence[dict[str, Any]]) -> dict[str, torch.Tensor]:
r"""Pad batched data to the longest sequence in the batch."""
chosen_features = []
for feature in features:
chosen_features.append(
@@ -68,8 +65,8 @@ def calculate_ppl(
max_samples: Optional[int] = None,
train_on_prompt: bool = False,
):
r"""
Calculates the ppl on the dataset of the pre-trained models.
r"""Calculate the ppl on the dataset of the pre-trained models.
Usage: export CUDA_VISIBLE_DEVICES=0
python cal_ppl.py --model_name_or_path path_to_model --dataset alpaca_en_demo --save_name ppl.json
"""
@@ -111,17 +108,17 @@ def calculate_ppl(
criterion = torch.nn.CrossEntropyLoss(reduction="none")
total_ppl = 0
perplexities = []
batch: Dict[str, "torch.Tensor"]
batch: dict[str, torch.Tensor]
with torch.no_grad():
for batch in tqdm(dataloader, desc="Computing perplexities"):
batch = batch.to(model.device)
outputs = model(**batch)
shift_logits: "torch.Tensor" = outputs["logits"][..., :-1, :]
shift_labels: "torch.Tensor" = batch["labels"][..., 1:]
shift_logits: torch.Tensor = outputs["logits"][..., :-1, :]
shift_labels: torch.Tensor = batch["labels"][..., 1:]
loss_mask = shift_labels != IGNORE_INDEX
flatten_logits = shift_logits.contiguous().view(shift_labels.size(0) * shift_labels.size(1), -1)
flatten_labels = shift_labels.contiguous().view(-1)
token_logps: "torch.Tensor" = criterion(flatten_logits, flatten_labels)
token_logps: torch.Tensor = criterion(flatten_logits, flatten_labels)
token_logps = token_logps.contiguous().view(shift_logits.size(0), -1)
sentence_logps = (token_logps * loss_mask).sum(-1) / loss_mask.sum(-1)
total_ppl += sentence_logps.exp().sum().item()

4
scripts/stat_utils/length_cdf.py
View File

@@ -29,8 +29,8 @@ def length_cdf(
template: str = "default",
interval: int = 1000,
):
r"""
Calculates the distribution of the input lengths in the dataset.
r"""Calculate the distribution of the input lengths in the dataset.
Usage: export CUDA_VISIBLE_DEVICES=0
python length_cdf.py --model_name_or_path path_to_model --dataset alpaca_en_demo --template default
"""

4
scripts/vllm_infer.py
View File

@@ -52,8 +52,8 @@ def vllm_infer(
image_max_pixels: int = 768 * 768,
image_min_pixels: int = 32 * 32,
):
r"""
Performs batch generation using vLLM engine, which supports tensor parallelism.
r"""Perform batch generation using vLLM engine, which supports tensor parallelism.
Usage: python vllm_infer.py --model_name_or_path meta-llama/Llama-2-7b-hf --template llama --dataset alpaca_en_demo
"""
check_version("vllm>=0.4.3,<=0.7.3")