add cal_lr.py

src/llmtuner/tuner/core/loader.py

@@ -84,10 +84,9 @@ def load_model_and_tokenizer(
         tokenizer._pad = MethodType(PreTrainedTokenizerBase._pad, tokenizer)
 
     # Set model dtype
-    if model_args.compute_dtype is not None: # for training
-        setattr(config, "torch_dtype", model_args.compute_dtype)
-    else: # for evaluation, priority: bf16 > fp16 > fp32
+    if model_args.compute_dtype is None: # priority: bf16 > fp16 > fp32
         model_args.compute_dtype = infer_optim_dtype(model_dtype=getattr(config, "torch_dtype", None))
+    setattr(config, "torch_dtype", model_args.compute_dtype)
 
     # Fix config (for Qwen)
     if getattr(config, "model_type", None) == "qwen":

tests/cal_flops.py

@@ -12,7 +12,7 @@ from deepspeed.profiling.flops_profiler import get_model_profile # type: ignore
 from llmtuner import ChatModel
 
 
-def calculate(
+def calculate_flops(
     model_name_or_path: str,
     batch_size: Optional[int] = 1,
     seq_length: Optional[int] = 256,
@@ -41,4 +41,4 @@ def calculate(
 
 
 if __name__ == "__main__":
-    fire.Fire(calculate)
+    fire.Fire(calculate_flops)

tests/cal_lr.py

@@ -0,0 +1,63 @@
+# coding=utf-8
+# Calculates 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 --cutoff_len 1024 --batch_size 16
+# Inspired by: https://github.com/imoneoi/openchat/blob/master/ochat/training_deepspeed/train.py
+
+import fire
+import math
+import torch
+from tqdm import tqdm
+from torch.utils.data import DataLoader
+from transformers import DataCollatorForSeq2Seq
+
+from llmtuner.dsets import get_dataset, preprocess_dataset
+from llmtuner.extras.constants import IGNORE_INDEX
+from llmtuner.tuner.core import get_train_args, load_model_and_tokenizer
+
+
+BASE_LR = 3e-4
+BASE_BS = 4_000_000
+
+
+def calculate_lr(
+    model_name_or_path: str,
+    dataset: str,
+    cutoff_len: int,
+    batch_size: int
+):
+    model_args, data_args, training_args, finetuning_args, _ = get_train_args(dict(
+        stage="sft",
+        model_name_or_path=model_name_or_path,
+        dataset=dataset,
+        template="default",
+        cutoff_len=cutoff_len,
+        output_dir="dummy_dir",
+        fp16=True
+    ))
+    trainset = get_dataset(model_args, data_args)
+    _, tokenizer = load_model_and_tokenizer(model_args, finetuning_args, is_trainable=False, stage="sft")
+    trainset = preprocess_dataset(trainset, tokenizer, data_args, training_args, stage="sft")
+    data_collator = DataCollatorForSeq2Seq(tokenizer=tokenizer, label_pad_token_id=IGNORE_INDEX)
+    dataloader = DataLoader(
+        dataset=trainset,
+        batch_size=batch_size,
+        shuffle=True,
+        collate_fn=data_collator,
+        pin_memory=True
+    )
+    valid_tokens, total_tokens = 0, 0
+    for batch in tqdm(dataloader):
+        valid_tokens += torch.sum(batch["labels"] != IGNORE_INDEX).item()
+        total_tokens += torch.numel(batch["labels"])
+
+    batch_max_len = cutoff_len * batch_size # max tokens in a batch
+    valid_ratio = valid_tokens / total_tokens
+    batch_valid_len = batch_max_len * valid_ratio
+    lr = BASE_LR * math.sqrt(batch_valid_len / BASE_BS)
+    print("Optimal learning rate is {:.2e} for valid ratio% {:.2f} and effective batch size {:.2f}".format(
+        lr, valid_ratio * 100, batch_valid_len
+    ))
+
+
+if __name__ == "__main__":
+    fire.Fire(calculate_lr)

tests/quantize.py

@@ -4,7 +4,6 @@
 #                           --max_length 1024 --max_samples 1024
 # dataset format: instruction (string), input (string), output (string), history (List[string])
 
-
 import fire
 from datasets import load_dataset
 from transformers import AutoTokenizer