Merge branch 'main' into patch-2

configs/galactica_1_3B.yml

@@ -34,7 +34,7 @@ tf32: false
 early_stopping_patience:
 resume_from_checkpoint:
 local_rank:
-special_tokens:
+tokens:
   pad_token: "[PAD]"
   bos_token: "<s>"
   eos_token: "</s>"

configs/llama_7B_jeopardy.yml

@@ -51,7 +51,7 @@ deepspeed:
 weight_decay: 0.0001
 fsdp:
 fsdp_config:
-special_tokens:
+tokens:
   pad_token: "[PAD]"
   bos_token: "<s>"
   eos_token: "</s>"

configs/stability_3b.yml

@@ -49,7 +49,7 @@ deepspeed:
 weight_decay: 0.01
 fsdp:
 fsdp_config:
-#special_tokens:
+#tokens:
 #  pad_token: "[PAD]"
 #  bos_token: "<s>"
 #  eos_token: "</s>"

examples/4bit-lora-7b/config.yml

@@ -55,7 +55,7 @@ deepspeed:
 weight_decay: 0.0001
 fsdp:
 fsdp_config:
-special_tokens:
+tokens:
   pad_token: "[PAD]"
   bos_token: "<s>"
   eos_token: "</s>"

examples/mpt-7b/config.yml

@@ -1,7 +1,6 @@
 base_model: mosaicml/mpt-7b
 base_model_config: mosaicml/mpt-7b
-model_type: AutoModelForCausalLM
-tokenizer_type: GPTNeoXTokenizer
+tokenizer_type: AutoTokenizer
 trust_remote_code: true  # required for mpt as their model class is not merged into transformers yet
 load_in_8bit: false
 datasets:
@@ -25,7 +24,7 @@ wandb_watch:
 wandb_run_id:
 wandb_log_model: checkpoint
 output_dir: ./mpt-alpaca-7b
-batch_size: 4
+batch_size: 1
 micro_batch_size: 1
 num_epochs: 3
 optimizer: adamw_bnb_8bit
@@ -52,7 +51,7 @@ deepspeed:
 weight_decay: 0.0001
 fsdp:
 fsdp_config:
-special_tokens:
+tokens:
   pad_token: "<|padding|>"
   bos_token: "<|endoftext|>"
   eos_token: "<|endoftext|>"

examples/redpajama/config-3b.yml

@@ -52,7 +52,7 @@ deepspeed:
 weight_decay: 0.0001
 fsdp:
 fsdp_config:
-special_tokens:
+tokens:
   pad_token: "<|padding|>"
   bos_token: "<|endoftext|>"
   eos_token: "<|endoftext|>"

scripts/finetune.py

@@ -191,7 +191,9 @@ def train(
     if cfg.debug:
         logging.info("check_dataset_labels...")
         check_dataset_labels(
-            train_dataset.select([random.randrange(0, len(train_dataset) - 1) for i in range(5)]),
+            train_dataset.select(
+                [random.randrange(0, len(train_dataset) - 1) for i in range(5)]
+            ),
             tokenizer,
         )
 
@@ -218,20 +220,23 @@ def train(
     logging.info("Starting trainer...")
     resume_from_checkpoint = cfg.resume_from_checkpoint
     if cfg.resume_from_checkpoint is None and cfg.auto_resume_from_checkpoints:
-        possible_checkpoints = [str(cp) for cp in Path(cfg.output_dir).glob("checkpoint-*")]
+        possible_checkpoints = [
+            str(cp) for cp in Path(cfg.output_dir).glob("checkpoint-*")
+        ]
         if len(possible_checkpoints) > 0:
-            sorted_paths = sorted(possible_checkpoints, key=lambda path: int(path.split('-')[-1]))
+            sorted_paths = sorted(
+                possible_checkpoints, key=lambda path: int(path.split("-")[-1])
+            )
             resume_from_checkpoint = sorted_paths[-1]
-            logging.info(f"Using Auto-resume functionality to start with checkpoint at {resume_from_checkpoint}")
+            logging.info(
+                f"Using Auto-resume functionality to start with checkpoint at {resume_from_checkpoint}"
+            )
     trainer.train(resume_from_checkpoint=resume_from_checkpoint)
 
-    logging.info(
-        f"Training Completed!!! Saving pre-trained model to {cfg.output_dir}"
-    )
-    
+    logging.info(f"Training Completed!!! Saving pre-trained model to {cfg.output_dir}")
 
     # TODO do we need this fix? https://huggingface.co/docs/accelerate/usage_guides/fsdp#saving-and-loading
-    trainer.save_pretrained(cfg.output_dir)
+    model.save_pretrained(cfg.output_dir)
     # trainer.save_model(cfg.output_dir)  # TODO this may be needed for deepspeed to work? need to review another time
 
 

setup.py

@@ -10,22 +10,22 @@ with open("./requirements.txt", "r") as requirements_file:
         install_requires.append(r)
 
 setup(
-    name='axolotl',
-    version='0.1',
+    name="axolotl",
+    version="0.1",
     description="You know you're going to axolotl questions",
-    package_dir={'': 'src'},
+    package_dir={"": "src"},
     packages=find_packages(),
     install_requires=install_requires,
     extras_require={
-        'int4': [
+        "int4": [
             "alpaca_lora_4bit @ git+https://github.com/winglian/alpaca_lora_4bit.git@setup_pip",
         ],
-        'int4_triton': [
+        "int4_triton": [
             "alpaca_lora_4bit[triton] @ git+https://github.com/winglian/alpaca_lora_4bit.git@setup_pip",
         ],
-        'extras': [
-            'flash-attn',
-            'deepspeed',
-        ]
+        "extras": [
+            "flash-attn",
+            "deepspeed",
+        ],
     },
 )

src/axolotl/datasets.py

@@ -31,6 +31,7 @@ class TokenizedPromptDataset(IterableDataset):
             except InvalidDataException:
                 pass
 
+
 # TODO this isn't the best since it can't interleave datasets
 class ConstantLengthDataset(IterableDataset):
     """
@@ -40,6 +41,7 @@ class ConstantLengthDataset(IterableDataset):
             dataset (dataset.Dataset): Dataset with text files.
             seq_length (int): Length of token sequences to return.
     """
+
     def __init__(
         self,
         tokenizer,
@@ -93,14 +95,19 @@ class ConstantLengthDataset(IterableDataset):
                             : self.seq_length
                         ]
                         labels = torch.cat(buffer["labels"], dim=-1)[: self.seq_length]
-                        if labels.size() == input_ids.size() and attention_mask.size() == input_ids.size():
+                        if (
+                            labels.size() == input_ids.size()
+                            and attention_mask.size() == input_ids.size()
+                        ):
                             yield {
                                 "input_ids": input_ids,
                                 "labels": labels,
                                 "attention_mask": attention_mask,
                             }
                         else:
-                            logging.warning("dropping batch due to tensor size mismatch")
+                            logging.warning(
+                                "dropping batch due to tensor size mismatch"
+                            )
                     buffer = {"input_ids": [], "attention_mask": [], "labels": []}
                     buffer_len = 0
 
@@ -116,11 +123,15 @@ class ConstantLengthDataset(IterableDataset):
                             attention_mask.append(1)
                             labels.append(self.concat_token_id)
 
-                        input_ids_with_concat = torch.tensor(input_ids, dtype=self.tokens_dtype)
+                        input_ids_with_concat = torch.tensor(
+                            input_ids, dtype=self.tokens_dtype
+                        )
                         attention_mask_with_concat = torch.tensor(
                             attention_mask, dtype=self.tokens_dtype
                         )
-                        labels_with_concat = torch.tensor(labels, dtype=self.tokens_dtype)
+                        labels_with_concat = torch.tensor(
+                            labels, dtype=self.tokens_dtype
+                        )
 
                         buffer["input_ids"].append(input_ids_with_concat)
                         buffer["attention_mask"].append(attention_mask_with_concat)

src/axolotl/prompt_tokenizers.py

@@ -126,10 +126,8 @@ class NomicGPT4AllPromptTokenizingStrategy(InstructionPromptTokenizingStrategy):
 
 
 class CompletionPromptTokenizingStrategy(InstructionPromptTokenizingStrategy):
-    def parse_instruction_fields(self, prompt) -> (str):
-        return (
-            prompt["text"]
-        )
+    def parse_instruction_fields(self, prompt) -> str:
+        return prompt["text"]
 
     def tokenize_prompt(self, prompt):
         instruction = self.parse_instruction_fields(prompt)
@@ -139,9 +137,7 @@ class CompletionPromptTokenizingStrategy(InstructionPromptTokenizingStrategy):
         return tokenized_full_prompt
 
     def _build_full_prompt(self, instruction):
-        return self.prompter.build_prompt(
-            instruction
-        )
+        return self.prompter.build_prompt(instruction)
 
 
 class ReflectionPromptTokenizingStrategy(PromptTokenizingStrategy):
@@ -149,8 +145,16 @@ class ReflectionPromptTokenizingStrategy(PromptTokenizingStrategy):
         raise NotImplementedError
 
     def tokenize_prompt(self, prompt):
-        instruction, input, output, reflection, corrected = self.parse_instruction_fields(prompt)
-        full_prompt = self._build_full_prompt(instruction, input, output, reflection, corrected)
+        (
+            instruction,
+            input,
+            output,
+            reflection,
+            corrected,
+        ) = self.parse_instruction_fields(prompt)
+        full_prompt = self._build_full_prompt(
+            instruction, input, output, reflection, corrected
+        )
         tokenized_full_prompt = self._tokenize(full_prompt)
         if not self.train_on_inputs:
             user_prompt = self.prompter.build_prompt(

src/axolotl/prompters.py

@@ -36,10 +36,7 @@ class JeopardyPrompter(AlpacaPrompter):
 
 
 class CompletionPrompter(AlpacaPrompter):
-    def build_prompt(
-        self,
-        instruction: str
-    ) -> str:
+    def build_prompt(self, instruction: str) -> str:
         return instruction
 
     def get_response(self, output: str) -> str:
@@ -75,7 +72,9 @@ class ReflectAlpacaPrompter:
         else:
             res = self.prompt_no_input.format(instruction=instruction)
         if output and reflection and corrected:
-            label = self.agent_label.format(output=output, reflection=reflection, corrected=corrected)
+            label = self.agent_label.format(
+                output=output, reflection=reflection, corrected=corrected
+            )
             res = f"{res}{label}"
         return res
 
@@ -200,9 +199,13 @@ class ShareGPTPrompter:
             if len(parts) != 2:
                 break
             parts[0] += sep
-            round_len = len(tokenizer(rou)["input_ids"]) - 1  # -1 ignores the bos_token generated for this
+            round_len = (
+                len(tokenizer(rou)["input_ids"]) - 1
+            )  # -1 ignores the bos_token generated for this
             # we have to strip the initial part, any dangling whitespace creates an additional ghost token
-            instruction_len = len(tokenizer(parts[0].strip())["input_ids"]) - 1  # -1 ignores the bos_token generated for this
+            instruction_len = (
+                len(tokenizer(parts[0].strip())["input_ids"]) - 1
+            )  # -1 ignores the bos_token generated for this
             target[cur_len : cur_len + instruction_len] = [
                 IGNORE_TOKEN_ID
             ] * instruction_len
@@ -212,7 +215,7 @@ class ShareGPTPrompter:
                 break
 
         # Fix: Truncate the target to have the same length as input_ids
-        target = target[:len(tokenized_result["input_ids"])]
+        target = target[: len(tokenized_result["input_ids"])]
         # target[cur_len:] = [IGNORE_TOKEN_ID] * (len(target) - cur_len)
 
         attention_mask = [

src/axolotl/utils/callbacks.py

@@ -1,8 +1,15 @@
 import os
 
-from transformers import Seq2SeqTrainer, TrainerCallback, TrainingArguments, TrainerState, TrainerControl
+from transformers import (
+    Seq2SeqTrainer,
+    TrainerCallback,
+    TrainingArguments,
+    TrainerState,
+    TrainerControl,
+)
 from transformers.trainer_utils import PREFIX_CHECKPOINT_DIR
 
+
 class SavePeftModelCallback(TrainerCallback):
     def on_save(
         self,
@@ -11,7 +18,9 @@ class SavePeftModelCallback(TrainerCallback):
         control: TrainerControl,
         **kwargs,
     ):
-        checkpoint_folder = os.path.join(args.output_dir, f"{PREFIX_CHECKPOINT_DIR}-{state.global_step}")
+        checkpoint_folder = os.path.join(
+            args.output_dir, f"{PREFIX_CHECKPOINT_DIR}-{state.global_step}"
+        )
 
         peft_model_path = os.path.join(checkpoint_folder, "adapter_model")
         kwargs["model"].save_pretrained(peft_model_path)

src/axolotl/utils/data.py

@@ -2,7 +2,13 @@ import logging
 from hashlib import md5
 from pathlib import Path
 
-from datasets import load_from_disk, load_dataset, IterableDataset, Dataset, concatenate_datasets
+from datasets import (
+    load_from_disk,
+    load_dataset,
+    IterableDataset,
+    Dataset,
+    concatenate_datasets,
+)
 from huggingface_hub import hf_hub_download
 
 from axolotl.datasets import TokenizedPromptDataset, ConstantLengthDataset
@@ -75,7 +81,9 @@ def load_prepare_datasets(tokenizer, cfg, default_dataset_prepared_path):
                 else:
                     ds = load_dataset(d.path, streaming=True)
             else:
-                fp = hf_hub_download(repo_id=d.path, repo_type="dataset", filename=d.data_files)
+                fp = hf_hub_download(
+                    repo_id=d.path, repo_type="dataset", filename=d.data_files
+                )
                 ds = load_dataset("json", data_files=fp, streaming=True, split=None)
             if not ds:
                 raise Exception("unhandled dataset load")
@@ -140,7 +148,9 @@ def load_prepare_datasets(tokenizer, cfg, default_dataset_prepared_path):
             samples = samples + [i for i in d]
         dataset = Dataset.from_list(samples).shuffle(seed=42)
         if cfg.local_rank == 0:
-            logging.info(f"Saving merged prepared dataset to disk... {prepared_ds_path}")
+            logging.info(
+                f"Saving merged prepared dataset to disk... {prepared_ds_path}"
+            )
             dataset.save_to_disk(prepared_ds_path)
 
     if cfg.max_packed_sequence_len is not None:
@@ -153,12 +163,14 @@ def load_prepare_datasets(tokenizer, cfg, default_dataset_prepared_path):
         dataset = Dataset.from_list([_ for _ in constant_len_dataset])
 
     if cfg.dataset_shard_num and cfg.dataset_shard_idx is not None:
-        logging.info(f"Using index #{cfg.dataset_shard_idx} of {cfg.dataset_shard_num} shards")
-        dataset = dataset.shard(num_shards=cfg.dataset_shard_num, index=cfg.dataset_shard_idx)
+        logging.info(
+            f"Using index #{cfg.dataset_shard_idx} of {cfg.dataset_shard_num} shards"
+        )
+        dataset = dataset.shard(
+            num_shards=cfg.dataset_shard_num, index=cfg.dataset_shard_idx
+        )
 
-    dataset = dataset.train_test_split(
-        test_size=cfg.val_set_size, shuffle=False
-    )
+    dataset = dataset.train_test_split(test_size=cfg.val_set_size, shuffle=False)
     train_dataset = dataset["train"]
     eval_dataset = dataset["test"]
 

src/axolotl/utils/models.py

@@ -9,14 +9,18 @@ from transformers import (
     AutoModelForCausalLM,
     AutoTokenizer,
     PreTrainedModel,
+    AutoConfig,
 )
+
 try:
     from transformers import (
         LlamaForCausalLM,
         LlamaTokenizer,
     )
 except:
-    logging.warning("This version of transformers does not support Llama. Consider upgrading.")
+    logging.warning(
+        "This version of transformers does not support Llama. Consider upgrading."
+    )
 
 from axolotl.prompt_tokenizers import LLAMA_DEFAULT_PAD_TOKEN
 
@@ -40,7 +44,9 @@ def load_model(
     # TODO refactor as a kwarg
     load_in_8bit = cfg.load_in_8bit
     tokenizer = None
-    is_llama_derived_model = "llama" in base_model or (cfg.model_type and "llama" in cfg.model_type.lower())
+    is_llama_derived_model = "llama" in base_model or (
+        cfg.model_type and "llama" in cfg.model_type.lower()
+    )
 
     if is_llama_derived_model and cfg.flash_attention:
         if cfg.device not in ["mps", "cpu"] and inference is False:
@@ -49,11 +55,16 @@ def load_model(
             logging.info("patching with flash attention")
             replace_llama_attn_with_flash_attn()
     elif is_llama_derived_model and cfg.xformers_attention:
-        from alpaca_lora_4bit.monkeypatch.llama_attn_hijack_xformers import hijack_llama_attention
+        from alpaca_lora_4bit.monkeypatch.llama_attn_hijack_xformers import (
+            hijack_llama_attention,
+        )
+
         logging.info("patching with xformers attention")
         hijack_llama_attention()
 
-    torch_dtype = torch.float16 if cfg.load_in_8bit or cfg.fp16 or cfg.bf16 else torch.float32
+    torch_dtype = (
+        torch.float16 if cfg.load_in_8bit or cfg.fp16 or cfg.bf16 else torch.float32
+    )
     try:
         if cfg.load_4bit:
             from alpaca_lora_4bit.monkeypatch.peft_tuners_lora_monkey_patch import (
@@ -74,8 +85,12 @@ def load_model(
             try:
                 snapshot_download_kwargs = {}
                 if cfg.base_model_ignore_patterns:
-                    snapshot_download_kwargs["ignore_patterns"] = cfg.base_model_ignore_patterns
-                cache_model_path = Path(snapshot_download(base_model, **snapshot_download_kwargs))
+                    snapshot_download_kwargs[
+                        "ignore_patterns"
+                    ] = cfg.base_model_ignore_patterns
+                cache_model_path = Path(
+                    snapshot_download(base_model, **snapshot_download_kwargs)
+                )
                 files = (
                     list(cache_model_path.glob("*.pt"))
                     + list(cache_model_path.glob("*.safetensors"))
@@ -116,8 +131,13 @@ def load_model(
                 trust_remote_code=True if cfg.trust_remote_code is True else False,
             )
         else:
+            config = AutoConfig.from_pretrained(
+                base_model,
+                trust_remote_code=True if cfg.trust_remote_code is True else False,
+            )
             model = AutoModelForCausalLM.from_pretrained(
                 base_model,
+                config=config,
                 load_in_8bit=cfg.load_in_8bit and cfg.adapter is not None,
                 torch_dtype=torch_dtype,
                 device_map=cfg.device_map,

src/axolotl/utils/schedulers.py

@@ -26,7 +26,10 @@ class InterpolatingLogScheduler(LRScheduler):
         if self.last_epoch <= 0:
             lrs = [self.min_lr for base_lr in self.base_lrs]
         elif self.last_epoch < self.num_steps:
-            lrs = [self.min_lr * (self.q ** (self.last_epoch - 1)) for base_lr in self.base_lrs]
+            lrs = [
+                self.min_lr * (self.q ** (self.last_epoch - 1))
+                for base_lr in self.base_lrs
+            ]
         else:
             lrs = [self.max_lr for base_lr in self.base_lrs]
 

src/axolotl/utils/tokenization.py

@@ -1,6 +1,7 @@
 from termcolor import colored
 import logging
 
+
 def check_dataset_labels(dataset, tokenizer):
     # the dataset is already shuffled, so let's just check the first 5 elements
     for idx in range(5):
@@ -11,7 +12,7 @@ def check_example_labels(example, tokenizer):
     # Get the input_ids, labels, and attention_mask from the dataset
     input_ids = example["input_ids"]
     labels = example["labels"]
-    attention_mask =example["attention_mask"]
+    attention_mask = example["attention_mask"]
 
     # You can compare the input_ids and labels element-wise
     # Remember to ignore positions with IGNORE_TOKEN_ID (if you use it) or attention_mask equal to 0
@@ -21,9 +22,7 @@ def check_example_labels(example, tokenizer):
     ):
         decoded_input_token = tokenizer.decode(input_id)
         # Choose the color based on whether the label has the ignore value or not
-        color = (
-            "red" if label_id == -100 else ("yellow" if label_id == 0 else "green")
-        )
+        color = "red" if label_id == -100 else ("yellow" if label_id == 0 else "green")
         colored_token = colored(decoded_input_token, color) + colored(
             f"({label_id}, {mask}, {input_id})", "white"
         )

src/axolotl/utils/trainer.py

@@ -30,16 +30,8 @@ def setup_trainer(cfg, train_dataset, eval_dataset, model, tokenizer):
         if cfg.logging_steps is not None
         else max(min(int(0.005 * total_num_steps), 10), 1)
     )
-    save_steps = (
-        cfg.save_steps
-        if cfg.save_steps is not None
-        else min(int(0.05 * total_num_steps), 200)
-    )
-    eval_steps = (
-        cfg.eval_steps
-        if cfg.eval_steps is not None and save_steps % cfg.eval_steps == 0
-        else save_steps
-    )
+    save_steps = cfg.save_steps
+    eval_steps = cfg.eval_steps
 
     training_arguments_kwargs = {}
     if cfg.bf16 == "full":
@@ -86,26 +78,33 @@ def setup_trainer(cfg, train_dataset, eval_dataset, model, tokenizer):
 
     training_args = transformers.TrainingArguments(
         per_device_train_batch_size=cfg.micro_batch_size,
-        per_device_eval_batch_size=cfg.eval_batch_size if cfg.eval_batch_size is not None else cfg.micro_batch_size,
+        per_device_eval_batch_size=cfg.eval_batch_size
+        if cfg.eval_batch_size is not None
+        else cfg.micro_batch_size,
         gradient_accumulation_steps=cfg.gradient_accumulation_steps,
         eval_accumulation_steps=cfg.gradient_accumulation_steps,
         num_train_epochs=cfg.num_epochs,
         learning_rate=cfg.learning_rate,
         evaluation_strategy="steps" if cfg.val_set_size > 0 else "no",
-        save_strategy="steps",
+        save_strategy="steps" if save_steps else "epoch",
         eval_steps=eval_steps if cfg.val_set_size > 0 else None,
         save_steps=save_steps,
         output_dir=cfg.output_dir,
         save_total_limit=3,
         load_best_model_at_end=True
-        if cfg.val_set_size > 0 and save_steps % eval_steps == 0 and cfg.load_in_8bit is not True
+        if cfg.val_set_size > 0
+        and save_steps is not None
+        and save_steps % eval_steps == 0
+        and cfg.load_in_8bit is not True
         else False,
         ddp_find_unused_parameters=False if cfg.ddp else None,
         group_by_length=cfg.group_by_length,
         report_to="wandb" if cfg.use_wandb else None,
         run_name=cfg.wandb_run_id if cfg.use_wandb else None,
         optim=cfg.optimizer if cfg.optimizer else "adamw_hf",
-        lr_scheduler_type=cfg.lr_scheduler if cfg.lr_scheduler and cfg.lr_scheduler not in ("one_cycle", "log_sweep") else "cosine",
+        lr_scheduler_type=cfg.lr_scheduler
+        if cfg.lr_scheduler and cfg.lr_scheduler not in ("one_cycle", "log_sweep")
+        else "cosine",
         weight_decay=cfg.weight_decay if cfg.weight_decay is not None else 0.0,
         **training_arguments_kwargs,
     )
@@ -158,6 +157,7 @@ def setup_trainer(cfg, train_dataset, eval_dataset, model, tokenizer):
                 cfg.learning_rate,
                 total_steps=total_num_steps,
                 epochs=cfg.num_epochs,
+                div_factor=10,
                 **lr_scheduler_kwargs,
             )
         elif cfg.lr_scheduler == "log_sweep":

src/axolotl/utils/wandb.py

@@ -2,7 +2,9 @@ import os
 
 
 def setup_wandb_env_vars(cfg):
-    if cfg.wandb_project and len(cfg.wandb_project) > 0:
+    if cfg.wandb_mode and cfg.wandb_mode == "offline":
+        os.environ["WANDB_MODE"] = cfg.wandb_mode
+    elif cfg.wandb_project and len(cfg.wandb_project) > 0:
         os.environ["WANDB_PROJECT"] = cfg.wandb_project
         cfg.use_wandb = True
         if cfg.wandb_watch and len(cfg.wandb_watch) > 0: