various bugfixes

configs/llama_65B_alpaca.yml

@@ -1,4 +1,4 @@
-base_model: decapoda-research/llama-65b-hf
+base_model: huggyllama/llama-7b
 model_type: LlamaForCausalLM
 tokenizer_type: LlamaTokenizer
 load_in_8bit: true
@@ -33,8 +33,8 @@ num_epochs: 5
 learning_rate: 0.00003
 train_on_inputs: false
 group_by_length: false
-bf16: True
-tf32: True
+bf16: true
+tf32: true
 resume_from_checkpoint:
 local_rank:
 deepspeed:

requirements.txt

@@ -10,3 +10,6 @@ accelerate
 sentencepiece
 wandb
 flash-attn
+deepspeed
+einops
+

scripts/finetune.py

@@ -11,7 +11,7 @@ import torch
 import transformers
 import yaml
 from attrdict import AttrDefault
-from datasets import load_dataset, IterableDataset, Dataset
+from datasets import load_dataset, IterableDataset, Dataset, load_from_disk
 from peft import (
     LoraConfig,
     get_peft_model,
@@ -52,8 +52,9 @@ def load_model(base_model, model_type, tokenizer_type, cfg, adapter="lora"):
     if adapter != "lora":
         raise NotImplementedError(f"{adapter} peft adapter not available")
     if "llama" in base_model:
-        from axolotl.flash_attn import replace_llama_attn_with_flash_attn
-        replace_llama_attn_with_flash_attn()
+        if cfg.device not in ["mps", "cpu"]:
+            from axolotl.flash_attn import replace_llama_attn_with_flash_attn
+            replace_llama_attn_with_flash_attn()
 
     try:
         if "llama" in base_model:
@@ -86,7 +87,7 @@ def load_model(base_model, model_type, tokenizer_type, cfg, adapter="lora"):
     except:
         tokenizer = AutoTokenizer.from_pretrained(base_model)
 
-    if tokenizer.__class__.__name__ == "LlamaTokenizer":
+    if tokenizer.__class__.__name__ in ["LlamaTokenizer", "LlamaTokenizerFast"]:
         tokenizer.pad_token = LLAMA_DEFAULT_PAD_TOKEN
 
     if tokenizer.__class__.__name__ == "GPTNeoXTokenizerFast":
@@ -255,8 +256,9 @@ def train(
         return
 
     datasets = []
-    if len(cfg.datasets) == 1 and cfg.datasets[0].type == "arrow":
-        dataset = load_dataset(cfg.datasets[0].path, split="train")
+    if not isinstance(cfg.datasets, list) and isinstance(cfg.datasets, str):
+        # assumption that we are loading a previously saved/cached dataset
+        dataset = load_from_disk(cfg.datasets)
     else:
         for d in cfg.datasets:
             ds: IterableDataset = load_dataset(
@@ -288,7 +290,6 @@ def train(
             [_ for _ in constant_len_dataset]
         ).train_test_split(test_size=cfg.val_set_size, shuffle=True, seed=42)
         dataset.save_to_disk("data/last_run")
-        print(dataset)
 
     train_dataset = dataset["train"]
     eval_dataset = dataset["test"]

setup.cfg

@@ -23,6 +23,7 @@ install_requires =
     sentencepiece
     wandb
     flash-attn
+    einops
 
 [options.packages.find]
 where = src

src/axolotl/datasets.py

@@ -93,22 +93,24 @@ class ConstantLengthDataset(IterableDataset):
                     buffer_len = 0
 
                 if example:
-                    input_ids = example["input_ids"]
-                    attention_mask = example["attention_mask"]
-                    labels = example["labels"]
-
-                    if add_concat_token:
-                        input_ids.append(self.concat_token_id)
-                        attention_mask.append(1)
-                        labels.append(self.concat_token_id)
-
-                    input_ids_with_concat = torch.tensor(input_ids, dtype=torch.long)
-                    attention_mask_with_concat = torch.tensor(
-                        attention_mask, dtype=torch.long
-                    )
-                    labels_with_concat = torch.tensor(labels, dtype=torch.long)
-
-                    buffer["input_ids"].append(input_ids_with_concat)
-                    buffer["attention_mask"].append(attention_mask_with_concat)
-                    buffer["labels"].append(labels_with_concat)
-                    buffer_len += len(input_ids)
+                    # just going to drop data points that are too long
+                    if len(example["input_ids"]) <= self.seq_length:
+                        input_ids = example["input_ids"]
+                        attention_mask = example["attention_mask"]
+                        labels = example["labels"]
+
+                        if add_concat_token:
+                            input_ids.append(self.concat_token_id)
+                            attention_mask.append(1)
+                            labels.append(self.concat_token_id)
+
+                        input_ids_with_concat = torch.tensor(input_ids, dtype=torch.long)
+                        attention_mask_with_concat = torch.tensor(
+                            attention_mask, dtype=torch.long
+                        )
+                        labels_with_concat = torch.tensor(labels, dtype=torch.long)
+
+                        buffer["input_ids"].append(input_ids_with_concat)
+                        buffer["attention_mask"].append(attention_mask_with_concat)
+                        buffer["labels"].append(labels_with_concat)
+                        buffer_len += len(input_ids)