yi-34b-dpo-rawrr-v2-2-hf.py

from unsloth import FastLanguageModel
from datasets import Dataset, load_dataset
from dataclasses import dataclass, field
from typing import Dict, Optional
import torch
max_seq_length = 4096 # Choose any! We auto support RoPE Scaling internally!
dtype = None # None for auto detection. Float16 for Tesla T4, V100, Bfloat16 for Ampere+
load_in_4bit = True # Use 4bit quantization to reduce memory usage. Can be False.

model, tokenizer = FastLanguageModel.from_pretrained(
    model_name = "larryvrh/Yi-34B-200K-Llamafied", # Choose ANY! eg mistralai/Mistral-7B-Instruct-v0.2
    max_seq_length = max_seq_length,
    attn_implementation="flash_attention_2",
    dtype = dtype,
    load_in_4bit = load_in_4bit,
    # token = "hf_...", # use one if using gated models like meta-llama/Llama-2-7b-hf
)



#@title Alignment Handbook utils
import os
import re
from typing import List, Literal, Optional

from datasets import DatasetDict, concatenate_datasets, load_dataset, load_from_disk
from datasets.builder import DatasetGenerationError


#DEFAULT_CHAT_TEMPLATE = "{% for message in messages %}\n{% if message['role'] == 'user' %}\n{{ '<|user|>\n' + message['content'] + eos_token }}\n{% elif message['role'] == 'system' %}\n{{ '<|system|>\n' + message['content'] + eos_token }}\n{% elif message['role'] == 'assistant' %}\n{{ '<|assistant|>\n'  + message['content'] + eos_token }}\n{% endif %}\n{% if loop.last and add_generation_prompt %}\n{{ '<|assistant|>' }}\n{% endif %}\n{% endfor %}"
tokenizer.chat_template = "{% if not add_generation_prompt is defined %}{% set add_generation_prompt = false %}{% endif %}{% for message in messages %}{{'<|im_start|>' + message['role'] + '\n' + message['content'] + '<|im_end|>' + '\n'}}{% endfor %}{% if add_generation_prompt %}{{ '<|im_start|>assistant\n' }}{% endif %}"

def chatml_format(example):
    # Format system
    if len(example['system']) > 0:
        message = {"role": "system", "content": example['system']}
        system = tokenizer.apply_chat_template([message], tokenize=False)
    else:
        system = ""

    # Format instruction
    message = {"role": "user", "content": example['prompt']}
    prompt = tokenizer.apply_chat_template([message], tokenize=False, add_generation_prompt=True)

    # Format chosen answer
    chosen = example['chosen'] + "<|im_end|>\n"

    # Format rejected answer
    rejected = example['rejected'] + "<|im_end|>\n"

    return {
        "prompt": system + prompt,
        "chosen": chosen,
        "rejected": rejected,
    }

# Load dataset
dataset = load_dataset("adamo1139/rawrr_v2", split="train")

import pprint
pprint.pprint("""NOT a formatted dataset
""")
pprint
pprint.pprint(dataset[250])
pprint.pprint(dataset[260])
pprint.pprint(dataset[270])
pprint.pprint(dataset[280])
pprint.pprint(dataset[290])
# Save columns
original_columns = dataset.column_names

# Format dataset
dataset = dataset.map(
    chatml_format,
    remove_columns=original_columns
)

# Print sample
pprint.pprint("""formatted dataset""")
pprint.pprint(dataset[250])
pprint.pprint(dataset[260])
pprint.pprint(dataset[270])
pprint.pprint(dataset[280])
pprint.pprint(dataset[290])


model = FastLanguageModel.get_peft_model(
    model,
    r = 16, # Choose any number > 0 ! Suggested 8, 16, 32, 64, 128
    target_modules = ["q_proj", "k_proj", "v_proj", "o_proj",
                      "gate_proj", "up_proj", "down_proj",],
    lora_alpha = 32,
    lora_dropout = 0, # Currently only supports dropout = 0
    bias = "none",    # Currently only supports bias = "none"
    use_gradient_checkpointing = True,
    random_state = 3407,
    use_rslora = False,  # We support rank stabilized LoRA
    loftq_config = None, # And LoftQ
)

from transformers import AutoModelForCausalLM, AutoTokenizer, BitsAndBytesConfig, HfArgumentParser, TrainingArguments
from trl import DPOTrainer

dpo_trainer = DPOTrainer(
    model = model,
    ref_model = None,
    args = TrainingArguments(
        per_device_train_batch_size = 1,
        gradient_accumulation_steps = 16,
        warmup_ratio = 0.05,
        num_train_epochs = 1,
        learning_rate = 0.000045,
        fp16 = not torch.cuda.is_bf16_supported(),
        bf16 = torch.cuda.is_bf16_supported(),
        logging_steps = 1,
        optim = "adamw_8bit",
        weight_decay = 0.0,
        lr_scheduler_type = "linear",
        seed = 42,
        save_strategy = "steps",
        save_steps = 100,
        save_total_limit = 10,
        output_dir = "rawrr_v2_run2",
    ),
    beta = 0.1,
    train_dataset = dataset,
    # eval_dataset = raw_datasets["test"],
    tokenizer = tokenizer,
    max_length = 700,
    max_prompt_length = 400,
)
dpo_trainer.train()
model.save_pretrained("yi-34b-200k_rawrr_v2_run2") # Local saving