Finetune Llama 3, Mistral & Gemma LLMs 2-5x faster with 80% less memory
Llama-3 (15 trillion tokens, GPT3.5 level) is fully supported! Get 2x faster, 60% less VRAM usage than HF + FA2!
Colab notebook: https://colab.research.google.com/drive/135ced7oHytdxu3N2DNe1Z0kqjyYIkDXp?usp=sharing
Pre-quantized 8b and 70b weights (4x faster downloading) via https://huggingface.co/unsloth
Full Changelog: https://github.com/unslothai/unsloth/compare/April-2024...April-Llama-3-2024
You can now 2x your batch size or train on long context windows with Unsloth! 228K context windows on H100s are now possible (4x longer than HF+FA2) with Mistral 7b.
How? We coded up an async offloaded gradient checkpointing in 20 loc of pure @PyTorch, reducing VRAM by >30% with +1.9% extra overhead. We carefully mask movement betw RAM<=>GPU. No extra dependencies needed.
Try our Colab notebook with Mistral's new long context v2 7b model + our new VRAM savings
You can turn it on with use_gradient_checkpointing = "unsloth"
:
model = FastLanguageModel.get_peft_model(
model,
r = 16,
target_modules = ["q_proj", "k_proj", "v_proj", "o_proj",
"gate_proj", "up_proj", "down_proj",],
lora_alpha = 16,
use_gradient_checkpointing = "unsloth", # True or "unsloth" for very long context
random_state = 3407,
)
Below shows the maximum possible sequence length with Mistral 7b QLoRA rank=32:
GPU | Memory | HF+FA2 | Unsloth | Unsloth New |
---|---|---|---|---|
RTX 4060 | 8 GB | 1,696 | 3,716 | 7,340 |
RTX 4070 | 12 GB | 4,797 | 11,055 | 19,610 |
RTX 4080 | 16 GB | 7,898 | 18,394 | 31,880 |
RTX 4090 | 24 GB | 14,099 | 33,073 | 56,420 |
A100 | 40 GB | 26,502 | 62,431 | 105,500 |
A6000 | 48 GB | 32,704 | 77,110 | 130,040 |
H100 | 80 GB | 57,510 | 135,826 | 228,199 |
We managed to smartly and on the fly convert a slow HF tokenizer to a fast one. We also automatically now load the tokenizer, and fix some dangling incorrect tokens. What can this be useful for?
@HuyNguyen-hust managed to make Unsloth RoPE Embeddings around 28% faster! This primarily is useful for long context windows. Via torch profiler, Unsloth's original kernel made RoPE use up less than 2% of total runtime, so you will see maybe 0.5 to 1% speedups especially for large training runs. Any speedup is vastly welcome! See #238 for more details.
protobuf<4.0.0
. We edited the pyproject.toml
to make it work.As always, Colab and Kaggle do not need updating. On local machines, please use pip install --upgrade --force-reinstall --no-cache-dir git+https://github.com/unslothai/unsloth.git
to update Unsloth with no dependency changes.
You can now finetune Gemma 7b 2.43x faster than HF + Flash Attention 2 with 57.5% less VRAM use. When compared to vanilla HF, Unsloth is 2.53x faster and uses 70% less VRAM. Blog post: https://unsloth.ai/blog/gemma. On local machines, update Unsloth via pip install --upgrade --force-reinstall --no-cache-dir git+https://github.com/unslothai/unsloth.git
On 1x A100 80GB GPU, Unsloth can fit 40K total tokens (8192 * bsz of 5), whilst FA2 can fit ~15K tokens and vanilla HF can fit 9K tokens.
Gemma 7b Colab Notebook free Tesla T4: https://colab.research.google.com/drive/10NbwlsRChbma1v55m8LAPYG15uQv6HLo?usp=sharing
Gemma 2b Colab Notebook free Tesla T4: https://colab.research.google.com/drive/15gGm7x_jTm017_Ic8e317tdIpDG53Mtu?usp=sharing
To use Gemma, simply use FastLanguageModel
:
# Load Llama model
model, tokenizer = FastLanguageModel.from_pretrained(
model_name = "unsloth/gemma-7b-bnb-4bit",
max_seq_length = max_seq_length,
dtype = None,
load_in_4bit = True,
)
Update Unsloth on local machines with no dependency updates with pip install --upgrade --force-reinstall --no-cache-dir git+https://github.com/unslothai/unsloth.git
Read our blog post for more info: https://unsloth.ai/blog/gemma. You can now finetune Gemma 7b 2.43x faster than HF + Flash Attention 2 with 57.5% less VRAM use. When compared to vanilla HF, Unsloth is 2.53x faster and uses 70% less VRAM. On 1x A100 80GB GPU, Unsloth can fit 40K total tokens (8192 * bsz of 5), whilst FA2 can fit ~15K tokens and vanilla HF can fit 9K tokens.
Unsloth supports natively 2x faster inference. All QLoRA, LoRA and non LoRA inference paths are 2x faster. This requires no change of code or any new dependencies.
from unsloth import FastLanguageModel
model, tokenizer = FastLanguageModel.from_pretrained(
model_name = "lora_model", # YOUR MODEL YOU USED FOR TRAINING
max_seq_length = max_seq_length,
dtype = dtype,
load_in_4bit = load_in_4bit,
)
FastLanguageModel.for_inference(model) # Enable native 2x faster inference
text_streamer = TextStreamer(tokenizer)
_ = model.generate(**inputs, streamer = text_streamer, max_new_tokens = 64)
Assuming your dataset is a list of list of dictionaries like the below:
[
[{'from': 'human', 'value': 'Hi there!'},
{'from': 'gpt', 'value': 'Hi how can I help?'},
{'from': 'human', 'value': 'What is 2+2?'}],
[{'from': 'human', 'value': 'What's your name?'},
{'from': 'gpt', 'value': 'I'm Daniel!'},
{'from': 'human', 'value': 'Ok! Nice!'},
{'from': 'gpt', 'value': 'What can I do for you?'},
{'from': 'human', 'value': 'Oh nothing :)'},],
]
You can use our get_chat_template
to format it. Select chat_template
to be any of zephyr, chatml, mistral, llama, alpaca, vicuna, vicuna_old, unsloth
, and use mapping
to map the dictionary values from
, value
etc. map_eos_token
allows you to map <|im_end|>
to EOS without any training.
from unsloth.chat_templates import get_chat_template
tokenizer = get_chat_template(
tokenizer,
chat_template = "chatml", # Supports zephyr, chatml, mistral, llama, alpaca, vicuna, vicuna_old, unsloth
mapping = {"role" : "from", "content" : "value", "user" : "human", "assistant" : "gpt"}, # ShareGPT style
map_eos_token = True, # Maps <|im_end|> to </s> instead
)
def formatting_prompts_func(examples):
convos = examples["conversations"]
texts = [tokenizer.apply_chat_template(convo, tokenize = False, add_generation_prompt = False) for convo in convos]
return { "text" : texts, }
pass
from datasets import load_dataset
dataset = load_dataset("philschmid/guanaco-sharegpt-style", split = "train")
dataset = dataset.map(formatting_prompts_func, batched = True,)
You can also make your own custom chat templates! For example our internal chat template we use is below. You must pass in a tuple
of (custom_template, eos_token)
where the eos_token
must be used inside the template.
unsloth_template = \
"{{ bos_token }}"\
"{{ 'You are a helpful assistant to the user\n' }}"\
"{% endif %}"\
"{% for message in messages %}"\
"{% if message['role'] == 'user' %}"\
"{{ '>>> User: ' + message['content'] + '\n' }}"\
"{% elif message['role'] == 'assistant' %}"\
"{{ '>>> Assistant: ' + message['content'] + eos_token + '\n' }}"\
"{% endif %}"\
"{% endfor %}"\
"{% if add_generation_prompt %}"\
"{{ '>>> Assistant: ' }}"\
"{% endif %}"
unsloth_eos_token = "eos_token"
tokenizer = get_chat_template(
tokenizer,
chat_template = (unsloth_template, unsloth_eos_token,), # You must provide a template and EOS token
mapping = {"role" : "from", "content" : "value", "user" : "human", "assistant" : "gpt"}, # ShareGPT style
map_eos_token = True, # Maps <|im_end|> to </s> instead
)
And many bug fixes!
Upgrade Unsloth via pip install --upgrade --force-reinstall --no-cache-dir git+https://github.com/unslothai/unsloth.git
. No dependency updates will be done.
# To merge to 16bit:
model.save_pretrained_merged("dir", tokenizer, save_method = "merged_16bit")
# To merge to 4bit:
model.save_pretrained_merged("dir", tokenizer, save_method = "merged_4bit")
# To save to GGUF:
model.save_pretrained_gguf("dir", tokenizer, quantization_method = "q4_k_m")
model.save_pretrained_gguf("dir", tokenizer, quantization_method = "q8_0")
model.save_pretrained_gguf("dir", tokenizer, quantization_method = "f16")
# All methods supported (listed below)
To push to HF:
model.push_to_hub_merged("hf_username/dir", tokenizer, save_method = "merged_16bit")
model.push_to_hub_merged("hf_username/dir", tokenizer, save_method = "merged_4bit")
model.push_to_hub_gguf("hf_username/dir", tokenizer, quantization_method = "q4_k_m")
model.push_to_hub_gguf("hf_username/dir", tokenizer, quantization_method = "q8_0")
"unsloth/mistral-7b-bnb-4bit",
"unsloth/llama-2-7b-bnb-4bit",
"unsloth/llama-2-13b-bnb-4bit",
"unsloth/codellama-34b-bnb-4bit",
"unsloth/tinyllama-bnb-4bit",
packing = True
support making training 5x faster via TRL.
GGUF:
Choose for `quantization_method` to be:
"not_quantized" : "Recommended. Fast conversion. Slow inference, big files.",
"fast_quantized" : "Recommended. Fast conversion. OK inference, OK file size.",
"quantized" : "Recommended. Slow conversion. Fast inference, small files.",
"f32" : "Not recommended. Retains 100% accuracy, but super slow and memory hungry.",
"f16" : "Fastest conversion + retains 100% accuracy. Slow and memory hungry.",
"q8_0" : "Fast conversion. High resource use, but generally acceptable.",
"q4_k_m" : "Recommended. Uses Q6_K for half of the attention.wv and feed_forward.w2 tensors, else Q4_K",
"q5_k_m" : "Recommended. Uses Q6_K for half of the attention.wv and feed_forward.w2 tensors, else Q5_K",
"q2_k" : "Uses Q4_K for the attention.vw and feed_forward.w2 tensors, Q2_K for the other tensors.",
"q3_k_l" : "Uses Q5_K for the attention.wv, attention.wo, and feed_forward.w2 tensors, else Q3_K",
"q3_k_m" : "Uses Q4_K for the attention.wv, attention.wo, and feed_forward.w2 tensors, else Q3_K",
"q3_k_s" : "Uses Q3_K for all tensors",
"q4_0" : "Original quant method, 4-bit.",
"q4_1" : "Higher accuracy than q4_0 but not as high as q5_0. However has quicker inference than q5 models.",
"q4_k_s" : "Uses Q4_K for all tensors",
"q5_0" : "Higher accuracy, higher resource usage and slower inference.",
"q5_1" : "Even higher accuracy, resource usage and slower inference.",
"q5_k_s" : "Uses Q5_K for all tensors",
"q6_k" : "Uses Q8_K for all tensors",
Use Mistral as follows:
pip install "unsloth[colab_ampere] @ git+https://github.com/unslothai/unsloth.git"
from unsloth import FastMistralModel
import torch
model, tokenizer = FastMistralModel.from_pretrained(
model_name = model_name,
max_seq_length = max_seq_length,
dtype = dtype,
load_in_4bit = load_in_4bit,
)
model = FastMistralModel.get_peft_model(
model,
r = 16,
target_modules = ["q_proj", "k_proj", "v_proj", "o_proj",
"gate_proj", "up_proj", "down_proj",],
lora_alpha = 16,
lora_dropout = 0, # Currently only supports dropout = 0
bias = "none", # Currently only supports bias = "none"
use_gradient_checkpointing = True,
random_state = 3407,
max_seq_length = max_seq_length,
)
https://unsloth.ai/blog/mistral-benchmark for full benchmarks and more details