unsloth/unsloth/models/vision.py

# Copyright 2023-present Daniel Han-Chen & the Unsloth team. All rights reserved.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
#     http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.

import torch
from transformers import (
    BitsAndBytesConfig,
    AutoModelForVision2Seq,
    AutoProcessor,
    AutoTokenizer,
    AutoModelForCausalLM,
)
from .llama import *
from ..kernels import (
    post_patch_loss_function,
)
from ._utils import __version__
from peft import LoraConfig, TaskType, get_peft_model
from transformers import set_seed as transformers_set_seed
from unsloth_zoo.peft_utils import (
    get_peft_regex,
    SKIP_QUANTIZATION_MODULES,
    requires_grad_for_gradient_checkpointing,
)
from triton import __version__ as triton_version
from unsloth_zoo.utils import _get_dtype
from unsloth_zoo.patching_utils import patch_model_and_tokenizer
import types
import functools

__all__ = [
    "FastBaseModel",
]


def unsloth_base_fast_generate(
    self,
    *args,
    **kwargs,
):
    FastBaseModel.for_inference(self)
    dtype = _get_dtype(self.config.torch_dtype)

    # Check if VLM
    is_vlm = (x.endswith("ForConditionalGeneration") for x in self.config.architectures)
    is_vlm = is_vlm or hasattr(self.config, "vision_config")

    # Remove token_type_ids
    kwargs.pop("token_type_ids", None)

    # VLMs do not allow logits_to_keep
    if not is_vlm: kwargs["logits_to_keep"] = 1

    # Check pad_token
    model_eos_token_id = getattr(self.config, "eos_token_id", None)
    if model_eos_token_id is not None and hasattr(model_eos_token_id, "__iter__"):
        model_eos_token_id = model_eos_token_id[0]

    kwargs["pad_token_id"] = kwargs.pop("pad_token_id", model_eos_token_id)

    # Get pixel values for VLMs
    try: kwargs["pixel_values"] = kwargs["pixel_values"].to(dtype)
    except: pass

    # Mixed precision autocast
    with torch.inference_mode(), torch.autocast(device_type = "cuda", dtype = dtype):
        output = self._old_generate(*args, **kwargs)
    pass

    FastBaseModel.for_training(self)
    return output
pass


class FastBaseModel:

    @staticmethod
    def from_pretrained(
        model_name        = "unsloth/Llama-3.2-1B-Instruct",
        max_seq_length    = None,
        dtype             = None,
        load_in_4bit      = True,
        token             = None,
        device_map        = "sequential",
        trust_remote_code = False,
        model_types       = None,
        tokenizer_name    = None,
        auto_model        = AutoModelForVision2Seq,
        **kwargs,
    ):
        if trust_remote_code:
            print(
                "Unsloth: WARNING `trust_remote_code` is True.\n"\
                "Are you certain you want to do remote code execution?"
            )
        pass
        if token is None: token = get_token()
        SUPPORTS_BFLOAT16 = is_bfloat16_supported()
        gpu_stats = torch.cuda.get_device_properties(0)
        max_memory = round(gpu_stats.total_memory / 1024 / 1024 / 1024, 3)

        from importlib.metadata import version as importlib_version
        try:    vllm_version = f" vLLM: {importlib_version('vllm')}."
        except: vllm_version = ""

        statistics = \
           f"==((====))==  Unsloth {__version__}: Fast {model_types[0].title()} patching. Transformers: {transformers_version}.{vllm_version}\n"\
           f"   {chr(92)}{chr(92)}   /|    {gpu_stats.name}. Num GPUs = {torch.cuda.device_count()}. Max memory: {max_memory} GB. Platform: {platform_system}.\n"\
           f"O^O/ {chr(92)}_/ {chr(92)}    Torch: {torch.__version__}. CUDA: {gpu_stats.major}.{gpu_stats.minor}. CUDA Toolkit: {torch.version.cuda}. Triton: {triton_version}\n"\
           f"{chr(92)}        /    Bfloat16 = {str(SUPPORTS_BFLOAT16).upper()}. FA [Xformers = {xformers_version}. FA2 = {HAS_FLASH_ATTENTION}]\n"\
           f' "-____-"     Free license: http://github.com/unslothai/unsloth'
        print(statistics)

        # Warn about fast transfers
        old_hf_transfer = os.environ.get("HF_HUB_ENABLE_HF_TRANSFER", "0")
        if os.environ.get("HF_HUB_ENABLE_HF_TRANSFER", "0") == "1":
            print("Unsloth: Fast downloading is enabled - ignore downloading bars which are red colored!")
        pass
        # Return old flag
        os.environ["HF_HUB_ENABLE_HF_TRANSFER"] = old_hf_transfer
        os.environ["HF_HUB_ENABLE_HF_TRANSFER"] = "1"

        get_statistics() # For debugging - we use a download counter to see if environments are not breaking 

        if dtype is None:
            dtype = torch.float16 if not SUPPORTS_BFLOAT16 else torch.bfloat16
        elif dtype == torch.bfloat16 and not SUPPORTS_BFLOAT16:
            logger.warning_once("Device does not support bfloat16. Will change to float16.")
            dtype = torch.float16

        assert(dtype == torch.float16 or dtype == torch.bfloat16 or dtype == torch.float32)

        bnb_config = None
        if load_in_4bit:
            bnb_config = BitsAndBytesConfig(
                load_in_4bit              = True,
                bnb_4bit_use_double_quant = True,
                bnb_4bit_quant_type       = "nf4",
                bnb_4bit_compute_dtype    = dtype,
                llm_int8_skip_modules     = SKIP_QUANTIZATION_MODULES,
            )
        pass

        kwargs.pop("attn_implementation", None); # No need since we auto call it

        # Cannot be None, since HF now checks for the config
        if load_in_4bit: kwargs["quantization_config"] = bnb_config

        model = auto_model.from_pretrained(
            model_name,
            device_map              = device_map,
            torch_dtype             = dtype,
            # quantization_config   = bnb_config,
            token                   = token,
            trust_remote_code       = trust_remote_code,
            # attn_implementation   = "sdpa", [TODO] Pixtral for eg fails
            **kwargs,
        )
        # Return old flag
        os.environ["HF_HUB_ENABLE_HF_TRANSFER"] = old_hf_transfer

        # Counteract saved tokenizers
        tokenizer_name = model_name if tokenizer_name is None else tokenizer_name
        auto_processor = AutoProcessor if auto_model is AutoModelForVision2Seq else AutoTokenizer
        tokenizer = auto_processor.from_pretrained(
            tokenizer_name,
            padding_side = "right",
            token        = token,
        )
        # Add padding side as well
        if hasattr(tokenizer, "tokenizer"):
            tokenizer.tokenizer.padding_side = "right"

        model, tokenizer = patch_tokenizer(model, tokenizer)
        model = post_patch_loss_function(model)
        # Fix other stuff like BnB compute data types
        model, tokenizer = patch_model_and_tokenizer(
            model,
            tokenizer,
            downcast_rope = False,
            fix_embeddings = False,
        )

        # Log Unsloth version for future fastpaths for inference
        if hasattr(model, "config"):
            model.config.update({"unsloth_version" : __version__})
        pass
        patch_saving_functions(model, vision = True)
        patch_saving_functions(tokenizer, vision = True)

        # Fix gradient accumulation
        from transformers.trainer import Trainer
        patch_gradient_accumulation_fix(Trainer)

        # Save tokenizer for inference purposes
        tokenizer.padding_side = "left" # Force inference
        tokenizer.tokenizer.padding_side = "left" # Force inference
        m = model
        while hasattr(m, "model"):
            m._saved_temp_tokenizer = tokenizer
            # Also set is_loaded_in_8bit to disable incorrect DDP
            m.is_loaded_in_8bit = True
            m = m.model
        pass
        m._saved_temp_tokenizer = tokenizer
        # Also set is_loaded_in_8bit to disable incorrect DDP
        m.is_loaded_in_8bit = True

        # Patch generate
        if model.generate.__name__ != "unsloth_base_fast_generate":
            model._old_generate = model.generate
            unsloth_base_fast_generate.__doc__ = model._old_generate.__doc__
            model.generate = types.MethodType(unsloth_base_fast_generate, model)
        return model, tokenizer
    pass


    @staticmethod
    def get_peft_model(
        model,
        r                          = 16,
        target_modules             = None,
        lora_alpha                 = 16,
        lora_dropout               = 0,
        bias                       = "none",
        finetune_vision_layers     = True,
        finetune_language_layers   = True,
        finetune_attention_modules = True,
        finetune_mlp_modules       = True,
        layers_to_transform        = None,
        layers_pattern             = None,
        use_gradient_checkpointing = True,
        random_state               = 3407,
        max_seq_length             = 2048, # not used anymore
        use_rslora                 = False,
        modules_to_save            = None,
        init_lora_weights          = True,
        loftq_config               = {},
        temporary_location         = "_unsloth_temporary_saved_buffers",
        **kwargs,
    ):
        transformers_set_seed(random_state)

        if type(r) is not int:
            raise TypeError(f"Unsloth: Rank of {str(r)} must be an integer.")
        if r <= 0:
            raise TypeError(f"Unsloth: Rank of {str(r)} must be larger than 0.")

        if isinstance(model, PeftModelForCausalLM):
            raise RuntimeError("Unsloth: You already added LoRA adapters to your model!")

        if target_modules == "all-linear":
            finetune_vision_layers     = True
            finetune_language_layers   = True
            finetune_attention_modules = True
            finetune_mlp_modules       = True
        pass
        if target_modules is None:
            target_modules = get_peft_regex(
                model,
                finetune_vision_layers     = finetune_vision_layers,
                finetune_language_layers   = finetune_language_layers,
                finetune_attention_modules = finetune_attention_modules,
                finetune_mlp_modules       = finetune_mlp_modules,
            )
        else:
            assert(type(target_modules) in (list, tuple,))
        pass

        # Clear deleted GPU items
        for _ in range(3):
            gc.collect()
            torch.cuda.empty_cache()
        pass

        lora_config = LoraConfig(
            r               = r,
            lora_alpha      = lora_alpha,
            target_modules  = target_modules,
            lora_dropout    = lora_dropout,
            bias            = bias,
            task_type       = TaskType.CAUSAL_LM,
        )
        model = prepare_model_for_kbit_training(
            model,
            use_gradient_checkpointing = use_gradient_checkpointing,
        )
        model = get_peft_model(model, lora_config)
        # Enable gradients on modules which are trainable
        requires_grad_for_gradient_checkpointing(model)

        model = FastBaseModel.patch_peft_model(model, use_gradient_checkpointing)

        # Clear deleted GPU items
        for _ in range(3):
            gc.collect()
            torch.cuda.empty_cache()
        pass
        patch_saving_functions(model, vision = True)

        # Add for_inference and for_training
        model.for_training  = functools.partial(FastBaseModel.for_training,  model)
        model.for_inference = functools.partial(FastBaseModel.for_inference, model)
        return model
    pass


    @staticmethod
    def patch_peft_model(
        model,
        use_gradient_checkpointing = True,
    ):
        if not isinstance(model, PeftModelForCausalLM):
            raise TypeError(
                "Unsloth: Your model needs to call `.get_peft_model` first!"
            )
        pass

        model = prepare_model_for_kbit_training(
            model,
            use_gradient_checkpointing = use_gradient_checkpointing,
            use_reentrant = True,
        )

        from transformers.trainer import Trainer 
        if Trainer._inner_training_loop.__name__ != "_fast_inner_training_loop":
            raise RuntimeError(
                'Unsloth currently does not work on multi GPU setups - sadly we are a 2 brother team so '\
                'enabling it will require much more work, so we have to prioritize. Please understand!\n'\
                'We do have a separate beta version, which you can contact us about!\n'\
                'Thank you for your understanding and we appreciate it immensely!'
            )
        pass
        patch_saving_functions(model, vision = True)

        # Patch tokenizer to pad to the right
        m = model
        while hasattr(m, "model"):
            if hasattr(m, "_saved_temp_tokenizer"):
                m._saved_temp_tokenizer.tokenizer.padding_side = "right"
            pass
            # Also set is_loaded_in_8bit to disable incorrect DDP
            m.is_loaded_in_8bit = True
            m = m.model
        pass
        if hasattr(m, "_saved_temp_tokenizer"):
            m._saved_temp_tokenizer.tokenizer.padding_side = "right"
        pass
        # Also set is_loaded_in_8bit to disable incorrect DDP
        m.is_loaded_in_8bit = True

        # Clear deleted GPU items
        for _ in range(3):
            gc.collect()
            torch.cuda.empty_cache()
        pass
        # Add for_inference and for_training
        model.for_training  = functools.partial(FastBaseModel.for_training,  model)
        model.for_inference = functools.partial(FastBaseModel.for_inference, model)

        # Patch generate
        if model.generate.__name__ != "unsloth_base_fast_generate":
            model._old_generate = model.generate
            unsloth_base_fast_generate.__doc__ = model._old_generate.__doc__
            model.generate = types.MethodType(unsloth_base_fast_generate, model)
        return model
    pass


    @staticmethod
    def for_inference(model):
        if not hasattr(model, "parameters"):
            raise TypeError("Unsloth: I think you're passing a tokenizer, not the model to for_inference!")

        def _for_inference(m):
            if hasattr(m, "gradient_checkpointing"): m.gradient_checkpointing = False
            if hasattr(m, "training"): m.training = False
            # Pad tokenizer to the left
            if hasattr(m, "_saved_temp_tokenizer"): m._saved_temp_tokenizer.padding_side = "left"
            # Set a flag for generation!
            m._flag_for_generation = True
        pass
        m = model
        while hasattr(m, "model"):
            _for_inference(m)
            m = m.model
        _for_inference(m)

        # Also disable training for embeddings for NEFTune
        if hasattr(model, "get_input_embeddings"):
            embeddings = model.get_input_embeddings()
            if hasattr(embeddings, "training"): embeddings.training = False
        pass
        if hasattr(model, "get_output_embeddings"):
            embeddings = model.get_output_embeddings()
            if hasattr(embeddings, "training"): embeddings.training = False
        pass
        return model
    pass


    @staticmethod
    def for_training(model, use_gradient_checkpointing = True):
        if not hasattr(model, "parameters"):
            raise TypeError("Unsloth: I think you're passing a tokenizer, not the model to for_training!")

        # Delete all fast inference loras
        for param in model.parameters():
            if hasattr(param, "_fast_lora"):
                del param._fast_lora
        pass

        def _for_training(m):
            if hasattr(m, "gradient_checkpointing"): m.gradient_checkpointing = use_gradient_checkpointing
            if hasattr(m, "training"): m.training = True
            # Pad tokenizer to the left
            if hasattr(m, "_saved_temp_tokenizer"): m._saved_temp_tokenizer.padding_side = "right"
            # Set a flag for generation!
            if hasattr(m, "_flag_for_generation"): del m._flag_for_generation
        pass
        m = model
        while hasattr(m, "model"):
            _for_training(m)
            m = m.model
        _for_training(m)

        # Also re-enable training for embeddings for NEFTune
        if hasattr(model, "get_input_embeddings"):
            embeddings = model.get_input_embeddings()
            if hasattr(embeddings, "training"): embeddings.training = True
        pass
        if hasattr(model, "get_output_embeddings"):
            embeddings = model.get_output_embeddings()
            if hasattr(embeddings, "training"): embeddings.training = True
        pass
        return model
    pass
pass