liger-kernel-nightly 0.0.1.dev20240819184814__py3-none-any.whl → 0.6.4.dev20251212103629__py3-none-any.whl

liger_kernel/transformers/model/mixtral.py

@@ -0,0 +1,293 @@
+from typing import List
+from typing import Optional
+from typing import Tuple
+from typing import Union
+
+import torch
+
+from torch.nn import CrossEntropyLoss
+from transformers.modeling_outputs import MoeCausalLMOutputWithPast
+from transformers.models.mixtral.modeling_mixtral import load_balancing_loss_func
+from transformers.utils.deprecation import deprecate_kwarg
+
+from liger_kernel.transformers.fused_linear_cross_entropy import LigerFusedLinearCrossEntropyLoss
+from liger_kernel.transformers.model.loss_utils import LigerForCausalLMLoss
+from liger_kernel.transformers.model.loss_utils import unpack_cross_entropy_result
+from liger_kernel.transformers.model.output_classes import LigerMoeCausalLMOutputWithPast
+
+
+def lce_forward_deprecated(
+    self,
+    input_ids: torch.LongTensor = None,
+    attention_mask: Optional[torch.Tensor] = None,
+    position_ids: Optional[torch.LongTensor] = None,
+    past_key_values: Optional[List[torch.FloatTensor]] = None,
+    inputs_embeds: Optional[torch.FloatTensor] = None,
+    labels: Optional[torch.LongTensor] = None,
+    use_cache: Optional[bool] = None,
+    output_attentions: Optional[bool] = None,
+    output_hidden_states: Optional[bool] = None,
+    output_router_logits: Optional[bool] = None,
+    return_dict: Optional[bool] = None,
+    cache_position: Optional[torch.LongTensor] = None,
+) -> Union[Tuple, MoeCausalLMOutputWithPast]:
+    r"""
+    Copy paste Mixtral's forward from transformers v4.44.2 but replace torch cross entropy with liger fused linear cross entropy
+
+
+    Args:
+        labels (`torch.LongTensor` of shape `(batch_size, sequence_length)`, *optional*):
+            Labels for computing the masked language modeling loss. Indices should either be in `[0, ...,
+            config.vocab_size]` or -100 (see `input_ids` docstring). Tokens with indices set to `-100` are ignored
+            (masked), the loss is only computed for the tokens with labels in `[0, ..., config.vocab_size]`.
+
+    Returns:
+
+    Example:
+
+    ```python
+    >>> from transformers import AutoTokenizer, MixtralForCausalLM
+
+    >>> model = MixtralForCausalLM.from_pretrained("mistralai/Mixtral-8x7B-v0.1")
+    >>> tokenizer = AutoTokenizer.from_pretrained("mistralai/Mixtral-8x7B-v0.1")
+
+    >>> prompt = "Hey, are you conscious? Can you talk to me?"
+    >>> inputs = tokenizer(prompt, return_tensors="pt")
+
+    >>> # Generate
+    >>> generate_ids = model.generate(inputs.input_ids, max_length=30)
+    >>> tokenizer.batch_decode(generate_ids, skip_special_tokens=True, clean_up_tokenization_spaces=False)[0]
+    "Hey, are you conscious? Can you talk to me?\nI'm not conscious, but I can talk to you."
+    ```"""
+
+    output_attentions = output_attentions if output_attentions is not None else self.config.output_attentions
+    output_router_logits = (
+        output_router_logits if output_router_logits is not None else self.config.output_router_logits
+    )
+
+    output_hidden_states = (
+        output_hidden_states if output_hidden_states is not None else self.config.output_hidden_states
+    )
+    return_dict = return_dict if return_dict is not None else self.config.use_return_dict
+
+    # decoder outputs consists of (dec_features, layer_state, dec_hidden, dec_attn)
+    outputs = self.model(
+        input_ids=input_ids,
+        attention_mask=attention_mask,
+        position_ids=position_ids,
+        past_key_values=past_key_values,
+        inputs_embeds=inputs_embeds,
+        use_cache=use_cache,
+        output_attentions=output_attentions,
+        output_hidden_states=output_hidden_states,
+        output_router_logits=output_router_logits,
+        return_dict=return_dict,
+        cache_position=cache_position,
+    )
+
+    hidden_states = outputs[0]
+    logits = self.lm_head(hidden_states)
+
+    loss = None
+    if self.training and (labels is not None):
+        shift_hidden_states = hidden_states[..., :-1, :].contiguous()
+        shift_labels = labels[..., 1:].contiguous()
+        # Flatten the tokens
+        shift_hidden_states = shift_hidden_states.view(-1, self.config.hidden_size)
+        shift_labels = shift_labels.view(-1)
+
+        lce = LigerFusedLinearCrossEntropyLoss()
+        loss = lce(self.lm_head.weight, shift_hidden_states, shift_labels)
+    elif labels is not None:
+        # Upcast to float if we need to compute the loss to avoid potential precision issues
+        logits = logits.float()
+        # Shift so that tokens < n predict n
+        shift_logits = logits[..., :-1, :].contiguous()
+        shift_labels = labels[..., 1:].contiguous()
+        # Flatten the tokens
+        shift_logits = shift_logits.view(-1, self.config.vocab_size)
+        shift_labels = shift_labels.view(-1)
+        # Enable model parallelism
+        shift_labels = shift_labels.to(shift_logits.device)
+
+        loss_fct = CrossEntropyLoss()
+        loss = loss_fct(logits.weight, shift_labels)
+
+    aux_loss = None
+    if output_router_logits:
+        aux_loss = load_balancing_loss_func(
+            outputs.router_logits if return_dict else outputs[-1],
+            self.num_experts,
+            self.num_experts_per_tok,
+            attention_mask,
+        )
+        if labels is not None:
+            loss += self.router_aux_loss_coef * aux_loss.to(loss.device)  # make sure to reside in the same device
+
+    if not return_dict:
+        output = (logits,) + outputs[1:]
+        if output_router_logits:
+            output = (aux_loss,) + output
+        return (loss,) + output if loss is not None else output
+
+    return MoeCausalLMOutputWithPast(
+        loss=loss,
+        aux_loss=aux_loss,
+        logits=logits,
+        past_key_values=outputs.past_key_values,
+        hidden_states=outputs.hidden_states,
+        attentions=outputs.attentions,
+        router_logits=outputs.router_logits,
+    )
+
+
+@deprecate_kwarg("num_logits_to_keep", version="4.50", new_name="logits_to_keep")
+# Ignore copy
+def lce_forward(
+    self,
+    input_ids: torch.LongTensor = None,
+    attention_mask: Optional[torch.Tensor] = None,
+    position_ids: Optional[torch.LongTensor] = None,
+    past_key_values: Optional[List[torch.FloatTensor]] = None,
+    inputs_embeds: Optional[torch.FloatTensor] = None,
+    labels: Optional[torch.LongTensor] = None,
+    use_cache: Optional[bool] = None,
+    output_attentions: Optional[bool] = None,
+    output_hidden_states: Optional[bool] = None,
+    output_router_logits: Optional[bool] = None,
+    return_dict: Optional[bool] = None,
+    cache_position: Optional[torch.LongTensor] = None,
+    logits_to_keep: Union[int, torch.Tensor] = 0,
+    skip_logits: Optional[bool] = None,
+    **kwargs,
+) -> Union[Tuple, LigerMoeCausalLMOutputWithPast]:
+    r"""
+    Args:
+        labels (`torch.LongTensor` of shape `(batch_size, sequence_length)`, *optional*):
+            Labels for computing the masked language modeling loss. Indices should either be in `[0, ...,
+            config.vocab_size]` or -100 (see `input_ids` docstring). Tokens with indices set to `-100` are ignored
+            (masked), the loss is only computed for the tokens with labels in `[0, ..., config.vocab_size]`.
+
+        logits_to_keep (`int` or `torch.Tensor`, *optional*):
+            If an `int`, compute logits for the last `logits_to_keep` tokens. If `0`, calculate logits for all
+            `input_ids` (special case). Only last token logits are needed for generation, and calculating them only for that
+            token can save memory, which becomes pretty significant for long sequences or large vocabulary size.
+            If a `torch.Tensor`, must be 1D corresponding to the indices to keep in the sequence length dimension.
+            This is useful when using packed tensor format (single dimension for batch and sequence length).
+
+    Returns:
+
+    Example:
+
+    ```python
+    >>> from transformers import AutoTokenizer, MixtralForCausalLM
+
+    >>> model = MixtralForCausalLM.from_pretrained("mistralai/Mixtral-8x7B-v0.1")
+    >>> tokenizer = AutoTokenizer.from_pretrained("mistralai/Mixtral-8x7B-v0.1")
+
+    >>> prompt = "Hey, are you conscious? Can you talk to me?"
+    >>> inputs = tokenizer(prompt, return_tensors="pt")
+
+    >>> # Generate
+    >>> generate_ids = model.generate(inputs.input_ids, max_length=30)
+    >>> tokenizer.batch_decode(generate_ids, skip_special_tokens=True, clean_up_tokenization_spaces=False)[0]
+    "Hey, are you conscious? Can you talk to me?\nI'm not conscious, but I can talk to you."
+    ```"""
+
+    output_attentions = output_attentions if output_attentions is not None else self.config.output_attentions
+    output_router_logits = (
+        output_router_logits if output_router_logits is not None else self.config.output_router_logits
+    )
+
+    output_hidden_states = (
+        output_hidden_states if output_hidden_states is not None else self.config.output_hidden_states
+    )
+    return_dict = return_dict if return_dict is not None else self.config.use_return_dict
+
+    # decoder outputs consists of (dec_features, layer_state, dec_hidden, dec_attn)
+    outputs = self.model(
+        input_ids=input_ids,
+        attention_mask=attention_mask,
+        position_ids=position_ids,
+        past_key_values=past_key_values,
+        inputs_embeds=inputs_embeds,
+        use_cache=use_cache,
+        output_attentions=output_attentions,
+        output_hidden_states=output_hidden_states,
+        output_router_logits=output_router_logits,
+        return_dict=return_dict,
+        cache_position=cache_position,
+        **kwargs,
+    )
+
+    hidden_states = outputs[0]
+    # Only compute necessary logits, and do not upcast them to float if we are not computing the loss
+    slice_indices = slice(-logits_to_keep, None) if isinstance(logits_to_keep, int) else logits_to_keep
+    kept_hidden_states = hidden_states[:, slice_indices, :]
+
+    shift_labels = kwargs.pop("shift_labels", None)
+    logits = None
+    loss = None
+    token_accuracy = None
+
+    if skip_logits and labels is None and shift_labels is None:
+        raise ValueError("skip_logits is True, but labels and shift_labels are None")
+
+    if skip_logits is None:
+        # By default, if in training mode, don't materialize logits
+        skip_logits = self.training and (labels is not None or shift_labels is not None)
+
+    # Compute loss
+    if skip_logits:
+        result = LigerForCausalLMLoss(
+            hidden_states=kept_hidden_states,
+            lm_head_weight=self.lm_head.weight,
+            labels=labels,
+            shift_labels=shift_labels,
+            hidden_size=self.config.hidden_size,
+            **kwargs,
+        )
+        loss, _, token_accuracy = unpack_cross_entropy_result(result)
+
+    else:
+        logits = self.lm_head(kept_hidden_states)
+
+        loss = None
+        if labels is not None or shift_labels is not None:
+            loss = self.loss_function(
+                logits=logits,
+                labels=labels,
+                shift_labels=shift_labels,
+                vocab_size=self.vocab_size,
+                **kwargs,
+            )
+    aux_loss = None
+    if output_router_logits:
+        aux_loss = load_balancing_loss_func(
+            outputs.router_logits if return_dict else outputs[-1],
+            self.num_experts,
+            self.num_experts_per_tok,
+            attention_mask,
+        )
+        if labels is not None:
+            loss += self.router_aux_loss_coef * aux_loss.to(loss.device)  # make sure to reside in the same device
+
+    if not return_dict:
+        output_tuple = (logits,) + outputs[1:]
+        if output_router_logits:
+            output_tuple = (aux_loss,) + output_tuple
+        if token_accuracy is not None:
+            output_tuple = output_tuple + (token_accuracy,)
+        return (loss,) + output_tuple if loss is not None else output_tuple
+
+    # Return custom output class with token_accuracy field
+    return LigerMoeCausalLMOutputWithPast(
+        loss=loss,
+        aux_loss=aux_loss,
+        logits=logits,
+        past_key_values=outputs.past_key_values,
+        hidden_states=outputs.hidden_states,
+        attentions=outputs.attentions,
+        router_logits=outputs.router_logits if return_dict else outputs[-1],
+        token_accuracy=token_accuracy,
+    )

liger_kernel/transformers/model/mllama.py

@@ -0,0 +1,269 @@
+from typing import List
+from typing import Optional
+from typing import Tuple
+from typing import Union
+
+import torch
+
+from torch.nn import CrossEntropyLoss
+from transformers.cache_utils import Cache
+from transformers.modeling_outputs import CausalLMOutputWithPast
+from transformers.utils.deprecation import deprecate_kwarg
+
+from liger_kernel.transformers.fused_linear_cross_entropy import LigerFusedLinearCrossEntropyLoss
+from liger_kernel.transformers.model.loss_utils import LigerForCausalLMLoss
+from liger_kernel.transformers.model.loss_utils import unpack_cross_entropy_result
+from liger_kernel.transformers.model.output_classes import LigerCausalLMOutputWithPast
+
+
+def lce_forward_deprecated(
+    self,
+    input_ids: torch.LongTensor = None,
+    attention_mask: Optional[torch.Tensor] = None,
+    position_ids: Optional[torch.LongTensor] = None,
+    cross_attention_states: Optional[torch.LongTensor] = None,
+    cross_attention_mask: Optional[torch.LongTensor] = None,
+    full_text_row_masked_out_mask: Optional[Tuple[torch.Tensor, torch.Tensor]] = None,
+    past_key_values: Optional[Union[Cache, List[torch.FloatTensor]]] = None,
+    inputs_embeds: Optional[torch.FloatTensor] = None,
+    labels: Optional[torch.LongTensor] = None,
+    use_cache: Optional[bool] = None,
+    output_attentions: Optional[bool] = None,
+    output_hidden_states: Optional[bool] = None,
+    return_dict: Optional[bool] = None,
+    cache_position: Optional[torch.LongTensor] = None,
+    num_logits_to_keep: int = 0,
+) -> Union[Tuple, CausalLMOutputWithPast]:
+    r"""
+    Copy paste mllama forward but replace torch cross entropy with liger fused linear cross entropy
+
+
+    Args:
+        labels (`torch.LongTensor` of shape `(batch_size, sequence_length)`, *optional*):
+            Labels for computing the masked language modeling loss. Indices should either be in `[0, ...,
+            config.vocab_size]` or -100 (see `input_ids` docstring). Tokens with indices set to `-100` are ignored
+            (masked), the loss is only computed for the tokens with labels in `[0, ..., config.vocab_size]`.
+        num_logits_to_keep (`int`, *optional*):
+            Calculate logits for the last `num_logits_to_keep` tokens. If `0`, calculate logits for all
+            `input_ids` (special case). Only last token logits are needed for generation, and calculating them only for that
+            token can save memory, which becomes pretty significant for long sequences or large vocabulary size.
+    Returns:
+    Example:
+    ```python
+    >>> from transformers import AutoTokenizer, MllamaForCausalLM
+    >>> model = MllamaForCausalLM.from_pretrained("Llama-3.2-11B-Vision")
+    >>> tokenizer = AutoTokenizer.from_pretrained("Llama-3.2-11B-Vision")
+    >>> prompt = "If I had to write a haiku, it would be:"
+    >>> inputs = tokenizer(prompt, return_tensors="pt")
+    >>> # Generate
+    >>> generate_ids = model.generate(inputs.input_ids, max_length=40, do_sample=True, temperature=0.6)
+    >>> result = tokenizer.batch_decode(generate_ids, skip_special_tokens=True, clean_up_tokenization_spaces=False)[0]
+    >>> print(result)
+    If I had to write a haiku, it would be: "Snowflakes gently fall" - simple, yet peaceful.
+    I love the idea of snowflakes gently falling, each one
+    ```
+    """
+    output_attentions = output_attentions if output_attentions is not None else self.config.output_attentions
+    output_hidden_states = (
+        output_hidden_states if output_hidden_states is not None else self.config.output_hidden_states
+    )
+    return_dict = return_dict if return_dict is not None else self.config.use_return_dict
+
+    # decoder outputs consists of (dec_features, layer_state, dec_hidden, dec_attn)
+    outputs = self.model(
+        input_ids=input_ids,
+        cross_attention_states=cross_attention_states,
+        attention_mask=attention_mask,
+        position_ids=position_ids,
+        cross_attention_mask=cross_attention_mask,
+        full_text_row_masked_out_mask=full_text_row_masked_out_mask,
+        past_key_values=past_key_values,
+        inputs_embeds=inputs_embeds,
+        use_cache=use_cache,
+        output_attentions=output_attentions,
+        output_hidden_states=output_hidden_states,
+        return_dict=return_dict,
+        cache_position=cache_position,
+    )
+
+    hidden_states = outputs[0]
+
+    loss = None
+    logits = None
+
+    if self.training and (labels is not None):
+        kept_hidden_states = hidden_states[:, -num_logits_to_keep:, :]
+
+        shift_hidden_states = kept_hidden_states[..., :-1, :].contiguous()
+        shift_labels = labels[..., 1:].contiguous()
+
+        # flatten tokens
+        shift_hidden_states = shift_hidden_states.view(-1, self.config.hidden_size)
+        shift_labels = shift_labels.view(-1)
+
+        lce = LigerFusedLinearCrossEntropyLoss()
+        loss = lce(self.lm_head.weight, shift_hidden_states, shift_labels)
+
+    else:
+        logits = self.lm_head(hidden_states[:, -num_logits_to_keep:, :]).float()
+        if labels is not None:
+            # Shift so that tokens < n predict n
+            shift_logits = logits[..., :-1, :].contiguous()
+            shift_labels = labels[..., 1:].contiguous()
+            # Flatten the tokens
+            loss_fct = CrossEntropyLoss()
+            shift_logits = shift_logits.view(-1, self.config.vocab_size)
+            shift_labels = shift_labels.view(-1)
+            # Enable model parallelism
+            shift_labels = shift_labels.to(shift_logits.device)
+            loss = loss_fct(shift_logits, shift_labels)
+
+    if not return_dict:
+        output = (logits,) + outputs[1:]
+        return (loss,) + output if loss is not None else output
+
+    return CausalLMOutputWithPast(
+        loss=loss,
+        logits=logits,
+        past_key_values=outputs.past_key_values,
+        hidden_states=outputs.hidden_states,
+        attentions=outputs.attentions,
+    )
+
+
+@deprecate_kwarg("num_logits_to_keep", version="4.50", new_name="logits_to_keep")
+def lce_forward(
+    self,
+    input_ids: torch.LongTensor = None,
+    attention_mask: Optional[torch.Tensor] = None,
+    position_ids: Optional[torch.LongTensor] = None,
+    cross_attention_states: Optional[torch.LongTensor] = None,
+    cross_attention_mask: Optional[torch.LongTensor] = None,
+    full_text_row_masked_out_mask: Optional[Tuple[torch.Tensor, torch.Tensor]] = None,
+    past_key_values: Optional[Union[Cache, List[torch.FloatTensor]]] = None,
+    inputs_embeds: Optional[torch.FloatTensor] = None,
+    labels: Optional[torch.LongTensor] = None,
+    use_cache: Optional[bool] = None,
+    output_attentions: Optional[bool] = None,
+    output_hidden_states: Optional[bool] = None,
+    return_dict: Optional[bool] = None,
+    cache_position: Optional[torch.LongTensor] = None,
+    logits_to_keep: Union[int, torch.Tensor] = 0,
+    skip_logits: Optional[bool] = None,
+    **kwargs,
+) -> Union[Tuple, LigerCausalLMOutputWithPast]:
+    r"""
+    Args:
+        labels (`torch.LongTensor` of shape `(batch_size, sequence_length)`, *optional*):
+            Labels for computing the masked language modeling loss. Indices should either be in `[0, ...,
+            config.vocab_size]` or -100 (see `input_ids` docstring). Tokens with indices set to `-100` are ignored
+            (masked), the loss is only computed for the tokens with labels in `[0, ..., config.vocab_size]`.
+
+        logits_to_keep (`int` or `torch.Tensor`, *optional*):
+            If an `int`, compute logits for the last `logits_to_keep` tokens. If `0`, calculate logits for all
+            `input_ids` (special case). Only last token logits are needed for generation, and calculating them only for that
+            token can save memory, which becomes pretty significant for long sequences or large vocabulary size.
+            If a `torch.Tensor`, must be 1D corresponding to the indices to keep in the sequence length dimension.
+            This is useful when using packed tensor format (single dimension for batch and sequence length).
+
+    Returns:
+
+    Example:
+
+    ```python
+    >>> from transformers import AutoTokenizer, MllamaForCausalLM
+
+    >>> model = MllamaForCausalLM.from_pretrained("Llama-3.2-11B-Vision")
+    >>> tokenizer = AutoTokenizer.from_pretrained("Llama-3.2-11B-Vision")
+
+    >>> prompt = "If I had to write a haiku, it would be:"
+    >>> inputs = tokenizer(prompt, return_tensors="pt")
+
+    >>> # Generate
+    >>> generate_ids = model.generate(inputs.input_ids, max_length=40, do_sample=True, temperature=0.6)
+    >>> result = tokenizer.batch_decode(generate_ids, skip_special_tokens=True, clean_up_tokenization_spaces=False)[0]
+    >>> print(result)
+    If I had to write a haiku, it would be: "Snowflakes gently fall" - simple, yet peaceful.
+    I love the idea of snowflakes gently falling, each one
+    ```
+    """
+    output_attentions = output_attentions if output_attentions is not None else self.config.output_attentions
+    output_hidden_states = (
+        output_hidden_states if output_hidden_states is not None else self.config.output_hidden_states
+    )
+    return_dict = return_dict if return_dict is not None else self.config.use_return_dict
+    # Filter out accum_dtype from kwargs for model call as MllamaTextModel doesn't accept it in transformers 4.49.0
+    # but preserve it for loss function calls
+    model_kwargs = {k: v for k, v in kwargs.items() if k != "accum_dtype"}
+    # decoder outputs consists of (dec_features, layer_state, dec_hidden, dec_attn)
+    outputs = self.model(
+        input_ids=input_ids,
+        cross_attention_states=cross_attention_states,
+        attention_mask=attention_mask,
+        position_ids=position_ids,
+        cross_attention_mask=cross_attention_mask,
+        full_text_row_masked_out_mask=full_text_row_masked_out_mask,
+        past_key_values=past_key_values,
+        inputs_embeds=inputs_embeds,
+        use_cache=use_cache,
+        output_attentions=output_attentions,
+        output_hidden_states=output_hidden_states,
+        return_dict=return_dict,
+        cache_position=cache_position,
+        **model_kwargs,
+    )
+
+    hidden_states = outputs[0]
+    # Only compute necessary logits, and do not upcast them to float if we are not computing the loss
+    slice_indices = slice(-logits_to_keep, None) if isinstance(logits_to_keep, int) else logits_to_keep
+    kept_hidden_states = hidden_states[:, slice_indices, :]
+
+    shift_labels = kwargs.pop("shift_labels", None)
+    logits = None
+    loss = None
+    token_accuracy = None
+
+    if skip_logits and labels is None and shift_labels is None:
+        raise ValueError("skip_logits is True, but labels and shift_labels are None")
+
+    if skip_logits is None:
+        # By default, if in training mode, don't materialize logits
+        skip_logits = self.training and (labels is not None or shift_labels is not None)
+
+    if skip_logits:
+        result = LigerForCausalLMLoss(
+            hidden_states=kept_hidden_states,
+            lm_head_weight=self.lm_head.weight,
+            labels=labels,
+            shift_labels=shift_labels,
+            hidden_size=self.config.hidden_size,
+            **kwargs,
+        )
+        loss, _, token_accuracy = unpack_cross_entropy_result(result)
+
+    else:
+        logits = self.lm_head(kept_hidden_states)
+        if labels is not None or shift_labels is not None:
+            loss = self.loss_function(
+                logits=logits,
+                labels=labels,
+                shift_labels=shift_labels,
+                vocab_size=self.config.vocab_size,
+                **kwargs,
+            )
+
+    if not return_dict:
+        output = (logits,) + outputs[1:]
+        output = (loss,) + output if loss is not None else output
+        output = output + (token_accuracy,) if token_accuracy is not None else output
+        return output
+
+    # Return custom output class with token_accuracy field
+    return LigerCausalLMOutputWithPast(
+        loss=loss,
+        logits=logits,
+        past_key_values=outputs.past_key_values,
+        hidden_states=outputs.hidden_states,
+        attentions=outputs.attentions,
+        token_accuracy=token_accuracy,
+    )