Idefics2.

server/text_generation_server/models/custom_modeling/idefics2_modeling.py

@@ -0,0 +1,457 @@
+import torch
+import torch.nn.functional as F
+import torch.utils.checkpoint
+from torch import nn
+from typing import Optional, List, Dict, Tuple
+
+
+class Idefics2VisionFlashAttention2(nn.Module):
+    """
+    Idefics2 flash attention module. This module inherits from `Idefics2VisionAttention` as the weights of the module stays
+    untouched. The only required change would be on the forward pass where it needs to correctly call the public API of
+    flash attention and deal with padding tokens in case the input contains any of them.
+    """
+
+    def __init__(self, config):
+        super().__init__()
+        self.config = config
+        self.embed_dim = config.hidden_size
+        self.num_heads = config.num_attention_heads
+        self.head_dim = self.embed_dim // self.num_heads
+        if self.head_dim * self.num_heads != self.embed_dim:
+            raise ValueError(
+                f"embed_dim must be divisible by num_heads (got `embed_dim`: {self.embed_dim} and `num_heads`:"
+                f" {self.num_heads})."
+            )
+        self.scale = self.head_dim**-0.5
+        self.dropout = config.attention_dropout
+
+        self.k_proj = nn.Linear(self.embed_dim, self.embed_dim)
+        self.v_proj = nn.Linear(self.embed_dim, self.embed_dim)
+        self.q_proj = nn.Linear(self.embed_dim, self.embed_dim)
+        self.out_proj = nn.Linear(self.embed_dim, self.embed_dim)
+
+    def __init__(self, *args, **kwargs):
+        super().__init__(*args, **kwargs)
+        self.is_causal = False  # Hack to make sure we don't use a causal mask
+
+        # TODO: Should be removed once Flash Attention for RoCm is bumped to 2.1.
+        # flash_attn<2.1 generates top-left aligned causal mask, while what is needed here is bottom-right alignement, that was made default for flash_attn>=2.1. This attribute is used to handle this difference. Reference: https://github.com/Dao-AILab/flash-attention/releases/tag/v2.1.0.
+        # Beware that with flash_attn<2.1, using q_seqlen != k_seqlen (except for the case q_seqlen == 1) produces a wrong mask (top-left).
+        self._flash_attn_uses_top_left_mask = not is_flash_attn_greater_or_equal_2_10()
+
+    def forward(
+        self,
+        hidden_states: torch.Tensor,
+        attention_mask: Optional[torch.LongTensor] = None,
+        position_ids: Optional[torch.LongTensor] = None,
+        past_key_value: Optional[Tuple[torch.Tensor]] = None,
+        output_attentions: bool = False,
+        use_cache: bool = False,
+        **kwargs,
+    ) -> Tuple[torch.Tensor, Optional[torch.Tensor], Optional[Tuple[torch.Tensor]]]:
+        output_attentions = False
+
+        bsz, q_len, _ = hidden_states.size()
+
+        query_states = self.q_proj(hidden_states)
+        key_states = self.k_proj(hidden_states)
+        value_states = self.v_proj(hidden_states)
+
+        # Flash attention requires the input to have the shape
+        # batch_size x seq_length x head_dim x hidden_dim
+        # therefore we just need to keep the original shape
+        query_states = query_states.view(
+            bsz, q_len, self.num_heads, self.head_dim
+        ).transpose(1, 2)
+        key_states = key_states.view(
+            bsz, q_len, self.num_heads, self.head_dim
+        ).transpose(1, 2)
+        value_states = value_states.view(
+            bsz, q_len, self.num_heads, self.head_dim
+        ).transpose(1, 2)
+
+        kv_seq_len = key_states.shape[-2]
+        if past_key_value is not None:
+            kv_seq_len += past_key_value.get_usable_length(kv_seq_len, self.layer_idx)
+
+        # TODO: These transpose are quite inefficient but Flash Attention requires the layout [batch_size, sequence_length, num_heads, head_dim]. We would need to refactor the KV cache
+        # to be able to avoid many of these transpose/reshape/view.
+        query_states = query_states.transpose(1, 2)
+        key_states = key_states.transpose(1, 2)
+        value_states = value_states.transpose(1, 2)
+
+        dropout_rate = self.dropout if self.training else 0.0
+
+        # In PEFT, usually we cast the layer norms in float32 for training stability reasons
+        # therefore the input hidden states gets silently casted in float32. Hence, we need
+        # cast them back in the correct dtype just to be sure everything works as expected.
+        # This might slowdown training & inference so it is recommended to not cast the LayerNorms
+        # in fp32. (Idefics2RMSNorm handles it correctly)
+
+        input_dtype = query_states.dtype
+        if input_dtype == torch.float32:
+            if torch.is_autocast_enabled():
+                target_dtype = torch.get_autocast_gpu_dtype()
+            # Handle the case where the model is quantized
+            elif hasattr(self.config, "_pre_quantization_dtype"):
+                target_dtype = self.config._pre_quantization_dtype
+            else:
+                target_dtype = self.q_proj.weight.dtype
+
+            logger.warning_once(
+                "The input hidden states seems to be silently casted in float32, this might be related to the fact"
+                " you have upcasted embedding or layer norm layers in float32. We will cast back the input in"
+                f" {target_dtype}."
+            )
+
+            query_states = query_states.to(target_dtype)
+            key_states = key_states.to(target_dtype)
+            value_states = value_states.to(target_dtype)
+
+        attn_output = self._flash_attention_forward(
+            query_states,
+            key_states,
+            value_states,
+            attention_mask,
+            q_len,
+            dropout=dropout_rate,
+        )
+
+        attn_output = attn_output.reshape(bsz, q_len, self.embed_dim).contiguous()
+        attn_output = self.out_proj(attn_output)
+
+        if not output_attentions:
+            attn_weights = None
+
+        return attn_output, attn_weights
+
+    # Copied from transformers.models.llama.modeling_llama.LlamaFlashAttention2._flash_attention_forward
+    def _flash_attention_forward(
+        self,
+        query_states,
+        key_states,
+        value_states,
+        attention_mask,
+        query_length,
+        dropout=0.0,
+        softmax_scale=None,
+    ):
+        """
+        Calls the forward method of Flash Attention - if the input hidden states contain at least one padding token
+        first unpad the input, then computes the attention scores and pad the final attention scores.
+
+        Args:
+            query_states (`torch.Tensor`):
+                Input query states to be passed to Flash Attention API
+            key_states (`torch.Tensor`):
+                Input key states to be passed to Flash Attention API
+            value_states (`torch.Tensor`):
+                Input value states to be passed to Flash Attention API
+            attention_mask (`torch.Tensor`):
+                The padding mask - corresponds to a tensor of size `(batch_size, seq_len)` where 0 stands for the
+                position of padding tokens and 1 for the position of non-padding tokens.
+            dropout (`int`, *optional*):
+                Attention dropout
+            softmax_scale (`float`, *optional*):
+                The scaling of QK^T before applying softmax. Default to 1 / sqrt(head_dim)
+        """
+        if not self._flash_attn_uses_top_left_mask:
+            causal = self.is_causal
+        else:
+            # TODO: Remove the `query_length != 1` check once Flash Attention for RoCm is bumped to 2.1. For details, please see the comment in LlamaFlashAttention2 __init__.
+            causal = self.is_causal and query_length != 1
+
+        # Contains at least one padding token in the sequence
+        if attention_mask is not None:
+            batch_size = query_states.shape[0]
+            (
+                query_states,
+                key_states,
+                value_states,
+                indices_q,
+                cu_seq_lens,
+                max_seq_lens,
+            ) = self._upad_input(
+                query_states, key_states, value_states, attention_mask, query_length
+            )
+
+            cu_seqlens_q, cu_seqlens_k = cu_seq_lens
+            max_seqlen_in_batch_q, max_seqlen_in_batch_k = max_seq_lens
+
+            attn_output_unpad = flash_attn_varlen_func(
+                query_states,
+                key_states,
+                value_states,
+                cu_seqlens_q=cu_seqlens_q,
+                cu_seqlens_k=cu_seqlens_k,
+                max_seqlen_q=max_seqlen_in_batch_q,
+                max_seqlen_k=max_seqlen_in_batch_k,
+                dropout_p=dropout,
+                softmax_scale=softmax_scale,
+                causal=causal,
+            )
+
+            attn_output = pad_input(
+                attn_output_unpad, indices_q, batch_size, query_length
+            )
+        else:
+            attn_output = flash_attn_func(
+                query_states,
+                key_states,
+                value_states,
+                dropout,
+                softmax_scale=softmax_scale,
+                causal=causal,
+            )
+
+        return attn_output
+
+    # Copied from transformers.models.llama.modeling_llama.LlamaFlashAttention2._upad_input
+    def _upad_input(
+        self, query_layer, key_layer, value_layer, attention_mask, query_length
+    ):
+        indices_k, cu_seqlens_k, max_seqlen_in_batch_k = _get_unpad_data(attention_mask)
+        batch_size, kv_seq_len, num_key_value_heads, head_dim = key_layer.shape
+
+        key_layer = index_first_axis(
+            key_layer.reshape(batch_size * kv_seq_len, num_key_value_heads, head_dim),
+            indices_k,
+        )
+        value_layer = index_first_axis(
+            value_layer.reshape(batch_size * kv_seq_len, num_key_value_heads, head_dim),
+            indices_k,
+        )
+        if query_length == kv_seq_len:
+            query_layer = index_first_axis(
+                query_layer.reshape(batch_size * kv_seq_len, self.num_heads, head_dim),
+                indices_k,
+            )
+            cu_seqlens_q = cu_seqlens_k
+            max_seqlen_in_batch_q = max_seqlen_in_batch_k
+            indices_q = indices_k
+        elif query_length == 1:
+            max_seqlen_in_batch_q = 1
+            cu_seqlens_q = torch.arange(
+                batch_size + 1, dtype=torch.int32, device=query_layer.device
+            )  # There is a memcpy here, that is very bad.
+            indices_q = cu_seqlens_q[:-1]
+            query_layer = query_layer.squeeze(1)
+        else:
+            # The -q_len: slice assumes left padding.
+            attention_mask = attention_mask[:, -query_length:]
+            query_layer, indices_q, cu_seqlens_q, max_seqlen_in_batch_q = unpad_input(
+                query_layer, attention_mask
+            )
+
+        return (
+            query_layer,
+            key_layer,
+            value_layer,
+            indices_q,
+            (cu_seqlens_q, cu_seqlens_k),
+            (max_seqlen_in_batch_q, max_seqlen_in_batch_k),
+        )
+
+
+IDEFICS_VISION_ATTENTION_CLASSES = {
+    "eager": Idefics2VisionAttention,
+    "flash_attention_2": Idefics2VisionFlashAttention2,
+}
+
+
+# Copied from transformers.models.siglip.modeling_siglip.SiglipMLP with Siglip->Idefics2Vision
+class Idefics2VisionMLP(nn.Module):
+    def __init__(self, config):
+        super().__init__()
+        self.config = config
+        self.activation_fn = ACT2FN[config.hidden_act]
+        self.fc1 = nn.Linear(config.hidden_size, config.intermediate_size)
+        self.fc2 = nn.Linear(config.intermediate_size, config.hidden_size)
+
+    def forward(self, hidden_states: torch.Tensor) -> torch.Tensor:
+        hidden_states = self.fc1(hidden_states)
+        hidden_states = self.activation_fn(hidden_states)
+        hidden_states = self.fc2(hidden_states)
+        return hidden_states
+
+
+# Copied from transformers.models.siglip.modeling_siglip.SiglipMultiheadAttentionPoolingHead with Siglip->Idefics2
+class Idefics2MultiheadAttentionPoolingHead(nn.Module):
+    """Multihead Attention Pooling."""
+
+    def __init__(self, config: Idefics2VisionConfig):
+        super().__init__()
+
+        self.probe = nn.Parameter(torch.randn(1, 1, config.hidden_size))
+        self.attention = torch.nn.MultiheadAttention(
+            config.hidden_size, config.num_attention_heads, batch_first=True
+        )
+        self.layernorm = nn.LayerNorm(config.hidden_size, eps=config.layer_norm_eps)
+        self.mlp = Idefics2MLP(config)
+
+    def forward(self, hidden_state):
+        batch_size = hidden_state.shape[0]
+        probe = self.probe.repeat(batch_size, 1, 1)
+
+        hidden_state = self.attention(probe, hidden_state, hidden_state)[0]
+
+        residual = hidden_state
+        hidden_state = self.layernorm(hidden_state)
+        hidden_state = residual + self.mlp(hidden_state)
+
+        return hidden_state[:, 0]
+
+
+class Idefics2EncoderLayer(nn.Module):
+    def __init__(self, config: Idefics2Config):
+        super().__init__()
+        self.embed_dim = config.hidden_size
+        self.self_attn = Idefics2VisionFlashAttention2(config)
+        self.layer_norm1 = nn.LayerNorm(self.embed_dim, eps=config.layer_norm_eps)
+        self.mlp = Idefics2VisionMLP(config)
+        self.layer_norm2 = nn.LayerNorm(self.embed_dim, eps=config.layer_norm_eps)
+
+    # Copied from transformers.models.siglip.modeling_siglip.SiglipEncoderLayer.forward
+    def forward(
+        self,
+        hidden_states: torch.Tensor,
+        attention_mask: torch.Tensor,
+        output_attentions: Optional[bool] = False,
+    ) -> Tuple[torch.FloatTensor]:
+        """
+        Args:
+            hidden_states (`torch.FloatTensor`):
+                Input to the layer of shape `(batch, seq_len, embed_dim)`.
+            attention_mask (`torch.FloatTensor`):
+                Attention mask of shape `(batch, 1, q_len, k_v_seq_len)` where padding elements are indicated by very large negative values.
+            output_attentions (`bool`, *optional*, defaults to `False`):
+                Whether or not to return the attentions tensors of all attention layers. See `attentions` under
+                returned tensors for more detail.
+        """
+        residual = hidden_states
+
+        hidden_states = self.layer_norm1(hidden_states)
+        hidden_states, attn_weights = self.self_attn(
+            hidden_states=hidden_states,
+            attention_mask=attention_mask,
+            output_attentions=output_attentions,
+        )
+        hidden_states = residual + hidden_states
+
+        residual = hidden_states
+        hidden_states = self.layer_norm2(hidden_states)
+        hidden_states = self.mlp(hidden_states)
+        hidden_states = residual + hidden_states
+
+        outputs = (hidden_states,)
+
+        if output_attentions:
+            outputs += (attn_weights,)
+
+        return outputs
+
+
+# Copied from transformers.models.siglip.modeling_siglip.SiglipEncoder with Siglip->Idefics2
+class Idefics2Encoder(nn.Module):
+    """
+    Transformer encoder consisting of `config.num_hidden_layers` self attention layers. Each layer is a
+    [`Idefics2EncoderLayer`].
+
+    Args:
+        config: Idefics2Config
+    """
+
+    def __init__(self, config: Idefics2Config):
+        super().__init__()
+        self.config = config
+        self.layers = nn.ModuleList(
+            [Idefics2EncoderLayer(config) for _ in range(config.num_hidden_layers)]
+        )
+        self.gradient_checkpointing = False
+
+    # Ignore copy
+    def forward(
+        self,
+        inputs_embeds,
+        attention_mask: Optional[torch.Tensor] = None,
+        output_attentions: Optional[bool] = None,
+        output_hidden_states: Optional[bool] = None,
+        return_dict: Optional[bool] = None,
+    ) -> Union[Tuple, BaseModelOutput]:
+        r"""
+        Args:
+            inputs_embeds (`torch.FloatTensor` of shape `(batch_size, sequence_length, hidden_size)`):
+                Optionally, instead of passing `input_ids` you can choose to directly pass an embedded representation.
+                This is useful if you want more control over how to convert `input_ids` indices into associated vectors
+                than the model's internal embedding lookup matrix.
+            attention_mask (`torch.Tensor` of shape `(batch_size, sequence_length)`, *optional*):
+                Mask to avoid performing attention on padding token indices. Mask values selected in `[0, 1]`:
+
+                - 1 for tokens that are **not masked**,
+                - 0 for tokens that are **masked**.
+
+                [What are attention masks?](../glossary#attention-mask)
+            output_attentions (`bool`, *optional*):
+                Whether or not to return the attentions tensors of all attention layers. See `attentions` under
+                returned tensors for more detail.
+            output_hidden_states (`bool`, *optional*):
+                Whether or not to return the hidden states of all layers. See `hidden_states` under returned tensors
+                for more detail.
+            return_dict (`bool`, *optional*):
+                Whether or not to return a [`~utils.ModelOutput`] instead of a plain tuple.
+        """
+        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
+        )
+
+        encoder_states = () if output_hidden_states else None
+        all_attentions = () if output_attentions else None
+
+        hidden_states = inputs_embeds
+        for encoder_layer in self.layers:
+            if output_hidden_states:
+                encoder_states = encoder_states + (hidden_states,)
+            if self.gradient_checkpointing and self.training:
+                layer_outputs = self._gradient_checkpointing_func(
+                    encoder_layer.__call__,
+                    hidden_states,
+                    attention_mask,
+                    output_attentions,
+                )
+            else:
+                layer_outputs = encoder_layer(
+                    hidden_states,
+                    attention_mask,
+                    output_attentions=output_attentions,
+                )
+
+            hidden_states = layer_outputs[0]
+
+            if output_attentions:
+                all_attentions = all_attentions + (layer_outputs[1],)
+
+        if output_hidden_states:
+            encoder_states = encoder_states + (hidden_states,)
+
+        if not return_dict:
+            return tuple(
+                v
+                for v in [hidden_states, encoder_states, all_attentions]
+                if v is not None
+            )
+        return BaseModelOutput(
+            last_hidden_state=hidden_states,
+            hidden_states=encoder_states,
+            attentions=all_attentions,
+        )

server/text_generation_server/models/idefics2.py

@@ -0,0 +1,93 @@
+import torch
+import torch.distributed
+
+from typing import List, Optional, Tuple
+
+from transformers import (
+    AutoTokenizer,
+    AutoConfig,
+    AutoProcessor,
+)
+
+from text_generation_server.models.custom_modeling.idefics2_config import IdeficsConfig
+from text_generation_server.models.custom_modeling.idefics_processing import (
+    IdeficsProcessor,
+)
+from transformers import LlamaTokenizerFast
+from text_generation_server.models.custom_modeling.idefics2_modeling import (
+    Idefics2ForVisionText2Text,
+)
+from text_generation_server.models.idefics_causal_lm import IdeficsCausalLM
+from text_generation_server.utils import (
+    initialize_torch_distributed,
+    weight_files,
+    Weights,
+)
+
+
+class IDEFICS2Sharded(IdeficsCausalLM):
+    def __init__(
+        self,
+        model_id: str,
+        revision: Optional[str] = None,
+        quantize: Optional[str] = None,
+        use_medusa: Optional[str] = None,
+        dtype: Optional[torch.dtype] = None,
+        trust_remote_code: bool = False,
+    ):
+        self.process_group, rank, world_size = initialize_torch_distributed()
+        if torch.cuda.is_available():
+            device = torch.device(f"cuda:{rank}")
+            # 9b seems to work correctly enough in float16, but 80b seems
+            # to be really saturating for f16.
+            dtype = torch.float16 if dtype is None else dtype
+        else:
+            device = torch.device("cpu")
+            dtype = torch.float32 if dtype is None else dtype
+        self.device, self.dtype = device, dtype
+
+        config = IdeficsConfig.from_pretrained(
+            model_id,
+            revision=revision,
+            trust_remote_code=trust_remote_code,
+        )
+        config.quantize = quantize
+        config.use_medusa = use_medusa
+        config.vision_config.quantize = quantize
+
+        tokenizer = LlamaTokenizerFast.from_pretrained(
+            model_id,
+            revision=revision,
+            padding_side="left",
+            truncation_side="left",
+            trust_remote_code=trust_remote_code,
+        )
+        self.processor = IdeficsProcessor.from_pretrained(
+            model_id,
+            revision=revision,
+            padding_side="left",
+            truncation_side="left",
+            trust_remote_code=trust_remote_code,
+        )
+
+        torch.distributed.barrier(group=self.process_group)
+        filenames = weight_files(model_id, revision=revision, extension=".safetensors")
+        weights = Weights(
+            filenames,
+            device=device,
+            dtype=dtype,
+            process_group=self.process_group,
+        )
+
+        model = IdeficsForVisionText2Text(config, weights)
+
+        torch.distributed.barrier(group=self.process_group)
+        super(IdeficsCausalLM, self).__init__(
+            model=model,
+            tokenizer=tokenizer,
+            requires_padding=True,
+            dtype=dtype,
+            device=device,
+            rank=rank,
+            world_size=world_size,
+        )