from typing import Callable, List, Optional, Tuple, Union import torch from torch import nn from transformers.activations import ACT2FN from transformers.cache_utils import Cache, DynamicCache, StaticCache from transformers.generation import GenerationMixin from transformers.modeling_attn_mask_utils import AttentionMaskConverter from transformers.modeling_flash_attention_utils import FlashAttentionKwargs from transformers.modeling_rope_utils import ROPE_INIT_FUNCTIONS from transformers.modeling_utils import PreTrainedModel, ALL_ATTENTION_FUNCTIONS from transformers.processing_utils import Unpack from transformers.utils import ( LossKwargs, add_code_sample_docstrings, add_start_docstrings, add_start_docstrings_to_model_forward, logging, replace_return_docstrings, ) from transformers.configuration_utils import PretrainedConfig from transformers.modeling_rope_utils import rope_config_validation class IvyConfig(PretrainedConfig): model_type = "ivy" def __init__( self, obs_size=16, out_size=8, hidden_size=4096, intermediate_size=22016, num_hidden_layers=32, num_attention_heads=32, num_key_value_heads=32, hidden_act="silu", max_position_embeddings=32768, initializer_range=0.02, rms_norm_eps=1e-6, use_cache=True, tie_word_embeddings=False, rope_theta=10000.0, rope_scaling=None, use_sliding_window=False, sliding_window=4096, max_window_layers=28, attention_dropout=0.0, **kwargs, ): self.obs_size = obs_size self.act_size = out_size self.max_position_embeddings = max_position_embeddings self.hidden_size = hidden_size self.intermediate_size = intermediate_size self.num_hidden_layers = num_hidden_layers self.num_attention_heads = num_attention_heads self.use_sliding_window = use_sliding_window self.sliding_window = sliding_window if use_sliding_window else None self.max_window_layers = max_window_layers # for backward compatibility if num_key_value_heads is None: num_key_value_heads = num_attention_heads self.num_key_value_heads = num_key_value_heads self.hidden_act = hidden_act self.initializer_range = initializer_range self.rms_norm_eps = rms_norm_eps self.use_cache = use_cache self.rope_theta = rope_theta self.rope_scaling = rope_scaling self.attention_dropout = attention_dropout # Validate the correctness of rotary position embeddings parameters # BC: if there is a 'type' field, move it to 'rope_type'. if self.rope_scaling is not None and "type" in self.rope_scaling: self.rope_scaling["rope_type"] = self.rope_scaling["type"] rope_config_validation(self) super().__init__( tie_word_embeddings=tie_word_embeddings, **kwargs, ) logger = logging.get_logger(__name__) _CONFIG_FOR_DOC = "IvyConfig" class IvyMLP(nn.Module): def __init__(self, config): super().__init__() self.config = config self.hidden_size = config.hidden_size self.intermediate_size = config.intermediate_size self.gate_proj = nn.Linear(self.hidden_size, self.intermediate_size, bias=False) self.up_proj = nn.Linear(self.hidden_size, self.intermediate_size, bias=False) self.down_proj = nn.Linear(self.intermediate_size, self.hidden_size, bias=False) self.act_fn = ACT2FN[config.hidden_act] def forward(self, x): down_proj = self.down_proj(self.act_fn(self.gate_proj(x)) * self.up_proj(x)) return down_proj def rotate_half(x): """Rotates half the hidden dims of the input.""" x1 = x[..., : x.shape[-1] // 2] x2 = x[..., x.shape[-1] // 2 :] return torch.cat((-x2, x1), dim=-1) def apply_rotary_pos_emb(q, k, cos, sin, position_ids=None, unsqueeze_dim=1): """Applies Rotary Position Embedding to the query and key tensors. Args: q (`torch.Tensor`): The query tensor. k (`torch.Tensor`): The key tensor. cos (`torch.Tensor`): The cosine part of the rotary embedding. sin (`torch.Tensor`): The sine part of the rotary embedding. position_ids (`torch.Tensor`, *optional*): Deprecated and unused. unsqueeze_dim (`int`, *optional*, defaults to 1): The 'unsqueeze_dim' argument specifies the dimension along which to unsqueeze cos[position_ids] and sin[position_ids] so that they can be properly broadcasted to the dimensions of q and k. For example, note that cos[position_ids] and sin[position_ids] have the shape [batch_size, seq_len, head_dim]. Then, if q and k have the shape [batch_size, heads, seq_len, head_dim], then setting unsqueeze_dim=1 makes cos[position_ids] and sin[position_ids] broadcastable to the shapes of q and k. Similarly, if q and k have the shape [batch_size, seq_len, heads, head_dim], then set unsqueeze_dim=2. Returns: `tuple(torch.Tensor)` comprising of the query and key tensors rotated using the Rotary Position Embedding. """ cos = cos.unsqueeze(unsqueeze_dim) sin = sin.unsqueeze(unsqueeze_dim) q_embed = (q * cos) + (rotate_half(q) * sin) k_embed = (k * cos) + (rotate_half(k) * sin) return q_embed, k_embed def repeat_kv(hidden_states: torch.Tensor, n_rep: int) -> torch.Tensor: """ This is the equivalent of torch.repeat_interleave(x, dim=1, repeats=n_rep). The hidden states go from (batch, num_key_value_heads, seqlen, head_dim) to (batch, num_attention_heads, seqlen, head_dim) """ batch, num_key_value_heads, slen, head_dim = hidden_states.shape if n_rep == 1: return hidden_states hidden_states = hidden_states[:, :, None, :, :].expand(batch, num_key_value_heads, n_rep, slen, head_dim) return hidden_states.reshape(batch, num_key_value_heads * n_rep, slen, head_dim) def eager_attention_forward( module: nn.Module, query: torch.Tensor, key: torch.Tensor, value: torch.Tensor, attention_mask: Optional[torch.Tensor], scaling: float, dropout: float = 0.0, **kwargs, ): key_states = repeat_kv(key, module.num_key_value_groups) value_states = repeat_kv(value, module.num_key_value_groups) attn_weights = torch.matmul(query, key_states.transpose(2, 3)) * scaling if attention_mask is not None: causal_mask = attention_mask[:, :, :, : key_states.shape[-2]] attn_weights = attn_weights + causal_mask attn_weights = nn.functional.softmax(attn_weights, dim=-1, dtype=torch.float32).to(query.dtype) attn_weights = nn.functional.dropout(attn_weights, p=dropout, training=module.training) attn_output = torch.matmul(attn_weights, value_states) attn_output = attn_output.transpose(1, 2).contiguous() return attn_output, attn_weights class IvyAttention(nn.Module): """Multi-headed attention from 'Attention Is All You Need' paper""" def __init__(self, config: IvyConfig, layer_idx: int): super().__init__() self.config = config self.layer_idx = layer_idx self.head_dim = getattr(config, "head_dim", config.hidden_size // config.num_attention_heads) self.num_key_value_groups = config.num_attention_heads // config.num_key_value_heads self.scaling = self.head_dim**-0.5 self.attention_dropout = config.attention_dropout self.is_causal = True self.q_proj = nn.Linear(config.hidden_size, config.num_attention_heads * self.head_dim, bias=True) self.k_proj = nn.Linear(config.hidden_size, config.num_key_value_heads * self.head_dim, bias=True) self.v_proj = nn.Linear(config.hidden_size, config.num_key_value_heads * self.head_dim, bias=True) self.o_proj = nn.Linear(config.num_attention_heads * self.head_dim, config.hidden_size, bias=False) def forward( self, hidden_states: torch.Tensor, position_embeddings: Tuple[torch.Tensor, torch.Tensor], attention_mask: Optional[torch.Tensor], past_key_value: Optional[Cache] = None, cache_position: Optional[torch.LongTensor] = None, **kwargs: Unpack[FlashAttentionKwargs], ) -> Tuple[torch.Tensor, Optional[torch.Tensor], Optional[Tuple[torch.Tensor]]]: input_shape = hidden_states.shape[:-1] hidden_shape = (*input_shape, -1, self.head_dim) query_states = self.q_proj(hidden_states).view(hidden_shape).transpose(1, 2) key_states = self.k_proj(hidden_states).view(hidden_shape).transpose(1, 2) value_states = self.v_proj(hidden_states).view(hidden_shape).transpose(1, 2) cos, sin = position_embeddings query_states, key_states = apply_rotary_pos_emb(query_states, key_states, cos, sin) if past_key_value is not None: # sin and cos are specific to RoPE models; cache_position needed for the static cache cache_kwargs = {"sin": sin, "cos": cos, "cache_position": cache_position} key_states, value_states = past_key_value.update(key_states, value_states, self.layer_idx, cache_kwargs) sliding_window = None if ( self.config.use_sliding_window and getattr(self.config, "sliding_window", None) is not None and self.layer_idx >= self.config.max_window_layers ): sliding_window = self.config.sliding_window attention_interface: Callable = eager_attention_forward if self.config._attn_implementation != "eager": if self.config._attn_implementation == "sdpa" and kwargs.get("output_attentions", False): logger.warning_once( "`torch.nn.functional.scaled_dot_product_attention` does not support `output_attentions=True`. Falling back to " 'eager attention. This warning can be removed using the argument `attn_implementation="eager"` when loading the model.' ) else: attention_interface = ALL_ATTENTION_FUNCTIONS[self.config._attn_implementation] attn_output, attn_weights = attention_interface( self, query_states, key_states, value_states, attention_mask, dropout=0.0 if not self.training else self.attention_dropout, scaling=self.scaling, sliding_window=sliding_window, # main diff with Llama **kwargs, ) attn_output = attn_output.reshape(*input_shape, -1).contiguous() attn_output = self.o_proj(attn_output) return attn_output, attn_weights class IvyRMSNorm(nn.Module): def __init__(self, hidden_size, eps=1e-6): """ IvyRMSNorm is equivalent to T5LayerNorm """ super().__init__() self.weight = nn.Parameter(torch.ones(hidden_size)) self.variance_epsilon = eps def forward(self, hidden_states): input_dtype = hidden_states.dtype hidden_states = hidden_states.to(torch.float32) variance = hidden_states.pow(2).mean(-1, keepdim=True) hidden_states = hidden_states * torch.rsqrt(variance + self.variance_epsilon) return self.weight * hidden_states.to(input_dtype) def extra_repr(self): return f"{tuple(self.weight.shape)}, eps={self.variance_epsilon}" class IvyDecoderLayer(nn.Module): def __init__(self, config: IvyConfig, layer_idx: int): super().__init__() self.hidden_size = config.hidden_size self.self_attn = IvyAttention(config=config, layer_idx=layer_idx) self.mlp = IvyMLP(config) self.input_layernorm = IvyRMSNorm(config.hidden_size, eps=config.rms_norm_eps) self.post_attention_layernorm = IvyRMSNorm(config.hidden_size, eps=config.rms_norm_eps) if config.sliding_window and config._attn_implementation != "flash_attention_2": logger.warning_once( f"Sliding Window Attention is enabled but not implemented for `{config._attn_implementation}`; " "unexpected results may be encountered." ) def forward( self, hidden_states: torch.Tensor, attention_mask: Optional[torch.Tensor] = None, position_ids: Optional[torch.LongTensor] = None, past_key_value: Optional[Cache] = None, output_attentions: Optional[bool] = False, use_cache: Optional[bool] = False, cache_position: Optional[torch.LongTensor] = None, position_embeddings: Optional[Tuple[torch.Tensor, torch.Tensor]] = None, # necessary, but kept here for BC **kwargs: Unpack[FlashAttentionKwargs], ) -> Tuple[torch.FloatTensor, Optional[Tuple[torch.FloatTensor, torch.FloatTensor]]]: residual = hidden_states hidden_states = self.input_layernorm(hidden_states) # Self Attention hidden_states, self_attn_weights = self.self_attn( hidden_states=hidden_states, attention_mask=attention_mask, position_ids=position_ids, past_key_value=past_key_value, output_attentions=output_attentions, use_cache=use_cache, cache_position=cache_position, position_embeddings=position_embeddings, **kwargs, ) hidden_states = residual + hidden_states # Fully Connected residual = hidden_states hidden_states = self.post_attention_layernorm(hidden_states) hidden_states = self.mlp(hidden_states) hidden_states = residual + hidden_states outputs = (hidden_states,) if output_attentions: outputs += (self_attn_weights,) return outputs class IvyRotaryEmbedding(nn.Module): def __init__(self, config: IvyConfig, device=None): super().__init__() # BC: "rope_type" was originally "type" if hasattr(config, "rope_scaling") and config.rope_scaling is not None: self.rope_type = config.rope_scaling.get("rope_type", config.rope_scaling.get("type")) else: self.rope_type = "default" self.max_seq_len_cached = config.max_position_embeddings self.original_max_seq_len = config.max_position_embeddings self.config = config self.rope_init_fn = ROPE_INIT_FUNCTIONS[self.rope_type] inv_freq, self.attention_scaling = self.rope_init_fn(self.config, device) self.register_buffer("inv_freq", inv_freq, persistent=False) self.original_inv_freq = self.inv_freq def _dynamic_frequency_update(self, position_ids, device): """ dynamic RoPE layers should recompute `inv_freq` in the following situations: 1 - growing beyond the cached sequence length (allow scaling) 2 - the current sequence length is in the original scale (avoid losing precision with small sequences) """ seq_len = torch.max(position_ids) + 1 if seq_len > self.max_seq_len_cached: # growth inv_freq, self.attention_scaling = self.rope_init_fn(self.config, device, seq_len=seq_len) self.register_buffer("inv_freq", inv_freq, persistent=False) # TODO joao: may break with compilation self.max_seq_len_cached = seq_len if seq_len < self.original_max_seq_len and self.max_seq_len_cached > self.original_max_seq_len: # reset # This .to() is needed if the model has been moved to a device after being initialized (because # the buffer is automatically moved, but not the original copy) self.original_inv_freq = self.original_inv_freq.to(device) self.register_buffer("inv_freq", self.original_inv_freq, persistent=False) self.max_seq_len_cached = self.original_max_seq_len @torch.no_grad() def forward(self, x, position_ids): if "dynamic" in self.rope_type: self._dynamic_frequency_update(position_ids, device=x.device) # Core RoPE block inv_freq_expanded = self.inv_freq[None, :, None].float().expand(position_ids.shape[0], -1, 1) position_ids_expanded = position_ids[:, None, :].float() # Force float32 (see https://github.com/huggingface/transformers/pull/29285) device_type = x.device.type device_type = device_type if isinstance(device_type, str) and device_type != "mps" else "cpu" with torch.autocast(device_type=device_type, enabled=False): freqs = (inv_freq_expanded.float() @ position_ids_expanded.float()).transpose(1, 2) emb = torch.cat((freqs, freqs), dim=-1) cos = emb.cos() sin = emb.sin() # Advanced RoPE types (e.g. yarn) apply a post-processing scaling factor, equivalent to scaling attention cos = cos * self.attention_scaling sin = sin * self.attention_scaling return cos.to(dtype=x.dtype), sin.to(dtype=x.dtype) class IvyPreTrainedModel(PreTrainedModel): config_class = IvyConfig base_model_prefix = "model" supports_gradient_checkpointing = True _no_split_modules = ["IvyDecoderLayer"] _skip_keys_device_placement = ["past_key_values"] _supports_flash_attn_2 = True _supports_sdpa = True _supports_flex_attn = True _supports_cache_class = True _supports_quantized_cache = True _supports_static_cache = True _supports_attention_backend = True def _init_weights(self, module): std = self.config.initializer_range if isinstance(module, nn.Linear): module.weight.data.normal_(mean=0.0, std=std) if module.bias is not None: module.bias.data.zero_() elif isinstance(module, nn.Embedding): module.weight.data.normal_(mean=0.0, std=std) if module.padding_idx is not None: module.weight.data[module.padding_idx].zero_() class Ivy4RL(IvyPreTrainedModel): def __init__(self, config: IvyConfig): super().__init__(config) self.padding_idx = config.pad_token_id self.embed_tokens = nn.Linear(config.obs_size, self.config.hidden_size) self.layers = nn.ModuleList( [IvyDecoderLayer(config, layer_idx) for layer_idx in range(config.num_hidden_layers)] ) self.norm = IvyRMSNorm(config.hidden_size, eps=config.rms_norm_eps) self.rotary_emb = IvyRotaryEmbedding(config=config) self.rl_head = nn.Linear(config.hidden_size, self.config.act_size) self.gradient_checkpointing = False # Initialize weights and apply final processing self.post_init() def get_input_embeddings(self): return self.embed_tokens def set_input_embeddings(self, value): self.embed_tokens = value def forward( self, input_ids: torch.LongTensor = None, attention_mask: Optional[torch.Tensor] = None, position_ids: Optional[torch.LongTensor] = None, past_key_values: Optional[Cache] = None, inputs_embeds: Optional[torch.FloatTensor] = 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, **flash_attn_kwargs: Unpack[FlashAttentionKwargs], ): use_cache = False # Yeah inputs_embeds = self.embed_tokens(inputs_embeds) # Yeah, ikr if use_cache and past_key_values is None: past_key_values = DynamicCache() if cache_position is None: past_seen_tokens = past_key_values.get_seq_length() if past_key_values is not None else 0 cache_position = torch.arange( past_seen_tokens, past_seen_tokens + inputs_embeds.shape[1], device=inputs_embeds.device ) if position_ids is None: position_ids = cache_position.unsqueeze(0) causal_mask = self._update_causal_mask( attention_mask, inputs_embeds, cache_position, past_key_values, output_attentions ) hidden_states = inputs_embeds # create position embeddings to be shared across the decoder layers position_embeddings = self.rotary_emb(hidden_states, position_ids) # decoder layers all_hidden_states = () if output_hidden_states else None all_self_attns = () if output_attentions else None for decoder_layer in self.layers[: self.config.num_hidden_layers]: if output_hidden_states: all_hidden_states += (hidden_states,) if self.gradient_checkpointing and self.training: layer_outputs = self._gradient_checkpointing_func( decoder_layer.__call__, hidden_states, causal_mask, position_ids, past_key_values, output_attentions, use_cache, cache_position, position_embeddings, ) else: layer_outputs = decoder_layer( hidden_states, attention_mask=causal_mask, position_ids=position_ids, past_key_value=past_key_values, output_attentions=output_attentions, use_cache=use_cache, cache_position=cache_position, position_embeddings=position_embeddings, **flash_attn_kwargs, ) hidden_states = layer_outputs[0] if output_attentions: all_self_attns += (layer_outputs[1],) hidden_states = self.norm(hidden_states) logits = self.rl_head(hidden_states[:, -1:, :]) logits = logits.squeeze(1) return logits def _update_causal_mask( self, attention_mask: torch.Tensor, input_tensor: torch.Tensor, cache_position: torch.Tensor, past_key_values: Cache, output_attentions: bool, ): if self.config._attn_implementation == "flash_attention_2": if attention_mask is not None and (attention_mask == 0.0).any(): return attention_mask return None # For SDPA, when possible, we will rely on its `is_causal` argument instead of its `attn_mask` argument, in # order to dispatch on Flash Attention 2. This feature is not compatible with static cache, as SDPA will fail # to infer the attention mask. past_seen_tokens = past_key_values.get_seq_length() if past_key_values is not None else 0 using_static_cache = isinstance(past_key_values, StaticCache) # When output attentions is True, sdpa implementation's forward method calls the eager implementation's forward if self.config._attn_implementation == "sdpa" and not using_static_cache and not output_attentions: if AttentionMaskConverter._ignore_causal_mask_sdpa( attention_mask, inputs_embeds=input_tensor, past_key_values_length=past_seen_tokens, is_training=self.training, ): return None dtype, device = input_tensor.dtype, input_tensor.device sequence_length = input_tensor.shape[1] if using_static_cache: target_length = past_key_values.get_max_cache_shape() else: target_length = ( attention_mask.shape[-1] if isinstance(attention_mask, torch.Tensor) else past_seen_tokens + sequence_length + 1 ) # In case the provided `attention` mask is 2D, we generate a causal mask here (4D). causal_mask = self._prepare_4d_causal_attention_mask_with_cache_position( attention_mask, sequence_length=sequence_length, target_length=target_length, dtype=dtype, device=device, cache_position=cache_position, batch_size=input_tensor.shape[0], ) if ( self.config._attn_implementation == "sdpa" and attention_mask is not None and attention_mask.device.type == "cuda" and not output_attentions ): # Attend to all tokens in fully masked rows in the causal_mask, for example the relevant first rows when # using left padding. This is required by F.scaled_dot_product_attention memory-efficient attention path. # Details: https://github.com/pytorch/pytorch/issues/110213 min_dtype = torch.finfo(dtype).min causal_mask = AttentionMaskConverter._unmask_unattended(causal_mask, min_dtype) return causal_mask @staticmethod def _prepare_4d_causal_attention_mask_with_cache_position( attention_mask: torch.Tensor, sequence_length: int, target_length: int, dtype: torch.dtype, device: torch.device, cache_position: torch.Tensor, batch_size: int, **kwargs, ): if attention_mask is not None and attention_mask.dim() == 4: # In this case we assume that the mask comes already in inverted form and requires no inversion or slicing. causal_mask = attention_mask else: min_dtype = torch.finfo(dtype).min causal_mask = torch.full( (sequence_length, target_length), fill_value=min_dtype, dtype=dtype, device=device ) if sequence_length != 1: causal_mask = torch.triu(causal_mask, diagonal=1) causal_mask *= torch.arange(target_length, device=device) > cache_position.reshape(-1, 1) causal_mask = causal_mask[None, None, :, :].expand(batch_size, 1, -1, -1) if attention_mask is not None: causal_mask = causal_mask.clone() # copy to contiguous memory for in-place edit mask_length = attention_mask.shape[-1] padding_mask = causal_mask[:, :, :, :mask_length] + attention_mask[:, None, None, :] padding_mask = padding_mask == 0 causal_mask[:, :, :, :mask_length] = causal_mask[:, :, :, :mask_length].masked_fill( padding_mask, min_dtype ) return causal_mask