From d5c00047da61a77923ffb351ff4de4ea0e2a38f3 Mon Sep 17 00:00:00 2001 From: Arthur Zucker Date: Wed, 15 May 2024 16:07:05 +0200 Subject: [PATCH] updates --- .../models/gemma/modeling_gemma.py | 196 +-- .../models/persimmon/modeling_persimmon.py | 1044 ++++++++++++++++ .../models/stablelm/modeling_stablelm.py | 1072 +++++++++++++++-- .../models/starcoder2/modeling_starcoder2.py | 246 ++-- utils/convert_diff_model.py | 24 +- 5 files changed, 2287 insertions(+), 295 deletions(-) diff --git a/src/transformers/models/gemma/modeling_gemma.py b/src/transformers/models/gemma/modeling_gemma.py index bcbe6dcfbd..968e142b3f 100644 --- a/src/transformers/models/gemma/modeling_gemma.py +++ b/src/transformers/models/gemma/modeling_gemma.py @@ -612,116 +612,118 @@ class LlamaModel(LlamaPreTrainedModel): hidden_states = inputs_embeds normalizer = torch.tensor(self.config.hidden_size**0.5, dtype=hidden_states.dtype) hidden_states = hidden_states * normalizer - @add_start_docstrings_to_model_forward(LLAMA_INPUTS_DOCSTRING) - def forward( - self, - input_ids: torch.LongTensor = None, - attention_mask: Optional[torch.Tensor] = None, - position_ids: Optional[torch.LongTensor] = None, - past_key_values: Optional[Union[Cache, List[torch.FloatTensor]]] = 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, - ) -> Union[Tuple, BaseModelOutputWithPast]: - 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 - ) - use_cache = use_cache if use_cache is not None else self.config.use_cache - return_dict = return_dict if return_dict is not None else self.config.use_return_dict - - if (input_ids is None) ^ (inputs_embeds is not None): - raise ValueError( - "You cannot specify both input_ids and inputs_embeds at the same time, and must specify either one" + return (None, attention_mask, position_ids, past_key_values, inputs_embeds, use_cache, output_attentions, output_hidden_states, return_dict, cache_position + @add_start_docstrings_to_model_forward(LLAMA_INPUTS_DOCSTRING) + def forward( + self, + input_ids: torch.LongTensor = None, + attention_mask: Optional[torch.Tensor] = None, + position_ids: Optional[torch.LongTensor] = None, + past_key_values: Optional[Union[Cache, List[torch.FloatTensor]]] = 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, + ) -> Union[Tuple, BaseModelOutputWithPast]: + 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 ) + use_cache = use_cache if use_cache is not None else self.config.use_cache + return_dict = return_dict if return_dict is not None else self.config.use_return_dict - if self.gradient_checkpointing and self.training and use_cache: - logger.warning_once( - "`use_cache=True` is incompatible with gradient checkpointing. Setting `use_cache=False`." - ) - use_cache = False + if (input_ids is None) ^ (inputs_embeds is not None): + raise ValueError( + "You cannot specify both input_ids and inputs_embeds at the same time, and must specify either one" + ) - if inputs_embeds is None: - inputs_embeds = self.embed_tokens(input_ids) + if self.gradient_checkpointing and self.training and use_cache: + logger.warning_once( + "`use_cache=True` is incompatible with gradient checkpointing. Setting `use_cache=False`." + ) + use_cache = False - return_legacy_cache = False - if use_cache and not isinstance(past_key_values, Cache): # kept for BC (non `Cache` `past_key_values` inputs) - return_legacy_cache = True - past_key_values = DynamicCache.from_legacy_cache(past_key_values) + if inputs_embeds is None: + inputs_embeds = self.embed_tokens(input_ids) - 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) + return_legacy_cache = False + if use_cache and not isinstance(past_key_values, Cache): # kept for BC (non `Cache` `past_key_values` inputs) + return_legacy_cache = True + past_key_values = DynamicCache.from_legacy_cache(past_key_values) - causal_mask = self._update_causal_mask(attention_mask, inputs_embeds, cache_position, past_key_values) + 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) - # embed positions - hidden_states = inputs_embeds + causal_mask = self._update_causal_mask(attention_mask, inputs_embeds, cache_position, past_key_values) - # decoder layers - all_hidden_states = () if output_hidden_states else None - all_self_attns = () if output_attentions else None - next_decoder_cache = None + # embed positions + hidden_states = inputs_embeds - for decoder_layer in self.layers: + # decoder layers + all_hidden_states = () if output_hidden_states else None + all_self_attns = () if output_attentions else None + next_decoder_cache = None + + for decoder_layer in self.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, + ) + 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, + ) + + hidden_states = layer_outputs[0] + + if use_cache: + next_decoder_cache = layer_outputs[2 if output_attentions else 1] + + if output_attentions: + all_self_attns += (layer_outputs[1],) + + hidden_states = self.norm(hidden_states) + + # add hidden states from the last decoder layer 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, - ) - 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, - ) + next_cache = next_decoder_cache if use_cache else None + if return_legacy_cache: + next_cache = next_cache.to_legacy_cache() - hidden_states = layer_outputs[0] - - if use_cache: - next_decoder_cache = layer_outputs[2 if output_attentions else 1] - - if output_attentions: - all_self_attns += (layer_outputs[1],) - - hidden_states = self.norm(hidden_states) - - # add hidden states from the last decoder layer - if output_hidden_states: - all_hidden_states += (hidden_states,) - - next_cache = next_decoder_cache if use_cache else None - if return_legacy_cache: - next_cache = next_cache.to_legacy_cache() - - if not return_dict: - return tuple(v for v in [hidden_states, next_cache, all_hidden_states, all_self_attns] if v is not None) - return BaseModelOutputWithPast( - last_hidden_state=hidden_states, - past_key_values=next_cache, - hidden_states=all_hidden_states, - attentions=all_self_attns, - )None, attention_mask, position_ids, past_key_values, inputs_embeds, use_cache, output_attentions, output_hidden_states, return_dict, cache_position) + if not return_dict: + return tuple(v for v in [hidden_states, next_cache, all_hidden_states, all_self_attns] if v is not None) + return BaseModelOutputWithPast( + last_hidden_state=hidden_states, + past_key_values=next_cache, + hidden_states=all_hidden_states, + attentions=all_self_attns, + ) + ) def _update_causal_mask( self, diff --git a/src/transformers/models/persimmon/modeling_persimmon.py b/src/transformers/models/persimmon/modeling_persimmon.py index e69de29bb2..b5c4807432 100644 --- a/src/transformers/models/persimmon/modeling_persimmon.py +++ b/src/transformers/models/persimmon/modeling_persimmon.py @@ -0,0 +1,1044 @@ +from transformers.models.llama.modeling_llama import * +import torch.nn as nn +from .configuration_persimmon import PersimmonConfig +from transformers.utils import ModelConverter + + +class PersimmonRotaryEmbedding(nn.Module): + def __init__(self, dim, max_position_embeddings=2048, base=10000, device=None, scaling_factor=1.0): + super().__init__() + self.scaling_factor = scaling_factor + self.dim = dim + self.max_position_embeddings = max_position_embeddings + self.base = base + inv_freq = 1.0 / (self.base ** (torch.arange(0, self.dim, 2, dtype=torch.int64).float().to(device) / self.dim)) + self.register_buffer("inv_freq", inv_freq, persistent=False) + + @torch.no_grad() + def compute_cos_sin(self, x, position_ids): + # x: [bs, num_attention_heads, seq_len, head_size] + 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 since bfloat16 loses precision on long contexts + # 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) + return emb.cos(), emb.sin() + + def rotate_half(self, 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) + + @torch.no_grad() + @add_start_docstrings("") + def forward(self, q, k, 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. + 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, sin = self.compute_cos_sin(q, position_ids) + cos = cos.unsqueeze(unsqueeze_dim).to(dtype=q.dtype) + sin = sin.unsqueeze(unsqueeze_dim).to(dtype=q.dtype) + q_embed = (q * cos) + (self.rotate_half(q) * sin) + k_embed = (k * cos) + (self.rotate_half(k) * sin) + return q_embed, k_embed + +class PersimmonMLP(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=config.mlp_bias) + self.up_proj = nn.Linear(self.hidden_size, self.intermediate_size, bias=config.mlp_bias) + self.down_proj = nn.Linear(self.intermediate_size, self.hidden_size, bias=config.mlp_bias) + 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 + + +class PersimmonAttention(nn.Module): + """Multi-headed attention from 'Attention Is All You Need' paper""" + def __init__(self, config: PersimmonConfig, layer_idx: Optional[int] = None): + + def __init__(self, config: LlamaConfig, layer_idx: Optional[int] = None): + super().__init__() + self.config = config + self.layer_idx = layer_idx + if layer_idx is None: + logger.warning_once( + f"Instantiating {self.__class__.__name__} without passing a `layer_idx` is not recommended and will " + "lead to errors during the forward call if caching is used. Please make sure to provide a `layer_idx` " + "when creating this class." + ) + + self.attention_dropout = config.attention_dropout + self.hidden_size = config.hidden_size + self.num_heads = config.num_attention_heads + self.head_dim = self.hidden_size // self.num_heads + self.num_key_value_heads = config.num_key_value_heads + self.num_key_value_groups = self.num_heads // self.num_key_value_heads + self.max_position_embeddings = config.max_position_embeddings + self.rope_theta = config.rope_theta + self.is_causal = True + + if (self.head_dim * self.num_heads) != self.hidden_size: + raise ValueError( + f"hidden_size must be divisible by num_heads (got `hidden_size`: {self.hidden_size}" + f" and `num_heads`: {self.num_heads})." + ) + + self.q_proj = nn.Linear(self.hidden_size, self.num_heads * self.head_dim, bias=config.attention_bias) + self.k_proj = nn.Linear(self.hidden_size, self.num_key_value_heads * self.head_dim, bias=config.attention_bias) + self.v_proj = nn.Linear(self.hidden_size, self.num_key_value_heads * self.head_dim, bias=config.attention_bias) + self.o_proj = nn.Linear(self.hidden_size, self.hidden_size, bias=config.attention_bias) + self._init_rope() + ... # copy before? add the line? how to best support this + self.query_key_value = nn.Linear(self.hidden_size, 3 * self.hidden_size, bias=True) + self.dense = nn.Linear(self.num_heads * self.head_dim, self.hidden_size, bias=True) + self.qk_layernorm = config.qk_layernorm + + if self.qk_layernorm: + self.q_layernorm = nn.LayerNorm( + config.hidden_size // self.num_heads, eps=config.layer_norm_eps, elementwise_affine=True + ) + self.k_layernorm = nn.LayerNorm( + config.hidden_size // self.num_heads, eps=config.layer_norm_eps, elementwise_affine=True + ) + self.attention_dropout = nn.Dropout(config.attention_dropout) + self._init_rope() + def _init_rope(self): + if self.config.rope_scaling is None: + self.rotary_emb = LlamaRotaryEmbedding( + self.head_dim, + max_position_embeddings=self.max_position_embeddings, + base=self.rope_theta, + ) + else: + scaling_type = self.config.rope_scaling["type"] + scaling_factor = self.config.rope_scaling["factor"] + if scaling_type == "linear": + self.rotary_emb = LlamaLinearScalingRotaryEmbedding( + self.head_dim, + max_position_embeddings=self.max_position_embeddings, + scaling_factor=scaling_factor, + base=self.rope_theta, + ) + elif scaling_type == "dynamic": + self.rotary_emb = LlamaDynamicNTKScalingRotaryEmbedding( + self.head_dim, + max_position_embeddings=self.max_position_embeddings, + scaling_factor=scaling_factor, + base=self.rope_theta, + ) + else: + raise ValueError(f"Unknown RoPE scaling type {scaling_type}") + 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: bool = False, + use_cache: bool = False, + ) -> Tuple[torch.Tensor, Optional[torch.Tensor], Optional[Tuple[torch.Tensor]]]: + bsz, q_len, _ = hidden_states.size() + + # [batch_size, seq_length, 3 x hidden_size] + fused_qkv = self.query_key_value(hidden_states) + + # 3 x [batch_size, seq_length, num_heads, head_dim] + (query_states, key_states, value_states) = self._split_heads(fused_qkv) + + if self.qk_layernorm: + query_states = self.q_layernorm(query_states) + key_states = self.k_layernorm(key_states) + + # [batch_size, num_heads, seq_length, head_dim] -> [batch_size, seq_length, num_heads, head_dim] + query_states = query_states.transpose(1, 2) + value_states = value_states.transpose(1, 2) + key_states = key_states.transpose(1, 2) + + os, sin = self.rotary_emb(value_states, seq_len=None) + + # Partial rotary embedding + query_rot, query_pass = ( + query_states[..., : self.rotary_emb.dim], + query_states[..., self.rotary_emb.dim :], + ) + key_rot, key_pass = ( + key_states[..., : self.rotary_emb.dim], + key_states[..., self.rotary_emb.dim :], + ) + # [batch_size, seq_length, num_heads, head_dim // config.partial_rotary_factor] + query_rot, key_rot = self.rotary_emb(query_rot, key_rot, position_ids) + + # [batch_size, seq_length, num_heads, head_dim] + query_states = torch.cat((query_rot, query_pass), dim=-1) + key_states = torch.cat((key_rot, key_pass), dim=-1) + ... # TODO copy the rest of the function? if we do this it's unusable + + """Multi-headed attention from 'Attention Is All You Need' paper""" + + def __init__(self, config: PersimmonConfig, layer_idx: Optional[int] = None): + super().__init__() + ... # copy before? add the line? how to best support this + self.query_key_value = nn.Linear(self.hidden_size, 3 * self.hidden_size, bias=True) + self.dense = nn.Linear(self.num_heads * self.head_dim, self.hidden_size, bias=True) + self.qk_layernorm = config.qk_layernorm + + if self.qk_layernorm: + self.q_layernorm = nn.LayerNorm( + config.hidden_size // self.num_heads, eps=config.layer_norm_eps, elementwise_affine=True + ) + self.k_layernorm = nn.LayerNorm( + config.hidden_size // self.num_heads, eps=config.layer_norm_eps, elementwise_affine=True + ) + self.attention_dropout = nn.Dropout(config.attention_dropout) + self._init_rope() + + 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: bool = False, + use_cache: bool = False, + ) -> Tuple[torch.Tensor, Optional[torch.Tensor], Optional[Tuple[torch.Tensor]]]: + bsz, q_len, _ = hidden_states.size() + + # [batch_size, seq_length, 3 x hidden_size] + fused_qkv = self.query_key_value(hidden_states) + + # 3 x [batch_size, seq_length, num_heads, head_dim] + (query_states, key_states, value_states) = self._split_heads(fused_qkv) + + if self.qk_layernorm: + query_states = self.q_layernorm(query_states) + key_states = self.k_layernorm(key_states) + + # [batch_size, num_heads, seq_length, head_dim] -> [batch_size, seq_length, num_heads, head_dim] + query_states = query_states.transpose(1, 2) + value_states = value_states.transpose(1, 2) + key_states = key_states.transpose(1, 2) + + os, sin = self.rotary_emb(value_states, seq_len=None) + + # Partial rotary embedding + query_rot, query_pass = ( + query_states[..., : self.rotary_emb.dim], + query_states[..., self.rotary_emb.dim :], + ) + key_rot, key_pass = ( + key_states[..., : self.rotary_emb.dim], + key_states[..., self.rotary_emb.dim :], + ) + # [batch_size, seq_length, num_heads, head_dim // config.partial_rotary_factor] + query_rot, key_rot = self.rotary_emb(query_rot, key_rot, position_ids) + + # [batch_size, seq_length, num_heads, head_dim] + query_states = torch.cat((query_rot, query_pass), dim=-1) + key_states = torch.cat((key_rot, key_pass), dim=-1) + ... # TODO copy the rest of the function? if we do this it's unusable + + + +class PersimmonSdpaAttention(nn.Module): + """Multi-headed attention from 'Attention Is All You Need' paper""" + + def __init__(self, config: PersimmonConfig, layer_idx: Optional[int] = None): + super().__init__() + self.config = config + self.layer_idx = layer_idx + if layer_idx is None: + logger.warning_once( + f"Instantiating {self.__class__.__name__} without passing a `layer_idx` is not recommended and will " + "lead to errors during the forward call if caching is used. Please make sure to provide a `layer_idx` " + "when creating this class." + ) + + self.attention_dropout = config.attention_dropout + self.hidden_size = config.hidden_size + self.num_heads = config.num_attention_heads + self.head_dim = self.hidden_size // self.num_heads + self.num_key_value_heads = config.num_key_value_heads + self.num_key_value_groups = self.num_heads // self.num_key_value_heads + self.max_position_embeddings = config.max_position_embeddings + self.rope_theta = config.rope_theta + self.is_causal = True + + if (self.head_dim * self.num_heads) != self.hidden_size: + raise ValueError( + f"hidden_size must be divisible by num_heads (got `hidden_size`: {self.hidden_size}" + f" and `num_heads`: {self.num_heads})." + ) + + self.q_proj = nn.Linear(self.hidden_size, self.num_heads * self.head_dim, bias=config.attention_bias) + self.k_proj = nn.Linear(self.hidden_size, self.num_key_value_heads * self.head_dim, bias=config.attention_bias) + self.v_proj = nn.Linear(self.hidden_size, self.num_key_value_heads * self.head_dim, bias=config.attention_bias) + self.o_proj = nn.Linear(self.hidden_size, self.hidden_size, bias=config.attention_bias) + self._init_rope() + + def _init_rope(self): + if self.config.rope_scaling is None: + self.rotary_emb = PersimmonRotaryEmbedding( + self.head_dim, + max_position_embeddings=self.max_position_embeddings, + base=self.rope_theta, + ) + else: + scaling_type = self.config.rope_scaling["type"] + scaling_factor = self.config.rope_scaling["factor"] + if scaling_type == "linear": + self.rotary_emb = PersimmonLinearScalingRotaryEmbedding( + self.head_dim, + max_position_embeddings=self.max_position_embeddings, + scaling_factor=scaling_factor, + base=self.rope_theta, + ) + elif scaling_type == "dynamic": + self.rotary_emb = PersimmonDynamicNTKScalingRotaryEmbedding( + self.head_dim, + max_position_embeddings=self.max_position_embeddings, + scaling_factor=scaling_factor, + base=self.rope_theta, + ) + else: + raise ValueError(f"Unknown RoPE scaling type {scaling_type}") + + 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: bool = False, + use_cache: bool = False, + cache_position: Optional[torch.LongTensor] = None, + ) -> Tuple[torch.Tensor, Optional[torch.Tensor], Optional[Tuple[torch.Tensor]]]: + 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) + + 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_key_value_heads, self.head_dim).transpose(1, 2) + value_states = value_states.view(bsz, q_len, self.num_key_value_heads, self.head_dim).transpose(1, 2) + + query_states, key_states = self.rotary_emb(query_states, key_states, position_ids) + + if past_key_value is not None: + # sin and cos are specific to RoPE models; cache_position needed for the static cache + cache_kwargs = {"cache_position": cache_position} + key_states, value_states = past_key_value.update(key_states, value_states, self.layer_idx, cache_kwargs) + + key_states = repeat_kv(key_states, self.num_key_value_groups) + value_states = repeat_kv(value_states, self.num_key_value_groups) + + attn_weights = torch.matmul(query_states, key_states.transpose(2, 3)) / math.sqrt(self.head_dim) + + if attention_mask is not None: # no matter the length, we just slice it + causal_mask = attention_mask[:, :, :, : key_states.shape[-2]] + attn_weights = attn_weights + causal_mask + + # upcast attention to fp32 + attn_weights = nn.functional.softmax(attn_weights, dim=-1, dtype=torch.float32).to(query_states.dtype) + attn_weights = nn.functional.dropout(attn_weights, p=self.attention_dropout, training=self.training) + attn_output = torch.matmul(attn_weights, value_states) + + if attn_output.size() != (bsz, self.num_heads, q_len, self.head_dim): + raise ValueError( + f"`attn_output` should be of size {(bsz, self.num_heads, q_len, self.head_dim)}, but is" + f" {attn_output.size()}" + ) + + attn_output = attn_output.transpose(1, 2).contiguous() + + attn_output = attn_output.reshape(bsz, q_len, self.hidden_size) + attn_output = self.o_proj(attn_output) + + if not output_attentions: + attn_weights = None + + return attn_output, attn_weights, past_key_value + +class PersimmonFlashAttention2(nn.Module): + """Multi-headed attention from 'Attention Is All You Need' paper""" + + def __init__(self, config: PersimmonConfig, layer_idx: Optional[int] = None): + super().__init__() + self.config = config + self.layer_idx = layer_idx + if layer_idx is None: + logger.warning_once( + f"Instantiating {self.__class__.__name__} without passing a `layer_idx` is not recommended and will " + "lead to errors during the forward call if caching is used. Please make sure to provide a `layer_idx` " + "when creating this class." + ) + + self.attention_dropout = config.attention_dropout + self.hidden_size = config.hidden_size + self.num_heads = config.num_attention_heads + self.head_dim = self.hidden_size // self.num_heads + self.num_key_value_heads = config.num_key_value_heads + self.num_key_value_groups = self.num_heads // self.num_key_value_heads + self.max_position_embeddings = config.max_position_embeddings + self.rope_theta = config.rope_theta + self.is_causal = True + + if (self.head_dim * self.num_heads) != self.hidden_size: + raise ValueError( + f"hidden_size must be divisible by num_heads (got `hidden_size`: {self.hidden_size}" + f" and `num_heads`: {self.num_heads})." + ) + + self.q_proj = nn.Linear(self.hidden_size, self.num_heads * self.head_dim, bias=config.attention_bias) + self.k_proj = nn.Linear(self.hidden_size, self.num_key_value_heads * self.head_dim, bias=config.attention_bias) + self.v_proj = nn.Linear(self.hidden_size, self.num_key_value_heads * self.head_dim, bias=config.attention_bias) + self.o_proj = nn.Linear(self.hidden_size, self.hidden_size, bias=config.attention_bias) + self._init_rope() + + def _init_rope(self): + if self.config.rope_scaling is None: + self.rotary_emb = PersimmonRotaryEmbedding( + self.head_dim, + max_position_embeddings=self.max_position_embeddings, + base=self.rope_theta, + ) + else: + scaling_type = self.config.rope_scaling["type"] + scaling_factor = self.config.rope_scaling["factor"] + if scaling_type == "linear": + self.rotary_emb = PersimmonLinearScalingRotaryEmbedding( + self.head_dim, + max_position_embeddings=self.max_position_embeddings, + scaling_factor=scaling_factor, + base=self.rope_theta, + ) + elif scaling_type == "dynamic": + self.rotary_emb = PersimmonDynamicNTKScalingRotaryEmbedding( + self.head_dim, + max_position_embeddings=self.max_position_embeddings, + scaling_factor=scaling_factor, + base=self.rope_theta, + ) + else: + raise ValueError(f"Unknown RoPE scaling type {scaling_type}") + + 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: bool = False, + use_cache: bool = False, + cache_position: Optional[torch.LongTensor] = None, + ) -> Tuple[torch.Tensor, Optional[torch.Tensor], Optional[Tuple[torch.Tensor]]]: + 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) + + 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_key_value_heads, self.head_dim).transpose(1, 2) + value_states = value_states.view(bsz, q_len, self.num_key_value_heads, self.head_dim).transpose(1, 2) + + query_states, key_states = self.rotary_emb(query_states, key_states, position_ids) + + if past_key_value is not None: + # sin and cos are specific to RoPE models; cache_position needed for the static cache + cache_kwargs = {"cache_position": cache_position} + key_states, value_states = past_key_value.update(key_states, value_states, self.layer_idx, cache_kwargs) + + key_states = repeat_kv(key_states, self.num_key_value_groups) + value_states = repeat_kv(value_states, self.num_key_value_groups) + + attn_weights = torch.matmul(query_states, key_states.transpose(2, 3)) / math.sqrt(self.head_dim) + + if attention_mask is not None: # no matter the length, we just slice it + causal_mask = attention_mask[:, :, :, : key_states.shape[-2]] + attn_weights = attn_weights + causal_mask + + # upcast attention to fp32 + attn_weights = nn.functional.softmax(attn_weights, dim=-1, dtype=torch.float32).to(query_states.dtype) + attn_weights = nn.functional.dropout(attn_weights, p=self.attention_dropout, training=self.training) + attn_output = torch.matmul(attn_weights, value_states) + + if attn_output.size() != (bsz, self.num_heads, q_len, self.head_dim): + raise ValueError( + f"`attn_output` should be of size {(bsz, self.num_heads, q_len, self.head_dim)}, but is" + f" {attn_output.size()}" + ) + + attn_output = attn_output.transpose(1, 2).contiguous() + + attn_output = attn_output.reshape(bsz, q_len, self.hidden_size) + attn_output = self.o_proj(attn_output) + + if not output_attentions: + attn_weights = None + + return attn_output, attn_weights, past_key_value + + +COHERE_ATTENTION_CLASSES = {"eager": PersimmonAttention, "flash_attention_2": PersimmonFlashAttention2, "sdpa": PersimmonSdpaAttention} + +class PersimmonDecoderLayer(nn.Module): + def __init__(self, config: PersimmonConfig, layer_idx: int): + super().__init__() + self.hidden_size = config.hidden_size + + self.self_attn = LLAMA_ATTENTION_CLASSES[config._attn_implementation](config=config, layer_idx=layer_idx) + + self.mlp = PersimmonMLP(config) + self.input_layernorm = PersimmonRMSNorm(config.hidden_size, eps=config.rms_norm_eps) + self.post_attention_layernorm = PersimmonRMSNorm(config.hidden_size, eps=config.rms_norm_eps) + + 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, + ) -> Tuple[torch.FloatTensor, Optional[Tuple[torch.FloatTensor, torch.FloatTensor]]]: + """ + Args: + hidden_states (`torch.FloatTensor`): input to the layer of shape `(batch, seq_len, embed_dim)` + attention_mask (`torch.FloatTensor`, *optional*): + attention mask of size `(batch_size, sequence_length)` if flash attention is used or `(batch_size, 1, + query_sequence_length, key_sequence_length)` if default attention is used. + output_attentions (`bool`, *optional*): + Whether or not to return the attentions tensors of all attention layers. See `attentions` under + returned tensors for more detail. + use_cache (`bool`, *optional*): + If set to `True`, `past_key_values` key value states are returned and can be used to speed up decoding + (see `past_key_values`). + past_key_value (`Tuple(torch.FloatTensor)`, *optional*): cached past key and value projection states + """ + + residual = hidden_states + + hidden_states = self.input_layernorm(hidden_states) + + # Self Attention + hidden_states, self_attn_weights, present_key_value = 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, + ) + 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,) + + if use_cache: + outputs += (present_key_value,) + + return outputs + +@add_start_docstrings( + "The bare LLaMA Model outputting raw hidden-states without any specific head on top.", + LLAMA_START_DOCSTRING, +) +class PersimmonPreTrainedModel(PreTrainedModel): + config_class = PersimmonConfig + base_model_prefix = "model" + supports_gradient_checkpointing = True + _no_split_modules = ["PersimmonDecoderLayer"] + _skip_keys_device_placement = ["past_key_values"] + _supports_flash_attn_2 = True + _supports_sdpa = True + _supports_cache_class = 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_() + + +@add_start_docstrings( + "The bare LLaMA Model outputting raw hidden-states without any specific head on top.", + LLAMA_START_DOCSTRING, +) +class PersimmonModel(PersimmonPreTrainedModel): + """ + Transformer decoder consisting of *config.num_hidden_layers* layers. Each layer is a [`PersimmonDecoderLayer`] + + Args: + config: PersimmonConfig + """ + + def __init__(self, config: PersimmonConfig): + super().__init__(config) + self.padding_idx = config.pad_token_id + self.vocab_size = config.vocab_size + + self.embed_tokens = nn.Embedding(config.vocab_size, config.hidden_size, self.padding_idx) + self.layers = nn.ModuleList( + [PersimmonDecoderLayer(config, layer_idx) for layer_idx in range(config.num_hidden_layers)] + ) + self.norm = PersimmonRMSNorm(config.hidden_size, eps=config.rms_norm_eps) + 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 + + @add_start_docstrings_to_model_forward(LLAMA_INPUTS_DOCSTRING) + def forward( + self, + input_ids: torch.LongTensor = None, + attention_mask: Optional[torch.Tensor] = None, + position_ids: Optional[torch.LongTensor] = None, + past_key_values: Optional[Union[Cache, List[torch.FloatTensor]]] = 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, + ) -> Union[Tuple, BaseModelOutputWithPast]: + 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 + ) + use_cache = use_cache if use_cache is not None else self.config.use_cache + return_dict = return_dict if return_dict is not None else self.config.use_return_dict + + if (input_ids is None) ^ (inputs_embeds is not None): + raise ValueError( + "You cannot specify both input_ids and inputs_embeds at the same time, and must specify either one" + ) + + if self.gradient_checkpointing and self.training and use_cache: + logger.warning_once( + "`use_cache=True` is incompatible with gradient checkpointing. Setting `use_cache=False`." + ) + use_cache = False + + if inputs_embeds is None: + inputs_embeds = self.embed_tokens(input_ids) + + return_legacy_cache = False + if use_cache and not isinstance(past_key_values, Cache): # kept for BC (non `Cache` `past_key_values` inputs) + return_legacy_cache = True + past_key_values = DynamicCache.from_legacy_cache(past_key_values) + + 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) + + # embed positions + hidden_states = inputs_embeds + + # decoder layers + all_hidden_states = () if output_hidden_states else None + all_self_attns = () if output_attentions else None + next_decoder_cache = None + + for decoder_layer in self.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, + ) + 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, + ) + + hidden_states = layer_outputs[0] + + if use_cache: + next_decoder_cache = layer_outputs[2 if output_attentions else 1] + + if output_attentions: + all_self_attns += (layer_outputs[1],) + + hidden_states = self.norm(hidden_states) + + # add hidden states from the last decoder layer + if output_hidden_states: + all_hidden_states += (hidden_states,) + + next_cache = next_decoder_cache if use_cache else None + if return_legacy_cache: + next_cache = next_cache.to_legacy_cache() + + if not return_dict: + return tuple(v for v in [hidden_states, next_cache, all_hidden_states, all_self_attns] if v is not None) + return BaseModelOutputWithPast( + last_hidden_state=hidden_states, + past_key_values=next_cache, + hidden_states=all_hidden_states, + attentions=all_self_attns, + ) + + def _update_causal_mask( + self, + attention_mask: torch.Tensor, + input_tensor: torch.Tensor, + cache_position: torch.Tensor, + past_key_values: Cache, + ): + # TODO: As of torch==2.2.0, the `attention_mask` passed to the model in `generate` is 2D and of dynamic length even when the static + # KV cache is used. This is an issue for torch.compile which then recaptures cudagraphs at each decode steps due to the dynamic shapes. + # (`recording cudagraph tree for symint key 13`, etc.), which is VERY slow. A workaround is `@torch.compiler.disable`, but this prevents using + # `fullgraph=True`. See more context in https://github.com/huggingface/transformers/pull/29114 + + if self.config._attn_implementation == "flash_attention_2": + if attention_mask is not None and 0.0 in attention_mask: + 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) + if self.config._attn_implementation == "sdpa" and not using_static_cache: + 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 + min_dtype = torch.finfo(dtype).min + sequence_length = input_tensor.shape[1] + if using_static_cache: + target_length = past_key_values.get_max_length() + else: + target_length = ( + attention_mask.shape[-1] + if isinstance(attention_mask, torch.Tensor) + else past_seen_tokens + sequence_length + 1 + ) + + 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(input_tensor.shape[0], 1, -1, -1) + if attention_mask is not None: + causal_mask = causal_mask.clone() # copy to contiguous memory for in-place edit + if attention_mask.dim() == 2: + 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 + ) + elif attention_mask.dim() == 4: + # backwards compatibility: we allow passing a 4D attention mask shorter than the input length with + # cache. In that case, the 4D attention mask attends to the newest tokens only. + if attention_mask.shape[-2] < cache_position[0] + sequence_length: + logger.warning_once( + "Passing a 4d mask shorter than the input length is deprecated and will be removed in " + "transformers v4.42.0" + ) + offset = cache_position[0] + else: + offset = 0 + mask_shape = attention_mask.shape + mask_slice = (attention_mask.eq(0.0)).to(dtype=dtype) * min_dtype + causal_mask[ + : mask_shape[0], : mask_shape[1], offset : mask_shape[2] + offset, : mask_shape[3] + ] = mask_slice + + if ( + self.config._attn_implementation == "sdpa" + and attention_mask is not None + and attention_mask.device.type == "cuda" + ): + # 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 + causal_mask = AttentionMaskConverter._unmask_unattended(causal_mask, min_dtype) + + return causal_mask + +class PersimmonForCausalLM(PersimmonPreTrainedModel): + _tied_weights_keys = ["lm_head.weight"] + + def __init__(self, config): + super().__init__(config) + self.model = PersimmonModel(config) + self.vocab_size = config.vocab_size + self.lm_head = nn.Linear(config.hidden_size, config.vocab_size, bias=False) + + # Initialize weights and apply final processing + self.post_init() + + def get_input_embeddings(self): + return self.model.embed_tokens + + def set_input_embeddings(self, value): + self.model.embed_tokens = value + + def get_output_embeddings(self): + return self.lm_head + + def set_output_embeddings(self, new_embeddings): + self.lm_head = new_embeddings + + def set_decoder(self, decoder): + self.model = decoder + + def get_decoder(self): + return self.model + + @add_start_docstrings_to_model_forward(LLAMA_INPUTS_DOCSTRING) + @replace_return_docstrings(output_type=CausalLMOutputWithPast, config_class=_CONFIG_FOR_DOC) + def forward( + self, + input_ids: torch.LongTensor = None, + attention_mask: Optional[torch.Tensor] = None, + position_ids: Optional[torch.LongTensor] = 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, + ) -> Union[Tuple, CausalLMOutputWithPast]: + 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]`. + + Returns: + + Example: + + ```python + >>> from transformers import AutoTokenizer, PersimmonForCausalLM + + >>> model = PersimmonForCausalLM.from_pretrained("meta-llama/Persimmon-2-7b-hf") + >>> tokenizer = AutoTokenizer.from_pretrained("meta-llama/Persimmon-2-7b-hf") + + >>> 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_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, + return_dict=return_dict, + cache_position=cache_position, + ) + + hidden_states = outputs[0] + if self.config.pretraining_tp > 1: + lm_head_slices = self.lm_head.weight.split(self.vocab_size // self.config.pretraining_tp, dim=0) + logits = [F.linear(hidden_states, lm_head_slices[i]) for i in range(self.config.pretraining_tp)] + logits = torch.cat(logits, dim=-1) + else: + logits = self.lm_head(hidden_states) + logits = logits.float() + + loss = None + 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, + ) + + def prepare_inputs_for_generation( + self, + input_ids, + past_key_values=None, + attention_mask=None, + inputs_embeds=None, + cache_position=None, + use_cache=True, + **kwargs, + ): + past_length = 0 + if past_key_values is not None: + if isinstance(past_key_values, Cache): + past_length = cache_position[0] if cache_position is not None else past_key_values.get_seq_length() + max_cache_length = ( + torch.tensor(past_key_values.get_max_length(), device=input_ids.device) + if past_key_values.get_max_length() is not None + else None + ) + cache_length = past_length if max_cache_length is None else torch.min(max_cache_length, past_length) + # TODO joao: remove this `else` after `generate` prioritizes `Cache` objects + else: + cache_length = past_length = past_key_values[0][0].shape[2] + max_cache_length = None + + # Keep only the unprocessed tokens: + # 1 - If the length of the attention_mask exceeds the length of input_ids, then we are in a setting where + # some of the inputs are exclusively passed as part of the cache (e.g. when passing input_embeds as input) + if attention_mask is not None and attention_mask.shape[1] > input_ids.shape[1]: + input_ids = input_ids[:, -(attention_mask.shape[1] - past_length) :] + # 2 - If the past_length is smaller than input_ids', then input_ids holds all input tokens. We can discard + # input_ids based on the past_length. + elif past_length < input_ids.shape[1]: + input_ids = input_ids[:, past_length:] + # 3 - Otherwise (past_length >= input_ids.shape[1]), let's assume input_ids only has unprocessed tokens. + + # If we are about to go beyond the maximum cache length, we need to crop the input attention mask. + if ( + max_cache_length is not None + and attention_mask is not None + and cache_length + input_ids.shape[1] > max_cache_length + ): + attention_mask = attention_mask[:, -max_cache_length:] + + position_ids = kwargs.get("position_ids", None) + if attention_mask is not None and position_ids is None: + # create position_ids on the fly for batch generation + position_ids = attention_mask.long().cumsum(-1) - 1 + position_ids.masked_fill_(attention_mask == 0, 1) + if past_key_values: + position_ids = position_ids[:, -input_ids.shape[1] :] + + # if `inputs_embeds` are passed, we only want to use them in the 1st generation step + if inputs_embeds is not None and past_key_values is None: + model_inputs = {"inputs_embeds": inputs_embeds} + else: + # The `contiguous()` here is necessary to have a static stride during decoding. torchdynamo otherwise + # recompiles graphs as the stride of the inputs is a guard. Ref: https://github.com/huggingface/transformers/pull/29114 + # TODO: use `next_tokens` directly instead. + model_inputs = {"input_ids": input_ids.contiguous()} + + input_length = position_ids.shape[-1] if position_ids is not None else input_ids.shape[-1] + if cache_position is None: + cache_position = torch.arange(past_length, past_length + input_length, device=input_ids.device) + elif use_cache: + cache_position = cache_position[-input_length:] + + model_inputs.update( + { + "position_ids": position_ids, + "cache_position": cache_position, + "past_key_values": past_key_values, + "use_cache": use_cache, + "attention_mask": attention_mask, + } + ) + return model_inputs + + @staticmethod + def _reorder_cache(past_key_values, beam_idx): + reordered_past = () + for layer_past in past_key_values: + reordered_past += ( + tuple(past_state.index_select(0, beam_idx.to(past_state.device)) for past_state in layer_past), + ) + return reordered_past diff --git a/src/transformers/models/stablelm/modeling_stablelm.py b/src/transformers/models/stablelm/modeling_stablelm.py index 3da496ce78..bc2673e6d5 100755 --- a/src/transformers/models/stablelm/modeling_stablelm.py +++ b/src/transformers/models/stablelm/modeling_stablelm.py @@ -113,7 +113,39 @@ class StableLmLayerNormPerHead(nn.Module): class StableLmAttention(nn.Module): """Multi-headed attention from 'Attention Is All You Need' paper""" def __init__(self, config: LlamaConfig, layer_idx: int | None = None): - super().__init__(config, layer_idx) # here call to super means + config, layer_idx + def __init__(self, config: LlamaConfig, layer_idx: Optional[int] = None): + super().__init__() + self.config = config + self.layer_idx = layer_idx + if layer_idx is None: + logger.warning_once( + f"Instantiating {self.__class__.__name__} without passing a `layer_idx` is not recommended and will " + "lead to errors during the forward call if caching is used. Please make sure to provide a `layer_idx` " + "when creating this class." + ) + + self.attention_dropout = config.attention_dropout + self.hidden_size = config.hidden_size + self.num_heads = config.num_attention_heads + self.head_dim = self.hidden_size // self.num_heads + self.num_key_value_heads = config.num_key_value_heads + self.num_key_value_groups = self.num_heads // self.num_key_value_heads + self.max_position_embeddings = config.max_position_embeddings + self.rope_theta = config.rope_theta + self.is_causal = True + + if (self.head_dim * self.num_heads) != self.hidden_size: + raise ValueError( + f"hidden_size must be divisible by num_heads (got `hidden_size`: {self.hidden_size}" + f" and `num_heads`: {self.num_heads})." + ) + + self.q_proj = nn.Linear(self.hidden_size, self.num_heads * self.head_dim, bias=config.attention_bias) + self.k_proj = nn.Linear(self.hidden_size, self.num_key_value_heads * self.head_dim, bias=config.attention_bias) + self.v_proj = nn.Linear(self.hidden_size, self.num_key_value_heads * self.head_dim, bias=config.attention_bias) + self.o_proj = nn.Linear(self.hidden_size, self.hidden_size, bias=config.attention_bias) + self._init_rope() # here call to super means # we should copy super self.qk_layernorm = config.qk_layernorm self.q_layernorm = StableLmLayerNormPerHead(self.head_dim, self.num_heads, eps=config.layer_norm_eps) @@ -377,102 +409,7 @@ class StableLmSdpaAttention(LlamaAttention): "StableLmModel is using StableLmSdpaAttention, but `torch.nn.functional.scaled_dot_product_attention` does not support `output_attentions=True`. Falling back to the manual attention implementation, " 'but specifying the manual implementation will be required from Transformers version v5.0.0 onwards. This warning can be removed using the argument `attn_implementation="eager"` when loading the model.' ) - 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: bool = False, - use_cache: bool = False, - ) -> Tuple[torch.Tensor, Optional[torch.Tensor], Optional[Tuple[torch.Tensor]]]: - 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) - - 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_key_value_heads, self.head_dim).transpose(1, 2) - value_states = value_states.view(bsz, q_len, self.num_key_value_heads, self.head_dim).transpose(1, 2) - - if self.qk_layernorm: - query_states = self.q_layernorm(query_states) - key_states = self.k_layernorm(key_states) - - kv_seq_len = key_states.shape[-2] - if past_key_value is not None: - if self.layer_idx is None: - raise ValueError( - f"The cache structure has changed since version v4.36. If you are using {self.__class__.__name__} " - "for auto-regressive decoding with k/v caching, please make sure to initialize the attention class " - "with a layer index." - ) - kv_seq_len += past_key_value.get_usable_length(kv_seq_len, self.layer_idx) - cos, sin = self.rotary_emb(value_states, seq_len=kv_seq_len) - - # Partial rotary embedding - query_rot, query_pass = ( - query_states[..., : self.rotary_emb.dim], - query_states[..., self.rotary_emb.dim :], - ) - key_rot, key_pass = ( - key_states[..., : self.rotary_emb.dim], - key_states[..., self.rotary_emb.dim :], - ) - # [batch_size, seq_length, num_heads, head_dim // config.partial_rotary_factor] - query_rot, key_rot = self.rotary_emb(query_rot, key_rot, cos, sin, position_ids) - - # [batch_size, seq_length, num_heads, head_dim] - query_states = torch.cat((query_rot, query_pass), dim=-1) - key_states = torch.cat((key_rot, key_pass), dim=-1) - - if past_key_value is not None: - # Specific to RoPE models with partial rotation - cache_kwargs = {"sin": sin, "cos": cos, "partial_rotation_size": self.rotary_emb.dim} - key_states, value_states = past_key_value.update(key_states, value_states, self.layer_idx, cache_kwargs) - - # Repeat k/v heads if n_kv_heads < n_heads - key_states = repeat_kv(key_states, self.num_key_value_groups) - value_states = repeat_kv(value_states, self.num_key_value_groups) - - attn_weights = torch.matmul(query_states, key_states.transpose(2, 3)) / math.sqrt(self.head_dim) - - if attn_weights.size() != (bsz, self.num_heads, q_len, kv_seq_len): - raise ValueError( - f"Attention weights should be of size {(bsz, self.num_heads, q_len, kv_seq_len)}, but is" - f" {attn_weights.size()}" - ) - - if attention_mask is not None: - if attention_mask.size() != (bsz, 1, q_len, kv_seq_len): - raise ValueError( - f"Attention mask should be of size {(bsz, 1, q_len, kv_seq_len)}, but is {attention_mask.size()}" - ) - attn_weights = attn_weights + attention_mask - - # upcast attention to fp32 - attn_weights = nn.functional.softmax(attn_weights, dtype=torch.float32, dim=-1).to(query_states.dtype) - attn_weights = self.attention_dropout(attn_weights) - - attn_output = torch.matmul(attn_weights, value_states) - - if attn_output.size() != (bsz, self.num_heads, q_len, self.head_dim): - raise ValueError( - f"`attn_output` should be of size {(bsz, self.num_heads, q_len, self.head_dim)}, but is" - f" {attn_output.size()}" - ) - - attn_output = attn_output.transpose(1, 2).contiguous() - attn_output = attn_output.reshape(bsz, q_len, self.hidden_size) - - attn_output = self.o_proj(attn_output) - - if not output_attentions: - attn_weights = None - - return attn_output, attn_weights, past_key_value - + return super().forward( hidden_states=hidden_states, attention_mask=attention_mask, position_ids=position_ids, @@ -654,3 +591,944 @@ class StableLmSdpaAttention(LlamaAttention): return attn_output, None, past_key_value class StableLmFlashAttention2(LlamaAttention): + """ + Llama flash attention module. This module inherits from `LlamaAttention` 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, *args, **kwargs): + super().__init__(*args, **kwargs) + + # 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[Cache] = 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_key_value_heads, self.head_dim).transpose(1, 2) + value_states = value_states.view(bsz, q_len, self.num_key_value_heads, self.head_dim).transpose(1, 2) + + if self.qk_layernorm: + query_states = self.q_layernorm(query_states) + key_states = self.k_layernorm(key_states) + + kv_seq_len = key_states.shape[-2] + if past_key_value is not None: + if self.layer_idx is None: + raise ValueError( + f"The cache structure has changed since version v4.36. If you are using {self.__class__.__name__} " + "for auto-regressive decoding with k/v caching, please make sure to initialize the attention class " + "with a layer index." + ) + kv_seq_len += past_key_value.get_usable_length(kv_seq_len, self.layer_idx) + cos, sin = self.rotary_emb(value_states, seq_len=kv_seq_len) + + # Partial rotary embedding + query_rot, query_pass = ( + query_states[..., : self.rotary_emb.dim], + query_states[..., self.rotary_emb.dim :], + ) + key_rot, key_pass = ( + key_states[..., : self.rotary_emb.dim], + key_states[..., self.rotary_emb.dim :], + ) + query_rot, key_rot = self.rotary_emb(query_rot, key_rot, cos, sin, position_ids) + + # [batch_size, seq_length, num_heads, head_dim] + query_states = torch.cat((query_rot, query_pass), dim=-1) + key_states = torch.cat((key_rot, key_pass), dim=-1) + + if past_key_value is not None: + cache_kwargs = {"sin": sin, "cos": cos, "partial_rotation_size": self.rotary_emb.dim} + key_states, value_states = past_key_value.update(key_states, value_states, self.layer_idx, cache_kwargs) + + # 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.attention_dropout.p if self.training else 0.0 + + 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.hidden_size).contiguous() + attn_output = self.o_proj(attn_output) + + if not output_attentions: + attn_weights = None + + return attn_output, attn_weights, past_key_value + + 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 (`float`): + 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 + 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), + ) + + """ + StableLM flash attention module. This module inherits from `StableLmAttention` 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. + """ + """ + StableLM flash attention module. This module inherits from `StableLmAttention` 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 forward( + self, + hidden_states: torch.Tensor, + attention_mask: Optional[torch.LongTensor] = None, + position_ids: Optional[torch.LongTensor] = None, + past_key_value: Optional[Cache] = 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_key_value_heads, self.head_dim).transpose(1, 2) + value_states = value_states.view(bsz, q_len, self.num_key_value_heads, self.head_dim).transpose(1, 2) + + if self.qk_layernorm: + query_states = self.q_layernorm(query_states) + key_states = self.k_layernorm(key_states) + + kv_seq_len = key_states.shape[-2] + if past_key_value is not None: + if self.layer_idx is None: + raise ValueError( + f"The cache structure has changed since version v4.36. If you are using {self.__class__.__name__} " + "for auto-regressive decoding with k/v caching, please make sure to initialize the attention class " + "with a layer index." + ) + kv_seq_len += past_key_value.get_usable_length(kv_seq_len, self.layer_idx) + cos, sin = self.rotary_emb(value_states, seq_len=kv_seq_len) + + # Partial rotary embedding + query_rot, query_pass = ( + query_states[..., : self.rotary_emb.dim], + query_states[..., self.rotary_emb.dim :], + ) + key_rot, key_pass = ( + key_states[..., : self.rotary_emb.dim], + key_states[..., self.rotary_emb.dim :], + ) + query_rot, key_rot = self.rotary_emb(query_rot, key_rot, cos, sin, position_ids) + + # [batch_size, seq_length, num_heads, head_dim] + query_states = torch.cat((query_rot, query_pass), dim=-1) + key_states = torch.cat((key_rot, key_pass), dim=-1) + + if past_key_value is not None: + cache_kwargs = {"sin": sin, "cos": cos, "partial_rotation_size": self.rotary_emb.dim} + key_states, value_states = past_key_value.update(key_states, value_states, self.layer_idx, cache_kwargs) + + # 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.attention_dropout.p if self.training else 0.0 + + 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.hidden_size).contiguous() + attn_output = self.o_proj(attn_output) + + if not output_attentions: + attn_weights = None + + return attn_output, attn_weights, past_key_value + + +StableLm_ATTENTION_CLASSES = {"eager": StableLmAttention, "flash_attention_2": StableLmFlashAttention2, "sdpa": StableLmSdpaAttention} + +class StableLmDecoderLayer(nn.Module): + def __init__(self, config: StableLmConfig, layer_idx: int): + super().__init__() + self.use_parallel_residual = config.use_parallel_residual + self.hidden_size = config.hidden_size + self.self_attn = StableLm_ATTENTION_CLASSES[config._attn_implementation](config, layer_idx=layer_idx) + self.mlp = StableLmMLP(config) + self.input_layernorm = nn.LayerNorm(config.hidden_size, eps=config.layer_norm_eps) + self.post_attention_layernorm = None + if not self.use_parallel_residual: + self.post_attention_layernorm = nn.LayerNorm(config.hidden_size, eps=config.layer_norm_eps) + self.dropout = nn.Dropout(config.hidden_dropout) + + def forward( + self, + hidden_states: torch.Tensor, + attention_mask: Optional[torch.Tensor] = None, + position_ids: Optional[torch.LongTensor] = None, + past_key_value: Optional[Tuple[torch.Tensor]] = None, + output_attentions: Optional[bool] = False, + use_cache: Optional[bool] = False, + ) -> Tuple[torch.FloatTensor, Optional[Tuple[torch.FloatTensor, torch.FloatTensor]]]: + residual = hidden_states + + hidden_states = self.input_layernorm(hidden_states) + + # Self Attention + self_attn_output, self_attn_weights, present_key_value = 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, + ) + + # copied from transformers.models.gpt_neox.modeling_gpt_neox.GPTNeoXLayer.forward + if self.use_parallel_residual: + # x = x + attn(ln1(x)) + mlp(ln1(x)) + # Fully Connected + mlp_output = self.mlp(hidden_states) + mlp_output = self.dropout(mlp_output) + hidden_states = residual + self_attn_output + mlp_output + else: + # x = x + attn(ln1(x)) + # x = x + mlp(ln2(x)) + residual = residual + self_attn_output + # Fully Connected + mlp_output = self.mlp(self.post_attention_layernorm(residual)) + mlp_output = self.dropout(mlp_output) + hidden_states = residual + mlp_output + + outputs = (hidden_states,) + + if output_attentions: + outputs += (self_attn_weights,) + + if use_cache: + outputs += (present_key_value,) + + return outputs + +@add_start_docstrings( + "The bare LLaMA Model outputting raw hidden-states without any specific head on top.", + LLAMA_START_DOCSTRING, +) +class StableLmPreTrainedModel(PreTrainedModel): + config_class = StableConfig + base_model_prefix = "model" + supports_gradient_checkpointing = True + _no_split_modules = ["StableDecoderLayer"] + _skip_keys_device_placement = ["past_key_values"] + _supports_flash_attn_2 = True + _supports_sdpa = True + _supports_cache_class = 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_() + +@add_start_docstrings( + "The bare LLaMA Model outputting raw hidden-states without any specific head on top.", + LLAMA_START_DOCSTRING, +) +class StableLmdModel(StablePreTrainedModel): + """ + Transformer decoder consisting of *config.num_hidden_layers* layers. Each layer is a [`StableDecoderLayer`] + + Args: + config: StableConfig + """ + + def __init__(self, config: StableConfig): + super().__init__(config) + self.padding_idx = config.pad_token_id + self.vocab_size = config.vocab_size + + self.embed_tokens = nn.Embedding(config.vocab_size, config.hidden_size, self.padding_idx) + self.layers = nn.ModuleList( + [StableDecoderLayer(config, layer_idx) for layer_idx in range(config.num_hidden_layers)] + ) + self.norm = StableRMSNorm(config.hidden_size, eps=config.rms_norm_eps) + 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 + + @add_start_docstrings_to_model_forward(LLAMA_INPUTS_DOCSTRING) + def forward( + self, + input_ids: torch.LongTensor = None, + attention_mask: Optional[torch.Tensor] = None, + position_ids: Optional[torch.LongTensor] = None, + past_key_values: Optional[Union[Cache, List[torch.FloatTensor]]] = 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, + ) -> Union[Tuple, BaseModelOutputWithPast]: + 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 + ) + use_cache = use_cache if use_cache is not None else self.config.use_cache + return_dict = return_dict if return_dict is not None else self.config.use_return_dict + + if (input_ids is None) ^ (inputs_embeds is not None): + raise ValueError( + "You cannot specify both input_ids and inputs_embeds at the same time, and must specify either one" + ) + + if self.gradient_checkpointing and self.training and use_cache: + logger.warning_once( + "`use_cache=True` is incompatible with gradient checkpointing. Setting `use_cache=False`." + ) + use_cache = False + + if inputs_embeds is None: + inputs_embeds = self.embed_tokens(input_ids) + + return_legacy_cache = False + if use_cache and not isinstance(past_key_values, Cache): # kept for BC (non `Cache` `past_key_values` inputs) + return_legacy_cache = True + past_key_values = DynamicCache.from_legacy_cache(past_key_values) + + 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) + + # embed positions + hidden_states = inputs_embeds + + # decoder layers + all_hidden_states = () if output_hidden_states else None + all_self_attns = () if output_attentions else None + next_decoder_cache = None + + for decoder_layer in self.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, + ) + 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, + ) + + hidden_states = layer_outputs[0] + + if use_cache: + next_decoder_cache = layer_outputs[2 if output_attentions else 1] + + if output_attentions: + all_self_attns += (layer_outputs[1],) + + hidden_states = self.norm(hidden_states) + + # add hidden states from the last decoder layer + if output_hidden_states: + all_hidden_states += (hidden_states,) + + next_cache = next_decoder_cache if use_cache else None + if return_legacy_cache: + next_cache = next_cache.to_legacy_cache() + + if not return_dict: + return tuple(v for v in [hidden_states, next_cache, all_hidden_states, all_self_attns] if v is not None) + return BaseModelOutputWithPast( + last_hidden_state=hidden_states, + past_key_values=next_cache, + hidden_states=all_hidden_states, + attentions=all_self_attns, + ) + + def _update_causal_mask( + self, + attention_mask: torch.Tensor, + input_tensor: torch.Tensor, + cache_position: torch.Tensor, + past_key_values: Cache, + ): + # TODO: As of torch==2.2.0, the `attention_mask` passed to the model in `generate` is 2D and of dynamic length even when the static + # KV cache is used. This is an issue for torch.compile which then recaptures cudagraphs at each decode steps due to the dynamic shapes. + # (`recording cudagraph tree for symint key 13`, etc.), which is VERY slow. A workaround is `@torch.compiler.disable`, but this prevents using + # `fullgraph=True`. See more context in https://github.com/huggingface/transformers/pull/29114 + + if self.config._attn_implementation == "flash_attention_2": + if attention_mask is not None and 0.0 in attention_mask: + 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) + if self.config._attn_implementation == "sdpa" and not using_static_cache: + 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 + min_dtype = torch.finfo(dtype).min + sequence_length = input_tensor.shape[1] + if using_static_cache: + target_length = past_key_values.get_max_length() + else: + target_length = ( + attention_mask.shape[-1] + if isinstance(attention_mask, torch.Tensor) + else past_seen_tokens + sequence_length + 1 + ) + + 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(input_tensor.shape[0], 1, -1, -1) + if attention_mask is not None: + causal_mask = causal_mask.clone() # copy to contiguous memory for in-place edit + if attention_mask.dim() == 2: + 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 + ) + elif attention_mask.dim() == 4: + # backwards compatibility: we allow passing a 4D attention mask shorter than the input length with + # cache. In that case, the 4D attention mask attends to the newest tokens only. + if attention_mask.shape[-2] < cache_position[0] + sequence_length: + logger.warning_once( + "Passing a 4d mask shorter than the input length is deprecated and will be removed in " + "transformers v4.42.0" + ) + offset = cache_position[0] + else: + offset = 0 + mask_shape = attention_mask.shape + mask_slice = (attention_mask.eq(0.0)).to(dtype=dtype) * min_dtype + causal_mask[ + : mask_shape[0], : mask_shape[1], offset : mask_shape[2] + offset, : mask_shape[3] + ] = mask_slice + + if ( + self.config._attn_implementation == "sdpa" + and attention_mask is not None + and attention_mask.device.type == "cuda" + ): + # 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 + causal_mask = AttentionMaskConverter._unmask_unattended(causal_mask, min_dtype) + + return causal_mask + +class StableLmForCausalLM(StablePreTrainedModel): + _tied_weights_keys = ["lm_head.weight"] + + def __init__(self, config): + super().__init__(config) + self.model = StableModel(config) + self.vocab_size = config.vocab_size + self.lm_head = nn.Linear(config.hidden_size, config.vocab_size, bias=False) + + # Initialize weights and apply final processing + self.post_init() + + def get_input_embeddings(self): + return self.model.embed_tokens + + def set_input_embeddings(self, value): + self.model.embed_tokens = value + + def get_output_embeddings(self): + return self.lm_head + + def set_output_embeddings(self, new_embeddings): + self.lm_head = new_embeddings + + def set_decoder(self, decoder): + self.model = decoder + + def get_decoder(self): + return self.model + + @add_start_docstrings_to_model_forward(LLAMA_INPUTS_DOCSTRING) + @replace_return_docstrings(output_type=CausalLMOutputWithPast, config_class=_CONFIG_FOR_DOC) + def forward( + self, + input_ids: torch.LongTensor = None, + attention_mask: Optional[torch.Tensor] = None, + position_ids: Optional[torch.LongTensor] = 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, + ) -> Union[Tuple, CausalLMOutputWithPast]: + 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]`. + + Returns: + + Example: + + ```python + >>> from transformers import AutoTokenizer, StableLmForCausalLM + + >>> model = StableLmForCausalLM.from_pretrained("meta-llama/Stable-2-7b-hf") + >>> tokenizer = AutoTokenizer.from_pretrained("meta-llama/Stable-2-7b-hf") + + >>> 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_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, + return_dict=return_dict, + cache_position=cache_position, + ) + + hidden_states = outputs[0] + if self.config.pretraining_tp > 1: + lm_head_slices = self.lm_head.weight.split(self.vocab_size // self.config.pretraining_tp, dim=0) + logits = [F.linear(hidden_states, lm_head_slices[i]) for i in range(self.config.pretraining_tp)] + logits = torch.cat(logits, dim=-1) + else: + logits = self.lm_head(hidden_states) + logits = logits.float() + + loss = None + 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, + ) + + def prepare_inputs_for_generation( + self, + input_ids, + past_key_values=None, + attention_mask=None, + inputs_embeds=None, + cache_position=None, + use_cache=True, + **kwargs, + ): + past_length = 0 + if past_key_values is not None: + if isinstance(past_key_values, Cache): + past_length = cache_position[0] if cache_position is not None else past_key_values.get_seq_length() + max_cache_length = ( + torch.tensor(past_key_values.get_max_length(), device=input_ids.device) + if past_key_values.get_max_length() is not None + else None + ) + cache_length = past_length if max_cache_length is None else torch.min(max_cache_length, past_length) + # TODO joao: remove this `else` after `generate` prioritizes `Cache` objects + else: + cache_length = past_length = past_key_values[0][0].shape[2] + max_cache_length = None + + # Keep only the unprocessed tokens: + # 1 - If the length of the attention_mask exceeds the length of input_ids, then we are in a setting where + # some of the inputs are exclusively passed as part of the cache (e.g. when passing input_embeds as input) + if attention_mask is not None and attention_mask.shape[1] > input_ids.shape[1]: + input_ids = input_ids[:, -(attention_mask.shape[1] - past_length) :] + # 2 - If the past_length is smaller than input_ids', then input_ids holds all input tokens. We can discard + # input_ids based on the past_length. + elif past_length < input_ids.shape[1]: + input_ids = input_ids[:, past_length:] + # 3 - Otherwise (past_length >= input_ids.shape[1]), let's assume input_ids only has unprocessed tokens. + + # If we are about to go beyond the maximum cache length, we need to crop the input attention mask. + if ( + max_cache_length is not None + and attention_mask is not None + and cache_length + input_ids.shape[1] > max_cache_length + ): + attention_mask = attention_mask[:, -max_cache_length:] + + position_ids = kwargs.get("position_ids", None) + if attention_mask is not None and position_ids is None: + # create position_ids on the fly for batch generation + position_ids = attention_mask.long().cumsum(-1) - 1 + position_ids.masked_fill_(attention_mask == 0, 1) + if past_key_values: + position_ids = position_ids[:, -input_ids.shape[1] :] + + # if `inputs_embeds` are passed, we only want to use them in the 1st generation step + if inputs_embeds is not None and past_key_values is None: + model_inputs = {"inputs_embeds": inputs_embeds} + else: + # The `contiguous()` here is necessary to have a static stride during decoding. torchdynamo otherwise + # recompiles graphs as the stride of the inputs is a guard. Ref: https://github.com/huggingface/transformers/pull/29114 + # TODO: use `next_tokens` directly instead. + model_inputs = {"input_ids": input_ids.contiguous()} + + input_length = position_ids.shape[-1] if position_ids is not None else input_ids.shape[-1] + if cache_position is None: + cache_position = torch.arange(past_length, past_length + input_length, device=input_ids.device) + elif use_cache: + cache_position = cache_position[-input_length:] + + model_inputs.update( + { + "position_ids": position_ids, + "cache_position": cache_position, + "past_key_values": past_key_values, + "use_cache": use_cache, + "attention_mask": attention_mask, + } + ) + return model_inputs + + @staticmethod + def _reorder_cache(past_key_values, beam_idx): + reordered_past = () + for layer_past in past_key_values: + reordered_past += ( + tuple(past_state.index_select(0, beam_idx.to(past_state.device)) for past_state in layer_past), + ) + return reordered_past + +@add_start_docstrings( + """ + The LLaMa Model transformer with a sequence classification head on top (linear layer). + + [`StableLmForSequenceClassification`] uses the last token in order to do the classification, as other causal models + (e.g. GPT-2) do. + + Since it does classification on the last token, it requires to know the position of the last token. If a + `pad_token_id` is defined in the configuration, it finds the last token that is not a padding token in each row. If + no `pad_token_id` is defined, it simply takes the last value in each row of the batch. Since it cannot guess the + padding tokens when `inputs_embeds` are passed instead of `input_ids`, it does the same (take the last value in + each row of the batch). + """, + LLAMA_START_DOCSTRING, +) +class StableLmForSequenceClassification(StablePreTrainedModel): + def __init__(self, config): + super().__init__(config) + self.num_labels = config.num_labels + self.model = StableModel(config) + self.score = nn.Linear(config.hidden_size, self.num_labels, bias=False) + + # Initialize weights and apply final processing + self.post_init() + + def get_input_embeddings(self): + return self.model.embed_tokens + + def set_input_embeddings(self, value): + self.model.embed_tokens = value + + @add_start_docstrings_to_model_forward(LLAMA_INPUTS_DOCSTRING) + def forward( + self, + input_ids: torch.LongTensor = None, + attention_mask: Optional[torch.Tensor] = None, + position_ids: Optional[torch.LongTensor] = 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, + ) -> Union[Tuple, SequenceClassifierOutputWithPast]: + r""" + labels (`torch.LongTensor` of shape `(batch_size,)`, *optional*): + Labels for computing the sequence classification/regression loss. Indices should be in `[0, ..., + config.num_labels - 1]`. If `config.num_labels == 1` a regression loss is computed (Mean-Square loss), If + `config.num_labels > 1` a classification loss is computed (Cross-Entropy). + """ + return_dict = return_dict if return_dict is not None else self.config.use_return_dict + + transformer_outputs = self.model( + 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, + return_dict=return_dict, + ) + hidden_states = transformer_outputs[0] + logits = self.score(hidden_states) + + if input_ids is not None: + batch_size = input_ids.shape[0] + else: + batch_size = inputs_embeds.shape[0] + + if self.config.pad_token_id is None and batch_size != 1: + raise ValueError("Cannot handle batch sizes > 1 if no padding token is defined.") + if self.config.pad_token_id is None: + sequence_lengths = -1 + else: + if input_ids is not None: + # if no pad token found, use modulo instead of reverse indexing for ONNX compatibility + sequence_lengths = torch.eq(input_ids, self.config.pad_token_id).int().argmax(-1) - 1 + sequence_lengths = sequence_lengths % input_ids.shape[-1] + sequence_lengths = sequence_lengths.to(logits.device) + else: + sequence_lengths = -1 + + pooled_logits = logits[torch.arange(batch_size, device=logits.device), sequence_lengths] + + loss = None + if labels is not None: + labels = labels.to(logits.device) + if self.config.problem_type is None: + if self.num_labels == 1: + self.config.problem_type = "regression" + elif self.num_labels > 1 and (labels.dtype == torch.long or labels.dtype == torch.int): + self.config.problem_type = "single_label_classification" + else: + self.config.problem_type = "multi_label_classification" + + if self.config.problem_type == "regression": + loss_fct = MSELoss() + if self.num_labels == 1: + loss = loss_fct(pooled_logits.squeeze(), labels.squeeze()) + else: + loss = loss_fct(pooled_logits, labels) + elif self.config.problem_type == "single_label_classification": + loss_fct = CrossEntropyLoss() + loss = loss_fct(pooled_logits.view(-1, self.num_labels), labels.view(-1)) + elif self.config.problem_type == "multi_label_classification": + loss_fct = BCEWithLogitsLoss() + loss = loss_fct(pooled_logits, labels) + if not return_dict: + output = (pooled_logits,) + transformer_outputs[1:] + return ((loss,) + output) if loss is not None else output + + return SequenceClassifierOutputWithPast( + loss=loss, + logits=pooled_logits, + past_key_values=transformer_outputs.past_key_values, + hidden_states=transformer_outputs.hidden_states, + attentions=transformer_outputs.attentions, + ) + diff --git a/src/transformers/models/starcoder2/modeling_starcoder2.py b/src/transformers/models/starcoder2/modeling_starcoder2.py index 5fcdfb48be..3cd0a15cbf 100644 --- a/src/transformers/models/starcoder2/modeling_starcoder2.py +++ b/src/transformers/models/starcoder2/modeling_starcoder2.py @@ -127,7 +127,39 @@ class Starcoder2MLP(nn.Module): class Starcoder2Attention(nn.Module): """Multi-headed attention from 'Attention Is All You Need' paper""" def __init__(self, config: LlamaConfig, layer_idx: int | None = None): - super().__init__(config, layer_idx) # here call to super means + config, layer_idx + def __init__(self, config: LlamaConfig, layer_idx: Optional[int] = None): + super().__init__() + self.config = config + self.layer_idx = layer_idx + if layer_idx is None: + logger.warning_once( + f"Instantiating {self.__class__.__name__} without passing a `layer_idx` is not recommended and will " + "lead to errors during the forward call if caching is used. Please make sure to provide a `layer_idx` " + "when creating this class." + ) + + self.attention_dropout = config.attention_dropout + self.hidden_size = config.hidden_size + self.num_heads = config.num_attention_heads + self.head_dim = self.hidden_size // self.num_heads + self.num_key_value_heads = config.num_key_value_heads + self.num_key_value_groups = self.num_heads // self.num_key_value_heads + self.max_position_embeddings = config.max_position_embeddings + self.rope_theta = config.rope_theta + self.is_causal = True + + if (self.head_dim * self.num_heads) != self.hidden_size: + raise ValueError( + f"hidden_size must be divisible by num_heads (got `hidden_size`: {self.hidden_size}" + f" and `num_heads`: {self.num_heads})." + ) + + self.q_proj = nn.Linear(self.hidden_size, self.num_heads * self.head_dim, bias=config.attention_bias) + self.k_proj = nn.Linear(self.hidden_size, self.num_key_value_heads * self.head_dim, bias=config.attention_bias) + self.v_proj = nn.Linear(self.hidden_size, self.num_key_value_heads * self.head_dim, bias=config.attention_bias) + self.o_proj = nn.Linear(self.hidden_size, self.hidden_size, bias=config.attention_bias) + self._init_rope() # here call to super means self.attention_dropout = config.attention_dropout def _init_rope(self): if self.config.rope_scaling is None: @@ -673,7 +705,21 @@ class LlamaModel(LlamaPreTrainedModel): config: LlamaConfig """ def __init__(self, config): - super().__init__(config) + config + def __init__(self, config: LlamaConfig): + super().__init__(config) + self.padding_idx = config.pad_token_id + self.vocab_size = config.vocab_size + + self.embed_tokens = nn.Embedding(config.vocab_size, config.hidden_size, self.padding_idx) + self.layers = nn.ModuleList( + [LlamaDecoderLayer(config, layer_idx) for layer_idx in range(config.num_hidden_layers)] + ) + self.norm = LlamaRMSNorm(config.hidden_size, eps=config.rms_norm_eps) + self.gradient_checkpointing = False + + # Initialize weights and apply final processing + self.post_init() self.embedding_dropout = config.embedding_dropout def get_input_embeddings(self): return self.embed_tokens @@ -684,116 +730,118 @@ class LlamaModel(LlamaPreTrainedModel): inputs_embeds = self.embed_tokens(input_ids) hidden_states = inputs_embeds hidden_states = nn.functional.dropout(hidden_states, p=self.embedding_dropout, training=self.training) - @add_start_docstrings_to_model_forward(LLAMA_INPUTS_DOCSTRING) - def forward( - self, - input_ids: torch.LongTensor = None, - attention_mask: Optional[torch.Tensor] = None, - position_ids: Optional[torch.LongTensor] = None, - past_key_values: Optional[Union[Cache, List[torch.FloatTensor]]] = 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, - ) -> Union[Tuple, BaseModelOutputWithPast]: - 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 - ) - use_cache = use_cache if use_cache is not None else self.config.use_cache - return_dict = return_dict if return_dict is not None else self.config.use_return_dict - - if (input_ids is None) ^ (inputs_embeds is not None): - raise ValueError( - "You cannot specify both input_ids and inputs_embeds at the same time, and must specify either one" + return (None, attention_mask, position_ids, past_key_values, inputs_embeds, use_cache, output_attentions, output_hidden_states, return_dict, cache_position + @add_start_docstrings_to_model_forward(LLAMA_INPUTS_DOCSTRING) + def forward( + self, + input_ids: torch.LongTensor = None, + attention_mask: Optional[torch.Tensor] = None, + position_ids: Optional[torch.LongTensor] = None, + past_key_values: Optional[Union[Cache, List[torch.FloatTensor]]] = 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, + ) -> Union[Tuple, BaseModelOutputWithPast]: + 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 ) + use_cache = use_cache if use_cache is not None else self.config.use_cache + return_dict = return_dict if return_dict is not None else self.config.use_return_dict - if self.gradient_checkpointing and self.training and use_cache: - logger.warning_once( - "`use_cache=True` is incompatible with gradient checkpointing. Setting `use_cache=False`." - ) - use_cache = False + if (input_ids is None) ^ (inputs_embeds is not None): + raise ValueError( + "You cannot specify both input_ids and inputs_embeds at the same time, and must specify either one" + ) - if inputs_embeds is None: - inputs_embeds = self.embed_tokens(input_ids) + if self.gradient_checkpointing and self.training and use_cache: + logger.warning_once( + "`use_cache=True` is incompatible with gradient checkpointing. Setting `use_cache=False`." + ) + use_cache = False - return_legacy_cache = False - if use_cache and not isinstance(past_key_values, Cache): # kept for BC (non `Cache` `past_key_values` inputs) - return_legacy_cache = True - past_key_values = DynamicCache.from_legacy_cache(past_key_values) + if inputs_embeds is None: + inputs_embeds = self.embed_tokens(input_ids) - 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) + return_legacy_cache = False + if use_cache and not isinstance(past_key_values, Cache): # kept for BC (non `Cache` `past_key_values` inputs) + return_legacy_cache = True + past_key_values = DynamicCache.from_legacy_cache(past_key_values) - causal_mask = self._update_causal_mask(attention_mask, inputs_embeds, cache_position, past_key_values) + 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) - # embed positions - hidden_states = inputs_embeds + causal_mask = self._update_causal_mask(attention_mask, inputs_embeds, cache_position, past_key_values) - # decoder layers - all_hidden_states = () if output_hidden_states else None - all_self_attns = () if output_attentions else None - next_decoder_cache = None + # embed positions + hidden_states = inputs_embeds - for decoder_layer in self.layers: + # decoder layers + all_hidden_states = () if output_hidden_states else None + all_self_attns = () if output_attentions else None + next_decoder_cache = None + + for decoder_layer in self.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, + ) + 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, + ) + + hidden_states = layer_outputs[0] + + if use_cache: + next_decoder_cache = layer_outputs[2 if output_attentions else 1] + + if output_attentions: + all_self_attns += (layer_outputs[1],) + + hidden_states = self.norm(hidden_states) + + # add hidden states from the last decoder layer 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, - ) - 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, - ) + next_cache = next_decoder_cache if use_cache else None + if return_legacy_cache: + next_cache = next_cache.to_legacy_cache() - hidden_states = layer_outputs[0] - - if use_cache: - next_decoder_cache = layer_outputs[2 if output_attentions else 1] - - if output_attentions: - all_self_attns += (layer_outputs[1],) - - hidden_states = self.norm(hidden_states) - - # add hidden states from the last decoder layer - if output_hidden_states: - all_hidden_states += (hidden_states,) - - next_cache = next_decoder_cache if use_cache else None - if return_legacy_cache: - next_cache = next_cache.to_legacy_cache() - - if not return_dict: - return tuple(v for v in [hidden_states, next_cache, all_hidden_states, all_self_attns] if v is not None) - return BaseModelOutputWithPast( - last_hidden_state=hidden_states, - past_key_values=next_cache, - hidden_states=all_hidden_states, - attentions=all_self_attns, - )None, attention_mask, position_ids, past_key_values, inputs_embeds, use_cache, output_attentions, output_hidden_states, return_dict, cache_position) + if not return_dict: + return tuple(v for v in [hidden_states, next_cache, all_hidden_states, all_self_attns] if v is not None) + return BaseModelOutputWithPast( + last_hidden_state=hidden_states, + past_key_values=next_cache, + hidden_states=all_hidden_states, + attentions=all_self_attns, + ) + ) def _update_causal_mask( self, diff --git a/utils/convert_diff_model.py b/utils/convert_diff_model.py index 588dbc556a..579e923e2a 100644 --- a/utils/convert_diff_model.py +++ b/utils/convert_diff_model.py @@ -17,7 +17,19 @@ import inspect # with open(CohereConverter.original_file, 'r') as file, open("result.py", "w+") as modeling: # pass # TODO also copy and import from all modules in CohereConverter.modules_to_import to be able to use inspect -import sys +def replace_super_calls_in_method(method_body, parent_method_body, method_name): + # Indent parent method body to match the child's method indentation + indent = re.match(r'(\s*)def ', method_body).group(1) + indented_parent_method_body = "\n".join([indent + line if line.strip() else line for line in parent_method_body.split('\n')]) + method_name = method_name.strip() + # Handle super().method_name(args) and return super().method_name(args) + super_call_pattern = re.compile(r'(\s*)return super\(\)\.' + method_name + r'\((.*?)\)') + method_body = super_call_pattern.sub(r'\1return (\2\n' + indented_parent_method_body + r'\1)', method_body) + + super_call_pattern_no_return = re.compile(r'(\s*)super\(\)\.' + method_name + r'\((.*?)\)') + method_body = super_call_pattern_no_return.sub(r'\1\2\n' + indented_parent_method_body, method_body) + + return method_body # 2. Write all the classes. Use the `CohereConverter` class for this. def create_single_model_file(converter): model_identifier = converter.diff_file.split("diff_") @@ -71,7 +83,15 @@ def create_single_model_file(converter): # TODO handle call to super! full_function = match.group() if "def" in full_function: - child_function_set[full_function.split("def")[1].split("(")[0]] = re.sub(r" return super\(\).forward\(", parent_function_set.get(function_name,""), full_function) + function_name = full_function.split("def")[1].split("(")[0] + if f"super()." in full_function or f"return super()." in full_function: + replaced_function = replace_super_calls_in_method(full_function, + parent_function_set.get(function_name, + ""), + function_name) + child_function_set[function_name] = replaced_function + else: + child_function_set[function_name] = full_function else: child_function_set[full_function] = full_function