Ring attention #181
Open
Ring attention #181
Conversation
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
3outeille
reviewed
Jul 24, 2024
Comment on lines
131
to
+197
|
|
||
| ## Copy from transformers. Non interleaved version of RoPE. Will be refactored later | ||
| def rotate_half(x): | ||
| """Rotates half the hidden dims of the input.""" | ||
| x1 = x[..., : x.shape[-1] // 2] | ||
| x2 = x[..., x.shape[-1] // 2 :] | ||
| return torch.cat((-x2, x1), dim=-1) | ||
|
|
||
|
|
||
| class LlamaRotaryEmbedding(nn.Module): | ||
| def __init__(self, dim: int, end: int, theta: float = 500000.0): | ||
| super().__init__() | ||
| self.dim = dim | ||
| self.end = end | ||
| self.theta = theta | ||
| self.init_rotary_embeddings() | ||
|
|
||
| def init_rotary_embeddings(self): | ||
| inv_freq = 1.0 / (self.theta ** (torch.arange(0, self.dim, 2, dtype=torch.float, device="cuda") / self.dim)) | ||
| self.register_buffer("inv_freq", inv_freq, persistent=False) | ||
|
|
||
| @torch.no_grad() | ||
| def forward( | ||
| self, | ||
| x: torch.Tensor, # [batch_size, seq_length, num_heads, d_qk] | ||
| position_ids: Optional[torch.LongTensor], # [batch_size, seq_length] | ||
| ): | ||
| # x: [bs, num_attention_heads, seq_len, head_size] | ||
| # print("rotary") | ||
| 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) | ||
| cos = emb.cos() | ||
| sin = emb.sin() | ||
| return cos.to(dtype=x.dtype), sin.to(dtype=x.dtype) | ||
|
|
||
|
|
||
| def apply_rotary_pos_emb(q, k, cos, sin, unsqueeze_dim=2): | ||
| """Applies Rotary Position Embedding to the query and key tensors. | ||
|
|
||
| Args: | ||
| q (`torch.Tensor`): The query tensor. | ||
| k (`torch.Tensor`): The key tensor. | ||
| cos (`torch.Tensor`): The cosine part of the rotary embedding. | ||
| sin (`torch.Tensor`): The sine part of the rotary embedding. | ||
| unsqueeze_dim (`int`, *optional*, defaults to 1): | ||
| The 'unsqueeze_dim' argument specifies the dimension along which to unsqueeze cos[position_ids] and | ||
| sin[position_ids] so that they can be properly broadcasted to the dimensions of q and k. For example, note | ||
| that cos[position_ids] and sin[position_ids] have the shape [batch_size, seq_len, head_dim]. Then, if q and | ||
| k have the shape [batch_size, heads, seq_len, head_dim], then setting unsqueeze_dim=1 makes | ||
| cos[position_ids] and sin[position_ids] broadcastable to the shapes of q and k. Similarly, if q and k have | ||
| the shape [batch_size, seq_len, heads, head_dim], then set unsqueeze_dim=2. | ||
| Returns: | ||
| `tuple(torch.Tensor)` comprising of the query and key tensors rotated using the Rotary Position Embedding. | ||
| """ | ||
| cos = cos.unsqueeze(unsqueeze_dim) | ||
| sin = sin.unsqueeze(unsqueeze_dim) | ||
| q_embed = (q * cos) + (rotate_half(q) * sin) | ||
| k_embed = (k * cos) + (rotate_half(k) * sin) | ||
| return q_embed, k_embed | ||
|
|
There was a problem hiding this comment.
can we open a separate PR first to replace current RotaryEmbedding
3outeille
reviewed
Jul 24, 2024
| init_distributed(tp=tp, dp=dp, pp=pp)(_test_save_zero_optimizer_and_load_optimizer)(test_context=test_context) | ||
| # Currently SP doesn't support zero. | ||
| if sp != 1: | ||
| return |
3outeille
reviewed
Jul 24, 2024
| # We use DP=2 as we're interested in testing that one | ||
| init_distributed(tp=tp, dp=dp, pp=pp)(_test_save_zero_optimizer_and_load_data_parallel_optimizer)( | ||
| if sp != 1: | ||
| return |
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
Ring attention for training on long sequences. Similar to Megatron context parallel. Idea from https://github.com/zhuzilin/ring-flash-attention