mirror of
https://github.com/hpcaitech/ColossalAI.git
synced 2025-09-23 10:30:03 +00:00
[upgrade]Upgrade transformers (#6320)
* fix for async io * test for upgrading transformers * add ci machine * fix * fix * fix * fix * fix * [pre-commit.ci] auto fixes from pre-commit.com hooks for more information, see https://pre-commit.ci * fix * [pre-commit.ci] auto fixes from pre-commit.com hooks for more information, see https://pre-commit.ci * Update test_fp16_torch.py * Update build_on_pr.yml * fix * [pre-commit.ci] auto fixes from pre-commit.com hooks for more information, see https://pre-commit.ci * fix * [pre-commit.ci] auto fixes from pre-commit.com hooks for more information, see https://pre-commit.ci * fix * [pre-commit.ci] auto fixes from pre-commit.com hooks for more information, see https://pre-commit.ci * fix * fix * [pre-commit.ci] auto fixes from pre-commit.com hooks for more information, see https://pre-commit.ci * fix * fix * [pre-commit.ci] auto fixes from pre-commit.com hooks for more information, see https://pre-commit.ci * fix * [pre-commit.ci] auto fixes from pre-commit.com hooks for more information, see https://pre-commit.ci * fix * [pre-commit.ci] auto fixes from pre-commit.com hooks for more information, see https://pre-commit.ci * fix * fix * fix * [pre-commit.ci] auto fixes from pre-commit.com hooks for more information, see https://pre-commit.ci * fix * [pre-commit.ci] auto fixes from pre-commit.com hooks for more information, see https://pre-commit.ci * fix * fiux * fix * fix * fix * upgrade llama * fix * fix * [pre-commit.ci] auto fixes from pre-commit.com hooks for more information, see https://pre-commit.ci * fix * fix * upgrade_bert * upgrade_bloom * [upgrade] upgrade gpt2 (#6291) * fix * fix * [pre-commit.ci] auto fixes from pre-commit.com hooks for more information, see https://pre-commit.ci * fix * fix --------- Co-authored-by: pre-commit-ci[bot] <66853113+pre-commit-ci[bot]@users.noreply.github.com> * upgrade command * [pre-commit.ci] auto fixes from pre-commit.com hooks for more information, see https://pre-commit.ci * fix * add explanation * [pre-commit.ci] auto fixes from pre-commit.com hooks for more information, see https://pre-commit.ci * fix * fix * fix * fix * [upgrade]Upgrade qwen2 (#6302) * upgrade qwen2 * fix * [pre-commit.ci] auto fixes from pre-commit.com hooks for more information, see https://pre-commit.ci --------- Co-authored-by: pre-commit-ci[bot] <66853113+pre-commit-ci[bot]@users.noreply.github.com> * update_bloom * fix * add explantion * [pre-commit.ci] auto fixes from pre-commit.com hooks for more information, see https://pre-commit.ci * upgrade_sam * add the explanation * upgrade_t * fix * fix * fix * upgrade_gptj * fix * fix * [pre-commit.ci] auto fixes from pre-commit.com hooks for more information, see https://pre-commit.ci * [upgrade]upgrade opt (#6307) * upgrade opt * fix * [upgrade]Upgrade mixtral (#6317) * upgrade mixtral * fix * [pre-commit.ci] auto fixes from pre-commit.com hooks for more information, see https://pre-commit.ci * fix * [pre-commit.ci] auto fixes from pre-commit.com hooks for more information, see https://pre-commit.ci * upgrade infer * [pre-commit.ci] auto fixes from pre-commit.com hooks for more information, see https://pre-commit.ci * fix * [pre-commit.ci] auto fixes from pre-commit.com hooks for more information, see https://pre-commit.ci * fix * upgrade drafter * [pre-commit.ci] auto fixes from pre-commit.com hooks for more information, see https://pre-commit.ci * fix * upgrade lazy * [pre-commit.ci] auto fixes from pre-commit.com hooks for more information, see https://pre-commit.ci * upgrade mixtral --------- Co-authored-by: pre-commit-ci[bot] <66853113+pre-commit-ci[bot]@users.noreply.github.com> * [upgrade]Upgrade vit (#6308) * fix * fix * fix rotate embedding test * [pre-commit.ci] auto fixes from pre-commit.com hooks for more information, see https://pre-commit.ci --------- Co-authored-by: pre-commit-ci[bot] <66853113+pre-commit-ci[bot]@users.noreply.github.com> * [upgrade]upgrade mistral (#6296) * upgrade mistral * fix * [pre-commit.ci] auto fixes from pre-commit.com hooks for more information, see https://pre-commit.ci --------- Co-authored-by: pre-commit-ci[bot] <66853113+pre-commit-ci[bot]@users.noreply.github.com> * fix * [pre-commit.ci] auto fixes from pre-commit.com hooks for more information, see https://pre-commit.ci * fix * fix falcon * fix * Update test_shard_deepseek.py * Update build_on_pr.yml * Update requirements.txt * fix (#6327) * fix (#6328) * [pre-commit.ci] auto fixes from pre-commit.com hooks for more information, see https://pre-commit.ci * Update bert.py * fix (#6329) --------- Co-authored-by: pre-commit-ci[bot] <66853113+pre-commit-ci[bot]@users.noreply.github.com> Co-authored-by: Hanks <hangxu0304@gmail.com> Co-authored-by: wangbluo <2538539015@qq.com> Co-authored-by: Wang Binluo <32676639+wangbluo@users.noreply.github.com>
This commit is contained in:
flybird11111
GitHub
@@ -128,7 +128,7 @@ class OPTPipelineForwards:
|
||||
# required mask seq length can be calculated via length of past
|
||||
mask_seq_length = past_key_values_length + seq_length
|
||||
# embed positions
|
||||
if self.decoder._use_flash_attention_2:
|
||||
if self.decoder.config._attn_implementation == "flash_attention_2":
|
||||
# 2d mask is passed through the layers
|
||||
causal_attention_mask = attention_mask if (attention_mask is not None and 0 in attention_mask) else None
|
||||
attention_mask = (
|
||||
@@ -542,6 +542,9 @@ class OPTPipelineForwards:
|
||||
def get_opt_flash_attention_forward(shard_config: ShardConfig):
|
||||
from transformers.models.opt.modeling_opt import OPTAttention
|
||||
|
||||
def _shape(tensor: torch.Tensor, seq_len: int, bsz: int, num_heads: int, head_dim: int):
|
||||
return tensor.view(bsz, seq_len, num_heads, head_dim).transpose(1, 2).contiguous()
|
||||
|
||||
def forward(
|
||||
self: OPTAttention,
|
||||
hidden_states: torch.Tensor,
|
||||
@@ -568,30 +571,20 @@ def get_opt_flash_attention_forward(shard_config: ShardConfig):
|
||||
value_states = past_key_value[1]
|
||||
elif is_cross_attention:
|
||||
# cross_attentions
|
||||
key_states = self._shape(self.k_proj(key_value_states), -1, bsz)
|
||||
value_states = self._shape(self.v_proj(key_value_states), -1, bsz)
|
||||
key_states = _shape(self.k_proj(key_value_states), -1, bsz, self.num_heads, self.head_dim)
|
||||
value_states = _shape(self.v_proj(key_value_states), -1, bsz, self.num_heads, self.head_dim)
|
||||
elif past_key_value is not None:
|
||||
# reuse k, v, self_attention
|
||||
key_states = self._shape(self.k_proj(hidden_states), -1, bsz)
|
||||
value_states = self._shape(self.v_proj(hidden_states), -1, bsz)
|
||||
key_states = _shape(self.k_proj(hidden_states), -1, bsz, self.num_heads, self.head_dim)
|
||||
value_states = _shape(self.v_proj(hidden_states), -1, bsz, self.num_heads, self.head_dim)
|
||||
key_states = torch.cat([past_key_value[0], key_states], dim=2)
|
||||
value_states = torch.cat([past_key_value[1], value_states], dim=2)
|
||||
else:
|
||||
# self_attention
|
||||
key_states = self._shape(self.k_proj(hidden_states), -1, bsz)
|
||||
value_states = self._shape(self.v_proj(hidden_states), -1, bsz)
|
||||
key_states = _shape(self.k_proj(hidden_states), -1, bsz, self.num_heads, self.head_dim)
|
||||
value_states = _shape(self.v_proj(hidden_states), -1, bsz, self.num_heads, self.head_dim)
|
||||
|
||||
if self.is_decoder:
|
||||
# if cross_attention save Tuple(torch.Tensor, torch.Tensor) of all cross attention key/value_states.
|
||||
# Further calls to cross_attention layer can then reuse all cross-attention
|
||||
# key/value_states (first "if" case)
|
||||
# if uni-directional self-attention (decoder) save Tuple(torch.Tensor, torch.Tensor) of
|
||||
# all previous decoder key/value_states. Further calls to uni-directional self-attention
|
||||
# can concat previous decoder key/value_states to current projected key/value_states (third "elif" case)
|
||||
# if encoder bi-directional self-attention `past_key_value` is always `None`
|
||||
past_key_value = (key_states, value_states)
|
||||
|
||||
query_states = self._shape(query_states, tgt_len, bsz)
|
||||
query_states = _shape(query_states, tgt_len, bsz, self.num_heads, self.head_dim)
|
||||
|
||||
dropout_p = self.dropout if self.training else 0.0
|
||||
attn_output = ColoAttention.attention(
|
||||
@@ -630,6 +623,8 @@ def get_opt_decoder_forward_for_flash_attention(shard_config: ShardConfig):
|
||||
output_attentions: Optional[bool] = None,
|
||||
output_hidden_states: Optional[bool] = None,
|
||||
return_dict: Optional[bool] = None,
|
||||
position_ids: Optional[torch.LongTensor] = None,
|
||||
cache_position: Optional[torch.Tensor] = None,
|
||||
) -> Union[Tuple, BaseModelOutputWithPast]:
|
||||
output_attentions = output_attentions if output_attentions is not None else self.config.output_attentions
|
||||
output_hidden_states = (
|
||||
|
||||
Reference in New Issue
Block a user