mirror of
https://github.com/hpcaitech/ColossalAI.git
synced 2025-09-02 01:28:31 +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
@@ -48,6 +48,7 @@ def _get_attention_mask(
|
||||
sp_mode = shard_config.sequence_parallelism_mode
|
||||
# If a 2D or 3D attention mask is provided for the cross-attention
|
||||
# we need to make broadcastable to [batch_size, num_heads, seq_length, seq_length]
|
||||
|
||||
if self.config.add_cross_attention and encoder_hidden_states is not None:
|
||||
assert not sp_mode == "ring_attn", "Ring Attention only supports decoder-only."
|
||||
encoder_batch_size, encoder_sequence_length, _ = encoder_hidden_states.size()
|
||||
@@ -55,7 +56,7 @@ def _get_attention_mask(
|
||||
encoder_attention_mask = ColoAttention.prepare_attn_kwargs(
|
||||
(encoder_batch_size, 1, seq_len, encoder_sequence_length),
|
||||
dtype=hidden_states.dtype,
|
||||
dtype2=encoder_hidden_states.dtype,
|
||||
device=encoder_hidden_states.device,
|
||||
q_padding_mask=attention_mask,
|
||||
kv_padding_mask=encoder_attention_mask,
|
||||
)
|
||||
@@ -77,7 +78,6 @@ def _get_attention_mask(
|
||||
if shard_config.enable_flash_attention:
|
||||
if attention_mask is not None:
|
||||
attention_mask = attention_mask.view(batch_size, -1)
|
||||
|
||||
attention_mask = ColoAttention.prepare_attn_kwargs(
|
||||
(batch_size, 1, seq_len, seq_len + past_key_values_length),
|
||||
hidden_states.dtype,
|
||||
@@ -835,9 +835,12 @@ def get_gpt2_flash_attention_forward(shard_config: Optional[ShardConfig] = None)
|
||||
attention_mask = encoder_attention_mask
|
||||
else:
|
||||
query, key, value = self.c_attn(hidden_states).split(self.split_size, dim=2)
|
||||
query = self._split_heads(query, self.num_heads, self.head_dim)
|
||||
key = self._split_heads(key, self.num_heads, self.head_dim)
|
||||
value = self._split_heads(value, self.num_heads, self.head_dim)
|
||||
|
||||
shape_q = (*query.shape[:-1], -1, self.head_dim)
|
||||
shape_kv = (*key.shape[:-1], -1, self.head_dim)
|
||||
query = query.view(shape_q).transpose(1, 2)
|
||||
key = key.view(shape_kv).transpose(1, 2)
|
||||
value = value.view(shape_kv).transpose(1, 2)
|
||||
|
||||
if layer_past is not None:
|
||||
past_key, past_value = layer_past
|
||||
@@ -871,7 +874,9 @@ def get_gpt2_flash_attention_forward(shard_config: Optional[ShardConfig] = None)
|
||||
)
|
||||
else:
|
||||
attn_output = ColoAttention.attention(query, key, value, **attention_mask, dropout_p=dropout_p, scale=scale)
|
||||
attn_output = self._merge_heads(attn_output, self.num_heads, self.head_dim)
|
||||
|
||||
attn_output = attn_output.permute(0, 2, 1, 3).contiguous()
|
||||
attn_output = attn_output.reshape(*attn_output.shape[:-2], -1).contiguous()
|
||||
attn_output = self.c_proj(attn_output)
|
||||
attn_output = self.resid_dropout(attn_output)
|
||||
outputs = (attn_output, present, None)
|
||||
|
||||
Reference in New Issue
Block a user