huggingface/text-generation-inference
5
0
Fork 0
mirror of https://github.com/huggingface/text-generation-inference.git synced 2025-09-09 19:34:53 +00:00
Code Issues Packages Projects Releases Wiki Activity

formatting

Browse Source
This commit is contained in:
OlivierDehaene 2023-02-01 11:48:18 +01:00
parent 1d0fa38cb8
commit 3149317fa1
1 changed files with 21 additions and 17 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

8
server/text_generation/models/gpt_neox.py
View File

@ -239,10 +239,14 @@ class GPTNeoxSharded(GPTNeox):
# Logits are sharded, so we need to gather them # Logits are sharded, so we need to gather them
logits = [torch.empty_like(outputs.logits) for _ in range(self.world_size)] logits = [torch.empty_like(outputs.logits) for _ in range(self.world_size)]
torch.distributed.all_gather(logits, outputs.logits, group=self.process_group) torch.distributed.all_gather(
logits, outputs.logits, group=self.process_group
)
logits = torch.cat(logits, dim=2) logits = torch.cat(logits, dim=2)
return logits, outputs.past_key_values return logits, outputs.past_key_values
# While the model itself is sharded, the embeddings might not as they might not be dividable by num-shard # While the model itself is sharded, the embeddings might not as they might not be dividable by num-shard
else: else:
return super(GPTNeoxSharded, self).forward(input_ids, attention_mask, position_ids, past_key_values) return super(GPTNeoxSharded, self).forward(
input_ids, attention_mask, position_ids, past_key_values
)