huggingface/text-generation-inference
5
0
Fork 0
mirror of https://github.com/huggingface/text-generation-inference.git synced 2025-06-21 16:40:20 +00:00
Code Issues Packages Projects Releases Wiki Activity

fix: check existance before accessing rope type in cuda warmup

Browse Source
This commit is contained in:
drbh 2025-01-28 22:54:34 +00:00
parent 79a2c956de
commit d0e2332d17
1 changed files with 5 additions and 2 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

7
server/text_generation_server/models/flash_causal_lm.py
View File

@ -1401,8 +1401,11 @@ class FlashCausalLM(Model):
if max_bs is None:
input_ids = torch.zeros(bs, dtype=torch.int64, device=self.device)
position_ids = torch.zeros(bs, dtype=torch.int32, device=self.device)
# mrope have position_ids per section, if so repeat n times
if self.model.config.rope_scaling["rope_type"] == "mrope":
if ( # mrope have position_ids per section, if so repeat n times
hasattr(self.model, "config")
and hasattr(self.model.config, "rope_scaling")
and self.model.config.rope_scaling["rope_type"] == "mrope"
):
n_sections = len(self.model.config.rope_scaling["mrope_section"])
position_ids = position_ids.unsqueeze(1).repeat(1, n_sections)
slots = torch.arange(bs, dtype=torch.int64, device=self.device)