huggingface/text-generation-inference
5
0
Fork 0
mirror of https://github.com/huggingface/text-generation-inference.git synced 2025-09-10 11:54:52 +00:00
Code Issues Packages Projects Releases Wiki Activity

optimize argmax

Browse Source
This commit is contained in:
OlivierDehaene 2023-05-24 16:28:16 +02:00
parent c59fb353a0
commit a62f14872e
1 changed files with 2 additions and 1 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

3
server/text_generation_server/utils/tokens.py
View File

@ -296,7 +296,8 @@ class HeterogeneousSampling:
def __call__(self, logits): def __call__(self, logits):
out = torch.empty(logits.shape[0], dtype=torch.int64, device=logits.device) out = torch.empty(logits.shape[0], dtype=torch.int64, device=logits.device)
if self.greedy_indices: if self.greedy_indices:
out[self.greedy_indices] = torch.argmax(logits[self.greedy_indices], -1) # Computing for all indices is faster than slicing
torch.argmax(logits, -1, out=out)
for i, sampling in self.sampling_mapping.items(): for i, sampling in self.sampling_mapping.items():
out[i] = sampling(logits[i]) out[i] = sampling(logits[i])