xinyun/vllm
1
0
Fork 0
mirror of https://git.datalinker.icu/vllm-project/vllm.git synced 2025-12-10 01:05:01 +08:00
Code Issues Packages Projects Releases Wiki Activity

[FIX] Minor bug fixes (#1035)

Browse Source 
* [FIX] Minor bug fixes

* Address review comments
This commit is contained in:
Zhuohan Li 2023-09-13 16:38:12 -07:00 committed by GitHub
parent ab019eea75
commit f04908cae7
No known key found for this signature in database
GPG Key ID: 4AEE18F83AFDEB23
2 changed files with 4 additions and 3 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

5
vllm/model_executor/layers/sampler.py
View File

@ -82,8 +82,9 @@ class Sampler(nn.Module):
# We use float32 for probabilities and log probabilities. # We use float32 for probabilities and log probabilities.
# Compute the probabilities. # Compute the probabilities.
probs = torch.softmax(logits, dim=-1, dtype=torch.float) probs = torch.softmax(logits, dim=-1, dtype=torch.float)
# Compute the log probabilities (before applying top-p and top-k). # Compute the log probabilities.
logprobs = torch.log(probs) # Use log_softmax to ensure numerical stability.
logprobs = torch.log_softmax(logits, dim=-1, dtype=torch.float)
# Sample the next tokens. # Sample the next tokens.
return _sample(probs, logprobs, input_metadata) return _sample(probs, logprobs, input_metadata)

2
vllm/sequence.py
View File

@ -350,7 +350,7 @@ class SequenceOutputs:
def __eq__(self, other: object) -> bool: def __eq__(self, other: object) -> bool:
if not isinstance(other, SequenceOutputs): if not isinstance(other, SequenceOutputs):
return NotImplementedError() raise NotImplementedError()
return (self.parent_seq_id == other.parent_seq_id return (self.parent_seq_id == other.parent_seq_id
and self.output_token == other.output_token and self.output_token == other.output_token
and self.logprobs == other.logprobs) and self.logprobs == other.logprobs)