EngineX/xc-llm-kunlun
3
0
Fork 0
Code Issues Pull Requests Actions Projects Releases Wiki Activity

Merge pull request #29 from chenyili0619/28-v0110-only-enble-top-p-or-k-occur-error

Browse Source 
[Bug] Fixed the issue where an error occurred when the request includ…
This commit is contained in:
Xinyu Dong
2025-12-18 13:04:39 +08:00
committed by GitHub
parent 6b5740ad0a 2e2933d217
commit ff7131678a
1 changed files with 2 additions and 2 deletions
Download Patch File Download Diff File Expand all files Collapse all files

4
vllm_kunlun/v1/sample/ops/topk_topp_sampler.py
View File

@@ -42,7 +42,7 @@ class TopKTopPSampler(nn.Module):
""" """
logits = apply_top_k_top_p(logits, k, p) logits = apply_top_k_top_p(logits, k, p)
probs = logits.softmax(dim=-1, dtype=torch.float32) probs = logits.softmax(dim=-1, dtype=torch.float32)
return random_sample(probs, generators) return random_sample(probs, generators), None
def forward_kunlun( def forward_kunlun(
self, self,
@@ -199,4 +199,4 @@ def flashinfer_sample(
next_token_ids = xtorch_ops.top_k_top_p_sampling_from_probs( next_token_ids = xtorch_ops.top_k_top_p_sampling_from_probs(
probs, top_k=k, top_p=p, deterministic=True) probs, top_k=k, top_p=p, deterministic=True)
return next_token_ids.view(-1) return next_token_ids.view(-1)