apps/FastDeploy
1
0
Fork 0
mirror of https://github.com/PaddlePaddle/FastDeploy.git synced 2025-10-04 16:22:57 +08:00
Code Issues Actions 2 Packages Projects Releases Wiki Activity

fix top_p_candidates and support separate setting of sampling params for mtp (#4189)
Some checks failed
CE Compile Job / ce_job_pre_check (push) Has been cancelled
Details
CE Compile Job / print_ce_job_pre_check_outputs (push) Has been cancelled
Details
CE Compile Job / FD-Clone-Linux (push) Has been cancelled
Details
CE Compile Job / Show Code Archive Output (push) Has been cancelled
Details
CE Compile Job / BUILD_SM8090 (push) Has been cancelled
Details
CE Compile Job / BUILD_SM8689 (push) Has been cancelled
Details
CE Compile Job / CE_UPLOAD (push) Has been cancelled
Details

Browse Source 
* fix top_p_candidates

* For separate setting params for mtp

* delete print

* fix
This commit is contained in:
GoldPancake
2025-09-28 11:41:20 +08:00
committed by GitHub
parent f8c6a354a1
commit 076c30cb0f
4 changed files with 29 additions and 5 deletions
Download Patch File Download Diff File Expand all files Collapse all files

12
custom_ops/gpu_ops/speculate_decoding/speculate_verify.cu
View File

@@ -38,14 +38,20 @@ __device__ int64_t topp_sampling_kernel(const int64_t *candidate_ids,
const int tid = threadIdx.x;
float sum_scores = 0.0f;
float rand_top_p = curand_uniform(dev_curand_states + tid) * topp;
for (int i = 0; i < candidate_len; i++) {
sum_scores += candidate_scores[i];
}
float tgt_topp = sum_scores < topp ? sum_scores : topp;
sum_scores = 0.0f;
float rand_top_p = curand_uniform(dev_curand_states + tid) * tgt_topp;
for (int i = 0; i < candidate_len; i++) {
sum_scores += candidate_scores[i];
if (rand_top_p <= sum_scores) {
return candidate_ids[i];
return candidate_ids[i];
}
}
return candidate_ids[0];
return candidate_ids[0];
}
__global__ void setup_kernel(curandState_t *state, const uint64_t seed,