tools : enable kvu in perplexity for hellaswag, winogrande, multiple-choice (#19954)

llama-perplexity -hf unsloth/Qwen3-0.6B-GGUF:Q4_K_M -f winogrande-debiased-eval.csv --winogrande

    winogrande_score : tokenizing selected tasks
    winogrande_score : calculating winogrande score over selected tasks.
    split_equal: sequential split is not supported when there are coupled sequences in the input batch (you may need to use the -kvu flag)
    decode: failed to find a memory slot for batch of size 46
    failed to decode the batch, n_batch = 2048, ret = 1
    winogrande_score: llama_decode() failed

same for hellaswag:

    split_equal: sequential split is not supported when there are coupled sequences in the input batch (you may need to use the -kvu flag)
    decode: failed to find a memory slot for batch of size 99
    failed to decode the batch, n_batch = 2048, ret = 1
    hellaswag_score: llama_decode() failed

Signed-off-by: Adrien Gallouët <redacted>

diff --git a/tools/perplexity/perplexity.cpp b/tools/perplexity/perplexity.cpp
index cc5ea99c4dfe05161a378812809654393fe82107..0eb062f05dd3f8a9a3b92ab94c3392a68378ab5a 100644 (file)
--- a/tools/perplexity/perplexity.cpp
+++ b/tools/perplexity/perplexity.cpp
@@ -2025,21 +2025,14 @@ int main(int argc, char ** argv) {
         return 1;
     }
 
-    const bool ppl = !params.hellaswag && !params.winogrande && !params.multiple_choice && !params.kl_divergence;
-
-    if (ppl || params.kl_divergence) {
-        const int32_t n_seq = std::max(1, params.n_batch / n_ctx);
-        const int32_t n_kv = n_seq * n_ctx;
-
-        params.n_parallel = n_seq;
-        params.n_ctx      = n_kv;
-
-        params.n_batch = std::min(params.n_batch, n_kv);
-    } else {
-        params.n_batch = std::min(params.n_batch, params.n_ctx);
-        // ensure there's at least enough seq_ids for HellaSwag
+    if (params.hellaswag || params.winogrande || params.multiple_choice) {
         params.n_parallel = std::max(4, params.n_parallel);
+        params.kv_unified = true;
+    } else { // Perplexity & KL divergence
+        params.n_parallel = std::max(1, params.n_batch / n_ctx);
     }
+    params.n_ctx = params.n_parallel * n_ctx;
+    params.n_batch = std::min(params.n_batch, params.n_ctx);
 
     if (params.ppl_stride > 0) {
         LOG_INF("Will perform strided perplexity calculation -> adjusting context size from %d to %d\n",