Update sampling.cpp

Olivier Chafik · Olivier Chafik · commit 31cc6df9bfd3 · 2024-05-20T19:37:22.000+01:00
diff --git a/common/sampling.cpp b/common/sampling.cpp
@@ -191,7 +191,7 @@ static llama_token llama_sampling_sample_impl(
 
     std::vector<float> original_logits;
     auto cur_p = llama_sampling_prepare(ctx_sampling, ctx_main, ctx_cfg, idx, /* apply_grammar= */ is_resampling, &original_logits);
-    if (!is_resampling) {
+    if (ctx_sampling->grammar != NULL && !is_resampling) {
         GGML_ASSERT(!original_logits.empty());
     }
     llama_token id = 0;

Original file line number	Diff line number	Diff line change
`@@ -191,7 +191,7 @@ static llama_token llama_sampling_sample_impl(`
`191`	`191`
`192`	`192`	`std::vector<float> original_logits;`
`193`	`193`	`auto cur_p = llama_sampling_prepare(ctx_sampling, ctx_main, ctx_cfg, idx, /* apply_grammar= */ is_resampling, &original_logits);`
`194`		`- if (!is_resampling) {`
	`194`	`+ if (ctx_sampling->grammar != NULL && !is_resampling) {`
`195`	`195`	`GGML_ASSERT(!original_logits.empty());`
`196`	`196`	`}`
`197`	`197`	`llama_token id = 0;`