context : fail upon construction if sequences exceed max value

ggerganov · ggerganov · commit 43b40d3f32dd · 2025-05-24T13:36:52.000+03:00
ggml-ci
diff --git a/src/llama-context.cpp b/src/llama-context.cpp
@@ -27,8 +27,7 @@ llama_context::llama_context(
 
     cparams.n_seq_max = std::max(1u, params.n_seq_max);
     if (cparams.n_seq_max > LLAMA_MAX_PARALLEL_SEQUENCES) {
-        LLAMA_LOG_WARN("%s: n_seq_max (%d) is larger than the maximum supported (%d) - clamping\n", __func__, cparams.n_seq_max, LLAMA_MAX_PARALLEL_SEQUENCES);
-        cparams.n_seq_max = LLAMA_MAX_PARALLEL_SEQUENCES;
+        throw std::runtime_error("n_seq_max must be <= " + std::to_string(LLAMA_MAX_PARALLEL_SEQUENCES));
     }
 
     cparams.n_threads        = params.n_threads;

Original file line number	Diff line number	Diff line change
`@@ -27,8 +27,7 @@ llama_context::llama_context(`
`27`	`27`
`28`	`28`	`cparams.n_seq_max = std::max(1u, params.n_seq_max);`
`29`	`29`	`if (cparams.n_seq_max > LLAMA_MAX_PARALLEL_SEQUENCES) {`
`30`		`- LLAMA_LOG_WARN("%s: n_seq_max (%d) is larger than the maximum supported (%d) - clamping\n", __func__, cparams.n_seq_max, LLAMA_MAX_PARALLEL_SEQUENCES);`
`31`		`- cparams.n_seq_max = LLAMA_MAX_PARALLEL_SEQUENCES;`
	`30`	`+ throw std::runtime_error("n_seq_max must be <= " + std::to_string(LLAMA_MAX_PARALLEL_SEQUENCES));`
`32`	`31`	`}`
`33`	`32`
`34`	`33`	`cparams.n_threads = params.n_threads;`