server : fix SWA condition for full context reprocess (#14163)

ggerganov · web-flow · commit ffad04397399 · 2025-06-13T11:18:25.000+03:00
ggml-ci
diff --git a/tools/server/server.cpp b/tools/server/server.cpp
@@ -3217,7 +3217,7 @@ struct server_context {
                                 }
 
                                 const auto n_swa = llama_model_n_swa(model);
-                                if (pos_min > slot.n_past - n_swa) {
+                                if (pos_min > std::max(0, slot.n_past - n_swa)) {
                                     SLT_WRN(slot, "n_past = %d, cache_tokens.size() = %d, seq_id = %d, pos_min = %d, n_swa = %d\n", slot.n_past, (int) slot.cache_tokens.size(), slot.id, pos_min, n_swa);
                                     SLT_WRN(slot, "forcing full prompt re-processing due to lack of cache data (likely due to SWA, see %s)\n",
                                             "https://github.com/ggml-org/llama.cpp/pull/13194#issuecomment-2868343055");

Original file line number	Diff line number	Diff line change
`@@ -3217,7 +3217,7 @@ struct server_context {`
`3217`	`3217`	`}`
`3218`	`3218`
`3219`	`3219`	`const auto n_swa = llama_model_n_swa(model);`
`3220`		`- if (pos_min > slot.n_past - n_swa) {`
	`3220`	`+ if (pos_min > std::max(0, slot.n_past - n_swa)) {`
`3221`	`3221`	`SLT_WRN(slot, "n_past = %d, cache_tokens.size() = %d, seq_id = %d, pos_min = %d, n_swa = %d\n", slot.n_past, (int) slot.cache_tokens.size(), slot.id, pos_min, n_swa);`
`3222`	`3222`	`SLT_WRN(slot, "forcing full prompt re-processing due to lack of cache data (likely due to SWA, see %s)\n",`
`3223`	`3223`	`"https://github.com/ggml-org/llama.cpp/pull/13194#issuecomment-2868343055");`