From: Adrien Gallouët <redacted>
Date: Sun, 28 Jun 2026 10:34:11 +0000 (+0200)
Subject: common : allow --offline in llama download (#25091)
X-Git-Tag: upstream/0.0.10438~608
X-Git-Url: https://git.djapps.eu/?a=commitdiff_plain;h=c1a1c8ee94e37ff7ba2c872e783aaf7f77e0f320;p=pkg%2Fggml%2Fsources%2Fllama.cpp

common : allow --offline in llama download (#25091)

Expose the existing --offline flag to `llama download` so a script can
run it to check whether a model is already cached and ready to be served
without touching the network.

Also fix a latent use-after-free in the URL-task on_done callback:
first_path is block-scoped and was captured by reference, but invoked
after the block ends.

Signed-off-by: Adrien Gallouët <redacted>
---

diff --git a/common/arg.cpp b/common/arg.cpp
index 841ca3ce2..c289ff713 100644
--- a/common/arg.cpp
+++ b/common/arg.cpp
@@ -467,7 +467,7 @@ void common_models_handler_apply(common_models_handler & handler, common_params
         // the first part is what gets loaded, so point params.model.path at it
         if (!url_tasks.empty()) {
             std::string first_path = url_tasks.front().local_path;
-            url_tasks.front().on_done = [&]() { params.model.path = first_path; };
+            url_tasks.front().on_done = [&, first_path]() { params.model.path = first_path; };
         }
         for (auto & task : url_tasks) {
             tasks.push_back(std::move(task));
@@ -3471,7 +3471,7 @@ common_params_context common_params_parser_init(common_params & params, llama_ex
         [](common_params & params) {
             params.offline = true;
         }
-    ).set_env("LLAMA_ARG_OFFLINE"));
+    ).set_examples({LLAMA_EXAMPLE_COMMON, LLAMA_EXAMPLE_DOWNLOAD}).set_env("LLAMA_ARG_OFFLINE"));
     add_opt(common_arg(
         {"-lv", "--verbosity", "--log-verbosity"}, "N",
         string_format("Set the verbosity threshold. Messages with a higher verbosity will be ignored. Values:\n"