Merge commit 'f534da26e4ab045b6899adc07cd2b9a065355ce9' into concedo_experimental

# Conflicts:
#	.github/workflows/release.yml
#	docs/ops.md
#	docs/ops/WebGPU.csv
#	ggml/src/CMakeLists.txt
#	ggml/src/ggml-cpu/kleidiai/kleidiai.cpp
#	ggml/src/ggml-hexagon/ggml-hexagon.cpp
#	ggml/src/ggml-openvino/ggml-openvino.cpp
#	ggml/src/ggml-webgpu/ggml-webgpu-shader-lib.hpp
#	ggml/src/ggml-webgpu/ggml-webgpu.cpp
#	models/templates/deepseek-ai-DeepSeek-V4.jinja
#	tests/test-backend-ops.cpp
#	tests/test-chat-auto-parser.cpp
#	tests/test-chat.cpp
#	tests/test-llama-archs.cpp

tools/server/server-context.cpp

@@ -1152,6 +1152,11 @@ private:
             return false;
         }
 
+        if (ctx_tgt == nullptr) {
+            SRV_ERR("failed to create_context with model '%s'\n", params_base.model.path.c_str());
+            return false;
+        }
+
         vocab = llama_model_get_vocab(model_tgt);
 
         n_ctx = llama_n_ctx(ctx_tgt);

tools/server/server-stream.cpp

@@ -632,6 +632,13 @@ void server_res_spipe::on_complete() {
     if (!spipe || next_finished) {
         return;
     }
+    // an empty next_orig means set_next() never ran: the request failed before streaming
+    // started, typically a params validation throw. evict the session installed by set_req()
+    // so the failed request leaves nothing behind for discovery or replay
+    if (!next_orig) {
+        g_stream_sessions.evict(server_stream_conv_id_from_headers(req->headers));
+        return;
+    }
     std::string chunk;
     while (!spipe->is_cancelled()) {
         chunk.clear();