ROCm · fredespi · Jul 22, 2026 · Jul 21, 2026
@@ -126,9 +126,11 @@ serve_timeout_secs = 90
 # On a lemonade LINUX host this needs a live lemonade serve, which reaches ready
 # then shuts down immediately (EAI-7423). os=linux: passes on native Windows
 # lemonade (run 29354305288 XPASS'd it), consistent with the other lemonade
-# serve scenarios.
+# serve scenarios. therock_family=gfx* scopes this to a REAL GPU host: the
+# scenario now also runs on the no-GPU mock lane (backed by MockServer, not a
+# real lemonade serve), where EAI-7423 does not apply and it must expect-pass.
 [["chat-tool-definitions-accepted"]]
-when = { effective_engine = "lemonade", os = "linux" }
+when = { effective_engine = "lemonade", os = "linux", therock_family = "gfx*" }
 bug = "EAI-7423"
 reason = "Lemonade managed serve reaches ready then shuts down immediately, so the chat request never reaches a live model."
 serve_timeout_secs = 90
@@ -140,9 +142,11 @@ serve_timeout_secs = 90
 # then 3/3 green re-runs), so there is NO vLLM xfail for this scenario — the CLI chat
 # path works. Only the lemonade LINUX path stays xfail: it needs a live lemonade
 # serve, which reaches ready then shuts down immediately (EAI-7423); native Windows
-# lemonade serve works (run 29338475133 XPASS).
+# lemonade serve works (run 29338475133 XPASS). therock_family=gfx* scopes this to
+# a REAL GPU host: the scenario also runs on the no-GPU mock lane (MockServer, not
+# a real lemonade serve), where EAI-7423 does not apply and it must expect-pass.
 [["chat-end-to-end-local-model"]]
-when = { effective_engine = "lemonade", os = "linux" }
+when = { effective_engine = "lemonade", os = "linux", therock_family = "gfx*" }
 bug = "EAI-7423"
 reason = "Lemonade managed serve reaches ready then shuts down immediately, so the chat request never reaches a live model."
 serve_timeout_secs = 90

@@ -41,14 +41,22 @@ Feature: Chat and endpoint detection
     When the user lists running services
     Then the served model endpoint is listed
 
-  @id:chat-tool-definitions-accepted @requires-gpu
+  # Runs on every lane: on a GPU host `a model is served in the background` does a
+  # real `rocm serve`, on the no-GPU mock lane it's backed by MockServer. The
+  # assertion (a tools-bearing request is accepted) is engine-agnostic, so no GPU
+  # is required — dropping @requires-gpu gives this per-PR mock-lane coverage.
+  @id:chat-tool-definitions-accepted
   Scenario: 5 - Chat requests that include tool definitions are accepted
     Given a managed runtime is active
     And a model is served in the background
     When a chat request with tool definitions is sent
     Then the chat response is successful
 
-  @id:chat-end-to-end-local-model @requires-gpu
+  # Runs on every lane (see scenario 5): real serve on a GPU host, MockServer on
+  # the no-GPU mock lane. Asserts only that a served model returns a non-empty
+  # reply, which is engine-agnostic — real generation is covered by the
+  # @requires-gpu serve-*-inference scenarios.
+  @id:chat-end-to-end-local-model
   Scenario: 6 - End-to-end chat through a locally served model
     Given a managed runtime is active
     And a model is served in the background

@@ -124,7 +124,9 @@ Feature: Model serving
 
   # Honest device selection: a `--gpu` index that does not exist on the host is
   # rejected outright, never silently remapped to another device (no device-0
-  # fallback). Runs on GPU hardware.
+  # fallback). Runs on GPU hardware: on a no-GPU host the GPU-required pre-flight
+  # refuses ("no usable AMD GPU") before the index is ever validated, so the
+  # index-specific rejection can only be observed where a real device is present.
   @id:serve-absent-gpu-index-rejected @requires-gpu @requires-os:linux
   Scenario: 13 - Serving pinned to a GPU that does not exist is refused
     When the user serves a model pinned to a GPU index that does not exist

@@ -20,6 +20,15 @@ async fn setup_standard_rocm(_world: &mut E2eWorld) {}
 
 #[given("a managed runtime is active")]
 async fn setup_active_runtime(world: &mut E2eWorld) {
+    // On a no-GPU host this is a no-op: a managed TheRock SDK runtime can only be
+    // installed where there's a GPU family to select wheels for (see
+    // `runtime-install-active`, @requires-gpu). The only scenarios that reach this
+    // step without `@requires-gpu` are the mock-lane chat scenarios, whose serve
+    // is backed by MockServer and needs no runtime at all — so skip the install
+    // rather than attempt a multi-GiB SDK pull that can't succeed here.
+    if !e2e_cucumber::capability::host_capability().has_amd_gpu {
+        return;
+    }
     // This precondition only needs *a* runtime present — it does not assert a
     // clean slate — so opt into the shared runtimes tree: the first scenario to
     // hit an empty shared tree installs once, and every later scenario finds the

@@ -412,7 +412,25 @@ async fn setup_large_gpu_model(world: &mut E2eWorld) {
 
 #[given("a model is served in the background")]
 async fn setup_background_model(world: &mut E2eWorld) {
-    setup_gpu_model(world).await;
+    // This precondition backs behavioural chat scenarios (tool-defs accepted,
+    // end-to-end reply) whose When/Then talk to the served endpoint over HTTP and
+    // never assert on real generation — only that a served model answers. So it
+    // doesn't need a real GPU serve: on a host WITH an AMD GPU we still exercise
+    // the real `rocm serve --managed` path (extra real coverage where hardware
+    // exists), but on a no-GPU host we back it with the in-process MockServer +
+    // a planted managed-service record. This lets the scenarios drop
+    // `@requires-gpu` and run on the GitHub-hosted mock lane every PR, with no
+    // coverage loss. Real inference stays covered by the `@requires-gpu`
+    // serve-*-inference scenarios.
+    if e2e_cucumber::capability::host_capability().has_amd_gpu {
+        setup_gpu_model(world).await;
+    } else {
+        let mock = MockServer::start("TestModel/E2E-1B").await;
+        world.endpoint = Some(mock.base_url());
+        world.model_name = Some("TestModel/E2E-1B".to_string());
+        world.mock = Some(mock);
+        world.register_mock_service();
+    }
 }
 
 #[given("the served model has been detected")]