Pin unslothai/llama.cpp#210 (server-side preemption and exact concurrency) at afd3248b82

scripts/unsloth/feature-checks.json

@@ -99,6 +99,7 @@
     "unslothai#149": "GGML_CUDA_ENABLE_UNIFIED_MEMORY=0 env parsing; needs a CUDA or HIP host, and no runner in the pipeline has one",
     "unslothai#152": "per-run mmap of a context's tensors; a memory-layout change with no observable output difference",
     "unslothai#157": "cudaMemcpyDefault in the ggml_cuda_cpy 2D fast path; needs a CUDA host",
-    "unslothai#158": "ROCm_Host compute buffer type on HIP integrated GPUs; needs a ROCm host"
+    "unslothai#158": "ROCm_Host compute buffer type on HIP integrated GPUs; needs a ROCm host",
+    "unslothai#210": "server-side KV preemption (--preempt-ram, --preempt-async) and opt-in exact concurrency (--exact-concurrency); the server parks and restores slots and pins per-sequence output, which no arch, backend-op or mtmd probe can observe. Covered by test-exact-geometry and the server preemption tests carried in unslothai#197, which need a CUDA host for the exact path"
   }
 }

scripts/unsloth/pr-set.json

@@ -30,6 +30,7 @@
     "https://github.com/unslothai/llama.cpp/pull/149/commits/b65a2dce12c14a489e19a059cb6ee59112f1b733",
     "https://github.com/unslothai/llama.cpp/pull/144/commits/a9e9c3c5fed8a0bb5cc617532d0d16b8f59c13e0",
     "https://github.com/unslothai/llama.cpp/pull/152/commits/b2b5ed9ff86427a530b762a45d3fdbd453bcd4e8",
-    "https://github.com/unslothai/llama.cpp/pull/176/commits/09ce1a4d2939844e211f7b4d30a296f4c1aed9a8"
+    "https://github.com/unslothai/llama.cpp/pull/176/commits/09ce1a4d2939844e211f7b4d30a296f4c1aed9a8",
+    "https://github.com/unslothai/llama.cpp/pull/210/commits/ae195390599e1ea8c4dbcc6b1d8ec77f5db90c3a"
   ]
 }