From de95de46124306178a86175a10cb4a2989742574 Mon Sep 17 00:00:00 2001 From: Daniel Han Date: Wed, 9 Sep 2026 13:20:39 +0000 Subject: [PATCH 01/14] Pin unslothai/llama.cpp#210 (server-side preemption and exact concurrency) at afd3248b82 --- scripts/unsloth/pr-set.json | 3 ++- 1 file changed, 2 insertions(+), 1 deletion(-) diff --git a/scripts/unsloth/pr-set.json b/scripts/unsloth/pr-set.json index 06d29286360f..aab5f61e18c5 100644 --- a/scripts/unsloth/pr-set.json +++ b/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/afd3248b8270bf4aa66f5f3fb0e2c9cbc64337a1" ] } From e2e03901b3d4b3879013765e2bc8c43530915cdd Mon Sep 17 00:00:00 2001 From: Daniel Han Date: Wed, 9 Sep 2026 13:57:01 +0000 Subject: [PATCH 02/14] Record unslothai#210 in feature-checks.json as knowingly unchecked --- scripts/unsloth/feature-checks.json | 3 ++- 1 file changed, 2 insertions(+), 1 deletion(-) diff --git a/scripts/unsloth/feature-checks.json b/scripts/unsloth/feature-checks.json index 93fe8d3c18f5..3792bb2a14ef 100644 --- a/scripts/unsloth/feature-checks.json +++ b/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" } } From 4a1409f94f32b6d151c0027f91c772ce533bf952 Mon Sep 17 00:00:00 2001 From: Daniel Han Date: Wed, 9 Sep 2026 14:27:59 +0000 Subject: [PATCH 03/14] Repin unslothai/llama.cpp#210 at 62cf502303, composed with ggml-org#25731 for b10871 --- scripts/unsloth/pr-set.json | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/scripts/unsloth/pr-set.json b/scripts/unsloth/pr-set.json index aab5f61e18c5..4fc0cc50590c 100644 --- a/scripts/unsloth/pr-set.json +++ b/scripts/unsloth/pr-set.json @@ -31,6 +31,6 @@ "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/210/commits/afd3248b8270bf4aa66f5f3fb0e2c9cbc64337a1" + "https://github.com/unslothai/llama.cpp/pull/210/commits/62cf502303c970f4782f822592d2d72f031df5ae" ] } From 85651e1658b41f021096a98d04d55d22ee5c1ffe Mon Sep 17 00:00:00 2001 From: Daniel Han Date: Wed, 9 Sep 2026 16:33:04 +0000 Subject: [PATCH 04/14] Repin unslothai/llama.cpp#210 at 17a0f3b967 with the two #197 follow-ups --- scripts/unsloth/pr-set.json | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/scripts/unsloth/pr-set.json b/scripts/unsloth/pr-set.json index 4fc0cc50590c..ba880afa5dd7 100644 --- a/scripts/unsloth/pr-set.json +++ b/scripts/unsloth/pr-set.json @@ -31,6 +31,6 @@ "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/210/commits/62cf502303c970f4782f822592d2d72f031df5ae" + "https://github.com/unslothai/llama.cpp/pull/210/commits/17a0f3b9678618ccbaa3887e3d179edd4a32fd57" ] } From 3bc18770084d8c43e9442398c704fae860e14478 Mon Sep 17 00:00:00 2001 From: Daniel Han Date: Thu, 10 Sep 2026 05:59:44 +0000 Subject: [PATCH 05/14] Repin unslothai/llama.cpp#210 at 1d5548978d: parking is off unless --preempt-ram is given --- scripts/unsloth/pr-set.json | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/scripts/unsloth/pr-set.json b/scripts/unsloth/pr-set.json index ba880afa5dd7..d6bead499d89 100644 --- a/scripts/unsloth/pr-set.json +++ b/scripts/unsloth/pr-set.json @@ -31,6 +31,6 @@ "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/210/commits/17a0f3b9678618ccbaa3887e3d179edd4a32fd57" + "https://github.com/unslothai/llama.cpp/pull/210/commits/1d5548978df86d27a3099f463120f0c30e1aca16" ] } From cff70853fe089babf6fb2735425482baa38cda5a Mon Sep 17 00:00:00 2001 From: Daniel Han Date: Thu, 10 Sep 2026 07:13:07 +0000 Subject: [PATCH 06/14] Repin unslothai/llama.cpp#210 at 15bcc6c7a8: the paged path reads src[5] only for its own op --- scripts/unsloth/pr-set.json | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/scripts/unsloth/pr-set.json b/scripts/unsloth/pr-set.json index d6bead499d89..ab0a1f65f562 100644 --- a/scripts/unsloth/pr-set.json +++ b/scripts/unsloth/pr-set.json @@ -31,6 +31,6 @@ "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/210/commits/1d5548978df86d27a3099f463120f0c30e1aca16" + "https://github.com/unslothai/llama.cpp/pull/210/commits/15bcc6c7a8bb5bbb732b904f28449c8be1fed2b9" ] } From 0585f093bfa86a385962ce6657067168bcca2dc3 Mon Sep 17 00:00:00 2001 From: Daniel Han Date: Thu, 10 Sep 2026 12:40:53 +0000 Subject: [PATCH 07/14] Repin #210 at 3e7d9530a: the CI fixes for the setenv, OpenVINO and park assertion legs --- scripts/unsloth/pr-set.json | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/scripts/unsloth/pr-set.json b/scripts/unsloth/pr-set.json index ab0a1f65f562..01fa9e153e37 100644 --- a/scripts/unsloth/pr-set.json +++ b/scripts/unsloth/pr-set.json @@ -31,6 +31,6 @@ "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/210/commits/15bcc6c7a8bb5bbb732b904f28449c8be1fed2b9" + "https://github.com/unslothai/llama.cpp/pull/210/commits/3e7d9530ac7cf9c21b30e97e6ca2901659cfda6c" ] } From 56476f267d13a3668765c2a5cc89b168ebbc94c4 Mon Sep 17 00:00:00 2001 From: Daniel Han Date: Thu, 10 Sep 2026 13:23:26 +0000 Subject: [PATCH 08/14] Repin #210 at 2176de1ad: the test-exact-buft placement and the two server test timings --- scripts/unsloth/pr-set.json | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/scripts/unsloth/pr-set.json b/scripts/unsloth/pr-set.json index 01fa9e153e37..4e995b2d1be2 100644 --- a/scripts/unsloth/pr-set.json +++ b/scripts/unsloth/pr-set.json @@ -31,6 +31,6 @@ "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/210/commits/3e7d9530ac7cf9c21b30e97e6ca2901659cfda6c" + "https://github.com/unslothai/llama.cpp/pull/210/commits/2176de1adffe533396a1359d03fe8dd4a9ce1310" ] } From 20a699aa3c0477c658befa18046199c3b56c4f22 Mon Sep 17 00:00:00 2001 From: Daniel Han Date: Thu, 10 Sep 2026 14:30:43 +0000 Subject: [PATCH 09/14] Repin #210 at 0beed5117: the rotation test timeout, recompute replay stats, streamed responses preempt record and the condensed weight check comment --- scripts/unsloth/pr-set.json | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/scripts/unsloth/pr-set.json b/scripts/unsloth/pr-set.json index 4e995b2d1be2..bca9f2057545 100644 --- a/scripts/unsloth/pr-set.json +++ b/scripts/unsloth/pr-set.json @@ -31,6 +31,6 @@ "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/210/commits/2176de1adffe533396a1359d03fe8dd4a9ce1310" + "https://github.com/unslothai/llama.cpp/pull/210/commits/0beed5117d54d7f99f7b3ec41fe8248b018ae4c6" ] } From 78c6d42b25b2f2a909ef974608bb4c4b6f1d5d2e Mon Sep 17 00:00:00 2001 From: Daniel Han Date: Thu, 10 Sep 2026 15:34:52 +0000 Subject: [PATCH 10/14] Repin #210 at 7eb9cb780: the started-slot test lengths, the recurrent transfer refusal and the condensed test preamble --- scripts/unsloth/pr-set.json | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/scripts/unsloth/pr-set.json b/scripts/unsloth/pr-set.json index bca9f2057545..cc4739f69238 100644 --- a/scripts/unsloth/pr-set.json +++ b/scripts/unsloth/pr-set.json @@ -31,6 +31,6 @@ "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/210/commits/0beed5117d54d7f99f7b3ec41fe8248b018ae4c6" + "https://github.com/unslothai/llama.cpp/pull/210/commits/7eb9cb780741dda641c321ccf16daa84725ee4e9" ] } From c391bb43c76356f3447f2557286dcbc68fc213c0 Mon Sep 17 00:00:00 2001 From: Daniel Han Date: Thu, 10 Sep 2026 16:29:17 +0000 Subject: [PATCH 11/14] Repin #210 at d8daa8fc5: the failed-save park test decided by lengths --- scripts/unsloth/pr-set.json | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/scripts/unsloth/pr-set.json b/scripts/unsloth/pr-set.json index cc4739f69238..043b383cab82 100644 --- a/scripts/unsloth/pr-set.json +++ b/scripts/unsloth/pr-set.json @@ -31,6 +31,6 @@ "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/210/commits/7eb9cb780741dda641c321ccf16daa84725ee4e9" + "https://github.com/unslothai/llama.cpp/pull/210/commits/d8daa8fc514ac964eff151bff299c9fe8bfda131" ] } From aa2b5dfa78521a7de4c2a1aeb50f562eed263ee8 Mon Sep 17 00:00:00 2001 From: Daniel Han Date: Thu, 10 Sep 2026 17:47:43 +0000 Subject: [PATCH 12/14] Repin #210 at 5b66acb4d: prompt counters reset when a recompute park restarts a prefill --- scripts/unsloth/pr-set.json | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/scripts/unsloth/pr-set.json b/scripts/unsloth/pr-set.json index 043b383cab82..b2075208b564 100644 --- a/scripts/unsloth/pr-set.json +++ b/scripts/unsloth/pr-set.json @@ -31,6 +31,6 @@ "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/210/commits/d8daa8fc514ac964eff151bff299c9fe8bfda131" + "https://github.com/unslothai/llama.cpp/pull/210/commits/5b66acb4d7904f231d264f685fca13cc3c840ea1" ] } From 099562d6b16631182315cfac7ce38d595ba1a9b2 Mon Sep 17 00:00:00 2001 From: Daniel Han Date: Thu, 10 Sep 2026 19:13:33 +0000 Subject: [PATCH 13/14] Repin #210 at 41abdfb23: batched notices carry index 0, tests wait for the log to settle --- scripts/unsloth/pr-set.json | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/scripts/unsloth/pr-set.json b/scripts/unsloth/pr-set.json index b2075208b564..5332ef282f08 100644 --- a/scripts/unsloth/pr-set.json +++ b/scripts/unsloth/pr-set.json @@ -31,6 +31,6 @@ "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/210/commits/5b66acb4d7904f231d264f685fca13cc3c840ea1" + "https://github.com/unslothai/llama.cpp/pull/210/commits/41abdfb23111fbcbf1bace38fb4655ed8cc7cc72" ] } From aecf89aa7ab3e78b4e081e89d2f2e12c762cf153 Mon Sep 17 00:00:00 2001 From: Daniel Han Date: Thu, 10 Sep 2026 20:45:46 +0000 Subject: [PATCH 14/14] Repin #210 at ae1953905: overlapping requests by lengths, whole-sentence host buffer comment --- scripts/unsloth/pr-set.json | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/scripts/unsloth/pr-set.json b/scripts/unsloth/pr-set.json index 5332ef282f08..3d95937df341 100644 --- a/scripts/unsloth/pr-set.json +++ b/scripts/unsloth/pr-set.json @@ -31,6 +31,6 @@ "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/210/commits/41abdfb23111fbcbf1bace38fb4655ed8cc7cc72" + "https://github.com/unslothai/llama.cpp/pull/210/commits/ae195390599e1ea8c4dbcc6b1d8ec77f5db90c3a" ] }