{ "ciru": { "protocol": { "classification": "local-custom HumanEval0-9 non-thinking MTP speed workload", "arm": "ciru", "host": "ciru", "command": [ "/srv/llm/work/sozo-adaptive-diagnosis-20260905/locked/qwen38-ciru-rocm10-20260905-rc2/run.sh", "-c", "262144", "--jinja", "--reasoning", "off", "--temp", "0", "--top-k", "1", "--top-p", "1", "--min-p", "0", "--host", "127.0.0.1", "--port", "18184", "--parallel", "1", "--metrics", "--slots", "--slot-save-path", "/srv/llm/work/qwen38-he-release-context-20260905/benchmarks/ciru/slots" ], "runtime": { "schema": "ciru.locked-runtime.v1", "candidate": "qwen38-ciru-rocm10-20260905-rc2", "locked_at": "2026-09-05T14:11:42.419157+00:00", "status": "LOCKED development candidate; Sozo ROCm10/gfx1151 qualified in stated scope; no production release or selector change", "base_commit": "baba5e0617ac40aa88b9ba96f4b90e584caec64e", "ancestry": [ "unified-next-20260904", "draft-shortlist", "retained-boundaries", "fixed6-graph-boundaries" ], "lane_checkpoints": [ { "id": "01a06e13-d2d8-7e83-82f3-8e8d1eb99868", "title": "sozo quality lane", "updated_at": "2026-09-05T07:15:56Z" }, { "id": "01a06e23-872a-7f52-8dbe-96b856f617e3", "title": "ciru speed lane", "updated_at": "2026-09-05T07:24:55Z" } ], "all_retained_lane_fixes_included": true, "missing_retained_fixes": [], "inclusions": [ { "fix": "H109C attention descriptor stride repair", "origin": "sozo quality lane", "included": "Inherited unified Next", "source": [ "ggml/src/ggml-cuda/fattn-tile.cuh" ], "evidence": "Six-source unified build identity; all 14 attention-header dependent objects verified in current HIP link" }, { "fix": "Long QSA TOP_K GPU admission", "origin": "sozo quality lane", "included": "Inherited and enabled: GGML_QSA_LONG_TOPK=1", "source": [ "ggml/src/ggml-cuda/ggml-cuda.cu", "ggml/src/ggml-cuda/top-k.cu" ], "evidence": "Qualified 35 selector cases / 17,920 rows; retained TOP_K recompiled unchanged during graph integration" }, { "fix": "Canonical restored-KV fast-path admission", "origin": "sozo quality lane", "included": "Inherited and enabled: GGML_QSA_RESTORE_FAST=1", "source": [ "src/llama-memory-hybrid-idx.cpp" ], "evidence": "Verified unified object in current libllama link; served recovery activation itself remains unproven" }, { "fix": "Benchmark PLE and context argument plumbing", "origin": "sozo quality lane", "included": "Inherited unified benchmark binary", "source": [ "tools/llama-bench/llama-bench.cpp" ], "evidence": "Same benchmark binary as retained shared ancestor" }, { "fix": "Seven-column Q5_K weight reuse", "origin": "ciru speed lane", "included": "Inherited unified HIP implementation", "source": [ "ggml/src/ggml-cuda/mmvq.cu" ], "evidence": "Source matches tested e0a39ef3...; verified unified mmvq object; historical 9/9 byte-identical GPU cases" }, { "fix": "GPU MTP top-10 selector and support-probe repair", "origin": "ciru speed lane", "included": "Inherited top10-v2, enabled: CIRU_MTP_TOPK10=1", "source": [ "ggml/src/ggml-cuda/ggml-cuda.cu", "ggml/src/ggml-cuda/top-k.cu", "ggml/src/ggml-cuda/top-k.cuh" ], "evidence": "189 direct cases / 1,323 rows and real sampler support/execution qualified; exact retained source preserved" }, { "fix": "Dynamic 32,768-row MTP output shortlist", "origin": "ciru speed lane", "included": "Inherited draft-shortlist, enabled: CIRU_MTP_SHORTLIST=32768", "source": [ "common/speculative.cpp", "src/ciru-mtp-shortlist.h", "src/models/qwen4exp.cpp" ], "evidence": "Original libllama-common unchanged; same source and retained model object. Target verification preserved" }, { "fix": "Decode-entry and internal-microbatch fences", "origin": "sozo quality lane; integrated and qualified by Profile Sozo tools", "included": "Added to direct shared shortlist ancestor", "source": [ "src/llama-context.cpp" ], "evidence": "22 production-shape numerical captures plus actual MTP and balanced 520-token serving; final tokens match; review proposals improve" }, { "fix": "Fresh HIP graph recapture and no redundant first update", "origin": "Profile Sozo tools", "included": "Added, fresh instantiation default; legacy update option cleared by launcher", "source": [ "ggml/src/ggml-cuda/ggml-cuda.cu" ], "evidence": "Same-policy MTP outputs/counters match; graph-only balanced serving; causal correlated profile. Retained for runtime-work reduction with modest TG" }, { "fix": "Set launcher working directory to its bundle", "origin": "Profile Sozo tools consolidation", "included": "Packaging fix in RC2", "source": [ "run.sh" ], "evidence": "Ciru version startup blocked on home/cirudata stat; same binary passed immediately with bundle cwd. Launcher now selects that cwd before exec." } ], "exclusions": [ { "candidate": "Adaptive confidence/draft-depth policy", "reason": "Corrected comparison is parity; fixed-six remains retained" }, { "candidate": "Broad Q4_1 expert reuse, route sorting and gate/up fusion", "reason": "No retained in-model win; broad reuse increases actual GPU execution despite isolated capture gains" }, { "candidate": "Q4_1 unroll, IQ4_NL expert representation, lookup/head replacements, depth eight", "reason": "No qualified retained upgrade in Ciru speed lane" }, { "candidate": "F32 alpha/beta quality experiment", "reason": "Failed its gate; no weight upgrade retained by Sozo quality lane" }, { "candidate": "DFlash/DFlash2 and feature-width prototype", "reason": "No compatible trained drafter or qualified speed gain; parked under same-day scope" }, { "candidate": "Cooperative expert-weight sharing across parallel token warps", "reason": "Future candidate only at latest Ciru checkpoint; no completed implementation to include" }, { "candidate": "Competitor factory tests and profiler installations", "reason": "Measurement/tooling work, not fixes to the retained model" } ], "launch": { "argv": [ "/bin/llama-server", "--model", "/srv/llm/models/Qwen3.8-Flash-CIRU-STRIX-IU4/Qwen3.8-Flash-CIRU-STRIX-IU4.gguf", "--alias", "qwen38-mtp-probe", "--host", "127.0.0.1", "--port", "18186", "--jinja", "--ple-sidecar", "/srv/llm/models/Qwen3.8-Flash-CIRU-STRIX-IU4/ple", "--ple-cache-mib", "4096", "-ngl", "all", "-sm", "none", "--fit", "off", "-c", "16384", "-b", "2048", "-ub", "512", "--parallel", "1", "-t", "8", "-tb", "8", "-ctk", "f16", "-ctv", "f16", "-fa", "on", "--cont-batching", "--cache-prompt", "--cache-ram", "8192", "--cache-idle-slots", "--ctx-checkpoints", "32", "--checkpoint-min-step", "8192", "--metrics", "--slots", "--spec-type", "draft-mtp", "--spec-draft-model", "/srv/llm/models/Qwen3.8-Flash-CIRU-STRIX-IU4/mtp/Qwen3.8-Flash-CIRU-STRIX-IU4-MTP-Q8_0.gguf", "--spec-draft-ngl", "all", "--spec-draft-device", "ROCm0", "--spec-draft-type-k", "q8_0", "--spec-draft-type-v", "q8_0", "--spec-draft-threads", "8", "--spec-draft-threads-batch", "8", "--spec-draft-n-max", "6", "--spec-draft-n-min", "0", "--spec-draft-p-min", "0", "--spec-draft-p-split", "0.10" ], "env": { "GGML_CUDA_Q41_MOE_FORCE_J": "32", "GGML_QWEN4EXP_PLE_WORKERS": "16", "GGML_QWEN4EXP_PLE_STRICT_SHA": "0", "ROCBLAS_USE_HIPBLASLT": "1", "GGML_QSA_LONG_TOPK": "1", "GGML_QSA_RESTORE_FAST": "1", "CIRU_MTP_TOPK10": "1", "CIRU_MTP_SHORTLIST": "32768" }, "unset": [ "GGML_HIP_GRAPH_EXEC_UPDATE", "CIRU_MTP_GPU_CONFIDENCE", "CIRU_MTP_GPU_ADAPTIVE", "CIRU_MTP_GPU_CONF_MIN", "CIRU_MOE_EXPERT_REUSE", "CIRU_MTP_TRACE", "CIRU_MTP_CONF_TRACE", "LD_PRELOAD" ], "library_path": "/bin plus installed ROCm10 libraries", "cwd": "" }, "models": "models.json; full current hashes verified at lock creation, no weight change", "runtime_hashes": { "bin/libgcc_s.so.1": "c2a120787046d833987189639ef707747f2840ec821b21456523c070030feb57", "bin/libggml-base.so.0.22.0": "277018e8a2c6653d5daa72a864d7b79f8cd82657fa28269689897a4b0082183a", "bin/libggml-cpu.so.0.22.0": "a5e48e3c50f174b25bf369f363db852dacd1f384a8e10dfaf18a392a2fd7aaaa", "bin/libggml-hip.so.0.22.0": "f60bfe62b7dfab548c5a30dcf7728f47b85f142c39ca9ae7377cc340b71c19d4", "bin/libggml.so.0.22.0": "0d80d5364e47fbf30304f754e0bc4dec514c8440e35e430f074a557344bd0df2", "bin/libgomp.so.1": "789f657765b7e6ea2bbc18408adf8bc123b1ea5b44ecb60f6673dd673effc953", "bin/libllama-bench-impl.so": "01feac4d93741273116ff4f5afc6e8398b3149983f0b68d94fa2511d60f7fd55", "bin/libllama-cli-impl.so": "6cad49e306929067c603477d53d56a5fcef13daef8ece5b30eeda804a11c3b17", "bin/libllama-common.so.0.3.0": "8961513d3c8d0d3df48e7197b4711a91c3f0da4da51db39b805a2825affcc3f4", "bin/libllama-server-impl.so": "c2d05f74db8da91d0e905778474d9bdfe554305441dece2117e12c4c617fddb7", "bin/libllama.so.0.3.0": "c29bec577633c8c0c0f0b08a2103b69135aff035aedd49eb70799310aef9576b", "bin/libmtmd.so.0.3.0": "fd0ce09eacd67eb8475facbe70d4cfe5d7a6e8593b3fe358f5857a2b43f65f3e", "bin/libstdc++.so.6": "91df7131df8e14c525527e436d7ea43f7d96ed880b5f3d650b09f8f9a4e53dd6", "bin/llama-bench": "0978408dcdd5d8a016d7072a6ea1f9466f2c5e637a6e6d61c45dc3ac6f9c3d84", "bin/llama-cli": "1276218095861368432ee7b8c9bbf8ddf58e74f16a204557747989ee4ca68b78", "bin/llama-server": "d522ca837e78a06347622b64d4e8f329c2bc1224675367331a20d561911aa5a8" }, "source_manifest": "evidence/inclusion-audit.json", "full_patch": "full.patch", "full_patch_sha256": "e4dd31d385dbe6e640be47cd185877d89cd3696272e80e2e38a58d6364377b3a", "incremental_patch": "evidence/retained-runtime.patch", "incremental_patch_sha256": "fad32d45623248c878bdeb7f470d8ab21839f44eabdd2a04694a318d697292df", "tested_performance": { "code_tg": [ 42.280053297306495, 42.314176153996854 ], "review_tg": [ 31.42178300879495, 31.314892421578826 ], "evidence": "evidence/CONTINUATION-RESULT.json", "milestone_50_plus_met": false }, "correctness": "22 full-vector production-shape captures plus same-policy MTP128 and balanced520 final-output equality. Graph-only acceptance counters equal; boundary correction improves review drafting.", "limits": [ "50+ TG remains unproven", "4K/8K prefixes within16K context and two fixed served prompts; not a full quality/long-session/multi-device release qualification", "Shared system RAM measured; separate VRAM counters unavailable", "Graph-only change +1.56% code/-0.34% review; do not advertise a large fixed6 speedup", "No new weight quantization or training in this lane" ], "lock_policy": "Read-only snapshot. Fork/copy for future changes; changed source, binary, environment or CLI settings must receive a new candidate/variant identity. No automatic updates from lane working trees.", "packaging_parent": "qwen38-ciru-rocm10-20260905-rc1", "launcher_sha256": "6178acd86bb5f9d1dc4300619f4dc798d2d71d3649c9f0e2232e4542c777eec5", "runtime_binary_delta_from_rc1": [] }, "artifacts": { "/srv/llm/models/Qwen3.8-Flash-CIRU-STRIX-IU4/Qwen3.8-Flash-CIRU-STRIX-IU4.gguf": { "resolved": "/home/halo/llm-models/Qwen3.8-Flash-CIRU-STRIX-IU4/Qwen3.8-Flash-CIRU-STRIX-IU4.gguf", "bytes": 79397818720, "link": "/home/halo/llm-models/Qwen3.8-Flash-CIRU-STRIX-IU4/Qwen3.8-Flash-CIRU-STRIX-IU4.gguf", "sha256": "c0ea11e4e24d0f909720b6c4e7462aa1e6fbf5e0f6acc796063f2aed4cf46ed0", "mtime_ns": 1788008884377941688, "device": 66307, "inode": 17838408 }, "/srv/llm/models/Qwen3.8-Flash-CIRU-STRIX-IU4/mtp/Qwen3.8-Flash-CIRU-STRIX-IU4-MTP-Q8_0.gguf": { "resolved": "/home/halo/llm-models/Qwen3.8-Flash-CIRU-STRIX-IU4/mtp/Qwen3.8-Flash-CIRU-STRIX-IU4-MTP-Q8_0.gguf", "bytes": 4135893248, "link": "/home/halo/llm-models/Qwen3.8-Flash-CIRU-STRIX-IU4/mtp/Qwen3.8-Flash-CIRU-STRIX-IU4-MTP-Q8_0.gguf", "sha256": "e6743badef1f2619fcb5addfa4344a2a3368cb75214735117e3af80c70b80642", "mtime_ns": 1788008885971892625, "device": 66307, "inode": 17838409 }, "/srv/llm/models/Qwen3.8-Flash-CIRU-STRIX-IU4/ple/ple.manifest.json": { "resolved": "/srv/llm/models/Qwen3.8-Flash-Next-FP8-CIRUPLE1-bcd9f01/ple.manifest.json", "bytes": 115213, "link": "/srv/llm/models/Qwen3.8-Flash-Next-FP8-CIRUPLE1-bcd9f01/ple.manifest.json", "sha256": "eb7404ce5ef056729452df10ee888e0c300cd0459121444be3313c51788cc171", "mtime_ns": 1787766334187770443, "device": 66308, "inode": 87687205 }, "/srv/llm/models/Qwen3.8-Flash-CIRU-STRIX-IU4/ple/ple.payload.bin": { "resolved": "/srv/llm/models/Qwen3.8-Flash-Next-FP8-CIRUPLE1-bcd9f01/ple.payload.bin", "bytes": 52429053952, "link": "/srv/llm/models/Qwen3.8-Flash-Next-FP8-CIRUPLE1-bcd9f01/ple.payload.bin", "sha256": "687fc742efb6888c6cd7cf9c80cb4b1ac8cb4707b9409c206699c43363e239b2", "mtime_ns": 1787766534066307056, "device": 66308, "inode": 87687204 }, "/srv/llm/models/Qwen3.8-Flash-CIRU-STRIX-IU4/ple/ple.scale.bf16": { "resolved": "/srv/llm/models/Qwen3.8-Flash-Next-FP8-CIRUPLE1-bcd9f01/ple.scale.bf16", "bytes": 2, "link": "/srv/llm/models/Qwen3.8-Flash-Next-FP8-CIRUPLE1-bcd9f01/ple.scale.bf16", "sha256": "c7c58bd6007672362da2106fdbfaf9f50629e4bdf8598169c598027394ef9791", "mtime_ns": 1787764234000000000, "device": 66308, "inode": 87687203 } }, "verified_binary_hashes": { "bin/libgcc_s.so.1": "c2a120787046d833987189639ef707747f2840ec821b21456523c070030feb57", "bin/libggml-base.so.0.22.0": "277018e8a2c6653d5daa72a864d7b79f8cd82657fa28269689897a4b0082183a", "bin/libggml-cpu.so.0.22.0": "a5e48e3c50f174b25bf369f363db852dacd1f384a8e10dfaf18a392a2fd7aaaa", "bin/libggml-hip.so.0.22.0": "f60bfe62b7dfab548c5a30dcf7728f47b85f142c39ca9ae7377cc340b71c19d4", "bin/libggml.so.0.22.0": "0d80d5364e47fbf30304f754e0bc4dec514c8440e35e430f074a557344bd0df2", "bin/libgomp.so.1": "789f657765b7e6ea2bbc18408adf8bc123b1ea5b44ecb60f6673dd673effc953", "bin/libllama-bench-impl.so": "01feac4d93741273116ff4f5afc6e8398b3149983f0b68d94fa2511d60f7fd55", "bin/libllama-cli-impl.so": "6cad49e306929067c603477d53d56a5fcef13daef8ece5b30eeda804a11c3b17", "bin/libllama-common.so.0.3.0": "8961513d3c8d0d3df48e7197b4711a91c3f0da4da51db39b805a2825affcc3f4", "bin/libllama-server-impl.so": "c2d05f74db8da91d0e905778474d9bdfe554305441dece2117e12c4c617fddb7", "bin/libllama.so.0.3.0": "c29bec577633c8c0c0f0b08a2103b69135aff035aedd49eb70799310aef9576b", "bin/libmtmd.so.0.3.0": "fd0ce09eacd67eb8475facbe70d4cfe5d7a6e8593b3fe358f5857a2b43f65f3e", "bin/libstdc++.so.6": "91df7131df8e14c525527e436d7ea43f7d96ed880b5f3d650b09f8f9a4e53dd6", "bin/llama-bench": "0978408dcdd5d8a016d7072a6ea1f9466f2c5e637a6e6d61c45dc3ac6f9c3d84", "bin/llama-cli": "1276218095861368432ee7b8c9bbf8ddf58e74f16a204557747989ee4ca68b78", "bin/llama-server": "d522ca837e78a06347622b64d4e8f329c2bc1224675367331a20d561911aa5a8" }, "dataset_sha256": "b796127e635a67f93fb35c04f4cb03cf06f38c8072ee7cee8833d7bee06979ef", "driver_sha256": "d4978a25fcb0dbaf7f9611b64ae6fc2327e9ac0f9d789afef439ae3efd93c71c", "sampler": { "temperature": 0, "top_k": 1, "top_p": 1, "min_p": 0, "repeat_penalty": 1, "presence_penalty": 0, "frequency_penalty": 0, "seed": 123 }, "reasoning": false, "task_protocol": "Canonical HumanEval prompt in one user message, embedded chat template with enable_thinking=false; no added instructions. One first sample, no retries. Speed and acceptance only; no quality score.", "output_policy": "Natural EOS; n_predict=-1; no custom stops. 1800-second transport timeout. Preserve unsuccessful/incomplete requests.", "aggregate_policy": "TG=sum(generated tokens-1) / sum server decode seconds. Request wall=sum measured HTTP request wall seconds. Panel wall=elapsed loop time including recorder overhead, excluding model load.", "settings_basis": "CIRU released d6 ROCm10 profile at262144 context, per runtime-v2/README.md and model card; Laurent publisher adaptive2\u20134 Q8 targetKV; Unsloth recommended fork sharedQ8 d2. Greedy non-thinking for MTP acceptance speed, supported by retained CIRU probe and published Unsloth greedy MTP recipe.", "governors": { "/sys/devices/system/cpu/cpu21/cpufreq/scaling_governor": "performance", "/sys/devices/system/cpu/cpu11/cpufreq/scaling_governor": "performance", "/sys/devices/system/cpu/cpu9/cpufreq/scaling_governor": "performance", "/sys/devices/system/cpu/cpu7/cpufreq/scaling_governor": "performance", "/sys/devices/system/cpu/cpu28/cpufreq/scaling_governor": "performance", "/sys/devices/system/cpu/cpu18/cpufreq/scaling_governor": "performance", "/sys/devices/system/cpu/cpu5/cpufreq/scaling_governor": "performance", "/sys/devices/system/cpu/cpu26/cpufreq/scaling_governor": "performance", "/sys/devices/system/cpu/cpu16/cpufreq/scaling_governor": "performance", "/sys/devices/system/cpu/cpu3/cpufreq/scaling_governor": "performance", "/sys/devices/system/cpu/cpu24/cpufreq/scaling_governor": "performance", "/sys/devices/system/cpu/cpu14/cpufreq/scaling_governor": "performance", "/sys/devices/system/cpu/cpu1/cpufreq/scaling_governor": "performance", "/sys/devices/system/cpu/cpu22/cpufreq/scaling_governor": "performance", "/sys/devices/system/cpu/cpu12/cpufreq/scaling_governor": "performance", "/sys/devices/system/cpu/cpu30/cpufreq/scaling_governor": "performance", "/sys/devices/system/cpu/cpu20/cpufreq/scaling_governor": "performance", "/sys/devices/system/cpu/cpu10/cpufreq/scaling_governor": "performance", "/sys/devices/system/cpu/cpu8/cpufreq/scaling_governor": "performance", "/sys/devices/system/cpu/cpu29/cpufreq/scaling_governor": "performance", "/sys/devices/system/cpu/cpu19/cpufreq/scaling_governor": "performance", "/sys/devices/system/cpu/cpu6/cpufreq/scaling_governor": "performance", "/sys/devices/system/cpu/cpu27/cpufreq/scaling_governor": "performance", "/sys/devices/system/cpu/cpu17/cpufreq/scaling_governor": "performance", "/sys/devices/system/cpu/cpu4/cpufreq/scaling_governor": "performance", "/sys/devices/system/cpu/cpu25/cpufreq/scaling_governor": "performance", "/sys/devices/system/cpu/cpu15/cpufreq/scaling_governor": "performance", "/sys/devices/system/cpu/cpu2/cpufreq/scaling_governor": "performance", "/sys/devices/system/cpu/cpu23/cpufreq/scaling_governor": "performance", "/sys/devices/system/cpu/cpu13/cpufreq/scaling_governor": "performance", "/sys/devices/system/cpu/cpu0/cpufreq/scaling_governor": "performance", "/sys/devices/system/cpu/cpu31/cpufreq/scaling_governor": "performance" }, "rendered_sha256": "2cc4c7910b4ad2e7f96e9f5f6ffda94e14ea458f110d302b2ff7a44358e92f2e", "template_sha256": "c3cf9e34abf4f9e36c2d72165aa9c132d3e2a725b6c2586aaa3a8af9d7a81041" }, "identity": { "arm": "ciru", "host": "ciru", "protocol_lock": "/srv/llm/work/qwen38-he-release-context-20260905/benchmarks/ciru/protocol.lock.json", "command": [ "/srv/llm/work/sozo-adaptive-diagnosis-20260905/locked/qwen38-ciru-rocm10-20260905-rc2/run.sh", "-c", "262144", "--jinja", "--reasoning", "off", "--temp", "0", "--top-k", "1", "--top-p", "1", "--min-p", "0", "--host", "127.0.0.1", "--port", "18184", "--parallel", "1", "--metrics", "--slots", "--slot-save-path", "/srv/llm/work/qwen38-he-release-context-20260905/benchmarks/ciru/slots" ], "maps": [ "7a5767c00000-7a576aecb000 r--p 00000000 103:03 97537285 /srv/llm/work/sozo-adaptive-diagnosis-20260905/locked/qwen38-ciru-rocm10-20260905-rc2/bin/libggml-hip.so.0.22.0", "7a576aecb000-7a576b9bc000 r-xp 032ca000 103:03 97537285 /srv/llm/work/sozo-adaptive-diagnosis-20260905/locked/qwen38-ciru-rocm10-20260905-rc2/bin/libggml-hip.so.0.22.0", "7a576b9bc000-7a576b9cc000 r--p 03dba000 103:03 97537285 /srv/llm/work/sozo-adaptive-diagnosis-20260905/locked/qwen38-ciru-rocm10-20260905-rc2/bin/libggml-hip.so.0.22.0", "7a576b9cc000-7a576b9d0000 rw-p 03dc9000 103:03 97537285 /srv/llm/work/sozo-adaptive-diagnosis-20260905/locked/qwen38-ciru-rocm10-20260905-rc2/bin/libggml-hip.so.0.22.0", "7a576ba3a000-7a576ba4d000 r--p 00000000 103:03 97537284 /srv/llm/work/sozo-adaptive-diagnosis-20260905/locked/qwen38-ciru-rocm10-20260905-rc2/bin/libggml-cpu.so.0.22.0", "7a576ba4d000-7a576bb58000 r-xp 00013000 103:03 97537284 /srv/llm/work/sozo-adaptive-diagnosis-20260905/locked/qwen38-ciru-rocm10-20260905-rc2/bin/libggml-cpu.so.0.22.0", "7a576bb58000-7a576bb78000 r--p 0011e000 103:03 97537284 /srv/llm/work/sozo-adaptive-diagnosis-20260905/locked/qwen38-ciru-rocm10-20260905-rc2/bin/libggml-cpu.so.0.22.0", "7a576bb78000-7a576bb7a000 r--p 0013d000 103:03 97537284 /srv/llm/work/sozo-adaptive-diagnosis-20260905/locked/qwen38-ciru-rocm10-20260905-rc2/bin/libggml-cpu.so.0.22.0", "7a576bb7a000-7a576bb7b000 rw-p 0013f000 103:03 97537284 /srv/llm/work/sozo-adaptive-diagnosis-20260905/locked/qwen38-ciru-rocm10-20260905-rc2/bin/libggml-cpu.so.0.22.0", "7a576bc00000-7a576bcb4000 r--p 00000000 103:03 97537292 /srv/llm/work/sozo-adaptive-diagnosis-20260905/locked/qwen38-ciru-rocm10-20260905-rc2/bin/libllama.so.0.3.0", "7a576bcb4000-7a576bf62000 r-xp 000b4000 103:03 97537292 /srv/llm/work/sozo-adaptive-diagnosis-20260905/locked/qwen38-ciru-rocm10-20260905-rc2/bin/libllama.so.0.3.0", "7a576bf62000-7a576bfe9000 r--p 00362000 103:03 97537292 /srv/llm/work/sozo-adaptive-diagnosis-20260905/locked/qwen38-ciru-rocm10-20260905-rc2/bin/libllama.so.0.3.0", "7a576bfe9000-7a576bff7000 r--p 003e8000 103:03 97537292 /srv/llm/work/sozo-adaptive-diagnosis-20260905/locked/qwen38-ciru-rocm10-20260905-rc2/bin/libllama.so.0.3.0", "7a576bff7000-7a576bffc000 rw-p 003f6000 103:03 97537292 /srv/llm/work/sozo-adaptive-diagnosis-20260905/locked/qwen38-ciru-rocm10-20260905-rc2/bin/libllama.so.0.3.0", "7a576c000000-7a576c0a8000 r--p 00000000 103:03 97537290 /srv/llm/work/sozo-adaptive-diagnosis-20260905/locked/qwen38-ciru-rocm10-20260905-rc2/bin/libllama-common.so.0.3.0", "7a576c0a8000-7a576c389000 r-xp 000a8000 103:03 97537290 /srv/llm/work/sozo-adaptive-diagnosis-20260905/locked/qwen38-ciru-rocm10-20260905-rc2/bin/libllama-common.so.0.3.0", "7a576c389000-7a576c426000 r--p 00389000 103:03 97537290 /srv/llm/work/sozo-adaptive-diagnosis-20260905/locked/qwen38-ciru-rocm10-20260905-rc2/bin/libllama-common.so.0.3.0", "7a576c426000-7a576c42e000 r--p 00425000 103:03 97537290 /srv/llm/work/sozo-adaptive-diagnosis-20260905/locked/qwen38-ciru-rocm10-20260905-rc2/bin/libllama-common.so.0.3.0", "7a576c42e000-7a576c433000 rw-p 0042d000 103:03 97537290 /srv/llm/work/sozo-adaptive-diagnosis-20260905/locked/qwen38-ciru-rocm10-20260905-rc2/bin/libllama-common.so.0.3.0", "7a576c600000-7a576c67b000 r--p 00000000 103:03 97537291 /srv/llm/work/sozo-adaptive-diagnosis-20260905/locked/qwen38-ciru-rocm10-20260905-rc2/bin/libllama-server-impl.so", "7a576c67b000-7a576c875000 r-xp 0007b000 103:03 97537291 /srv/llm/work/sozo-adaptive-diagnosis-20260905/locked/qwen38-ciru-rocm10-20260905-rc2/bin/libllama-server-impl.so", "7a576c875000-7a576cbc3000 r--p 00275000 103:03 97537291 /srv/llm/work/sozo-adaptive-diagnosis-20260905/locked/qwen38-ciru-rocm10-20260905-rc2/bin/libllama-server-impl.so", "7a576cbc3000-7a576cbc8000 r--p 005c3000 103:03 97537291 /srv/llm/work/sozo-adaptive-diagnosis-20260905/locked/qwen38-ciru-rocm10-20260905-rc2/bin/libllama-server-impl.so", "7a576cbc8000-7a576cbcc000 rw-p 005c8000 103:03 97537291 /srv/llm/work/sozo-adaptive-diagnosis-20260905/locked/qwen38-ciru-rocm10-20260905-rc2/bin/libllama-server-impl.so", "7a576cc03000-7a576cc18000 r--p 00000000 103:03 97537283 /srv/llm/work/sozo-adaptive-diagnosis-20260905/locked/qwen38-ciru-rocm10-20260905-rc2/bin/libggml-base.so.0.22.0", "7a576cc18000-7a576ccb1000 r-xp 00015000 103:03 97537283 /srv/llm/work/sozo-adaptive-diagnosis-20260905/locked/qwen38-ciru-rocm10-20260905-rc2/bin/libggml-base.so.0.22.0", "7a576ccb1000-7a576ccd1000 r--p 000ae000 103:03 97537283 /srv/llm/work/sozo-adaptive-diagnosis-20260905/locked/qwen38-ciru-rocm10-20260905-rc2/bin/libggml-base.so.0.22.0", "7a576ccd1000-7a576ccd3000 r--p 000ce000 103:03 97537283 /srv/llm/work/sozo-adaptive-diagnosis-20260905/locked/qwen38-ciru-rocm10-20260905-rc2/bin/libggml-base.so.0.22.0", "7a576ccd3000-7a576ccd5000 rw-p 000d0000 103:03 97537283 /srv/llm/work/sozo-adaptive-diagnosis-20260905/locked/qwen38-ciru-rocm10-20260905-rc2/bin/libggml-base.so.0.22.0", "7a576ccd5000-7a576ccdd000 rw-p 000e3000 103:03 97537283 /srv/llm/work/sozo-adaptive-diagnosis-20260905/locked/qwen38-ciru-rocm10-20260905-rc2/bin/libggml-base.so.0.22.0" ], "pid": 3061397, "slots": [ { "id": 0, "n_ctx": 262144, "speculative": true, "is_processing": false } ], "props": { "default_generation_settings": { "params": { "seed": 4294967295, "temperature": 0.0, "dynatemp_range": 0.0, "dynatemp_exponent": 1.0, "top_k": 1, "top_p": 1.0, "min_p": 0.0, "top_n_sigma": -1.0, "xtc_probability": 0.0, "xtc_threshold": 0.10000000149011612, "typical_p": 1.0, "repeat_last_n": 64, "repeat_penalty": 1.0, "presence_penalty": 0.0, "frequency_penalty": 0.0, "dry_multiplier": 0.0, "dry_base": 1.75, "dry_allowed_length": 2, "dry_penalty_last_n": 64, "mirostat": 0, "mirostat_tau": 5.0, "mirostat_eta": 0.10000000149011612, "adaptive_target": -1.0, "adaptive_decay": 0.8999999761581421, "max_tokens": -1, "n_predict": -1, "n_keep": 0, "n_discard": 0, "ignore_eos": false, "stream": false, "n_probs": 0, "min_keep": 0, "chat_format": "Content-only", "reasoning_format": "none", "reasoning_in_content": false, "generation_prompt": "", "samplers": [ "penalties", "dry", "top_n_sigma", "top_k", "typ_p", "top_p", "min_p", "xtc", "temperature" ], "speculative.types": "none", "timings_per_token": false, "post_sampling_probs": false, "backend_sampling": false, "lora": [] }, "n_ctx": 262144 }, "total_slots": 1, "model_alias": "qwen38-mtp-probe", "model_ftype": "Q5_K - Medium", "model_path": "/srv/llm/models/Qwen3.8-Flash-CIRU-STRIX-IU4/Qwen3.8-Flash-CIRU-STRIX-IU4.gguf", "modalities": { "vision": false, "video": false, "audio": false }, "media_marker": "<__media_6eQTpvIWFsxSW9heZdJedSgS5qjjXKUx__>", "endpoint_slots": true, "endpoint_props": false, "endpoint_metrics": true, "ui": true, "ui_settings": {}, "chat_template": "{%- set image_count = namespace(value=0) %}\n{%- set video_count = namespace(value=0) %}\n{%- macro render_content(content, do_vision_count, is_system_content=false) %}\n {%- if content is string %}\n {{- content }}\n {%- elif content is iterable and content is not mapping %}\n {%- for item in content %}\n {%- if 'image' in item or 'image_url' in item or item.type == 'image' %}\n {%- if is_system_content %}\n {{- raise_exception('System message cannot contain images.') }}\n {%- endif %}\n {%- if do_vision_count %}\n {%- set image_count.value = image_count.value + 1 %}\n {%- endif %}\n {%- if add_vision_id %}\n {{- 'Picture ' ~ image_count.value ~ ': ' }}\n {%- endif %}\n {{- '<|vision_start|><|image_pad|><|vision_end|>' }}\n {%- elif 'video' in item or item.type == 'video' %}\n {%- if is_system_content %}\n {{- raise_exception('System message cannot contain videos.') }}\n {%- endif %}\n {%- if do_vision_count %}\n {%- set video_count.value = video_count.value + 1 %}\n {%- endif %}\n {%- if add_vision_id %}\n {{- 'Video ' ~ video_count.value ~ ': ' }}\n {%- endif %}\n {{- '<|vision_start|><|video_pad|><|vision_end|>' }}\n {%- elif 'text' in item %}\n {{- item.text }}\n {%- else %}\n {{- raise_exception('Unexpected item type in content.') }}\n {%- endif %}\n {%- endfor %}\n {%- elif content is none or content is undefined %}\n {{- '' }}\n {%- else %}\n {{- raise_exception('Unexpected content type.') }}\n {%- endif %}\n{%- endmacro %}\n{%- if not messages %}\n {{- raise_exception('No messages provided.') }}\n{%- endif %}\n{%- set reasoning_instructions = '' %}\n{%- if enable_thinking is undefined or enable_thinking is true %}\n {%- set resolved_reasoning_effort = reasoning_effort|default('xhigh') %}\n {%- if resolved_reasoning_effort not in ('xhigh', 'medium', 'low') %}\n {{- raise_exception('Unexpected reasoning effort ' ~ reasoning_effort ~ '. Supported types are xhigh (default), medium, and low.') }}\n {%- endif %}\n {%- if resolved_reasoning_effort == 'xhigh' %}\n {%- set reasoning_instructions = 'Reasoning effort is set to xhigh. Please think carefully through the task, validate key assumptions, consider plausible alternatives, and prioritize correctness, consistency, and clarity in the final answer.' %}\n {%- elif resolved_reasoning_effort == 'low' %}\n {%- set reasoning_instructions = 'Reasoning effort is set to low. Keep your thinking brief and focused, moving directly to the conclusion without unnecessary elaboration.' %}\n {%- endif %}\n{%- endif %}\n{%- if tools and tools is iterable and tools is not mapping %}\n {{- '<|im_start|>system\\n' }}\n {%- if reasoning_instructions %}\n {{- reasoning_instructions + '\\n\\n' }}\n {%- endif %}\n {{- \"# Tools\\n\\nYou have access to the following functions:\\n\\n\" }}\n {%- for tool in tools %}\n {{- \"\\n\" }}\n {{- tool | tojson }}\n {%- endfor %}\n {{- \"\\n\" }}\n {{- '\\n\\nIf you choose to call a function ONLY reply in the following format with NO suffix:\\n\\n\\n\\n\\nvalue_1\\n\\n\\nThis is the value for the second parameter\\nthat can span\\nmultiple lines\\n\\n\\n\\n\\n\\nReminder:\\n- Function calls MUST follow the specified format: an inner block must be nested within XML tags\\n- Required parameters MUST be specified\\n- You may provide optional reasoning for your function call in natural language BEFORE the function call, but NOT after\\n- If there is no function call available, answer the question like normal with your current knowledge and do not tell the user about function calls\\n' }}\n {%- if messages[0].role == 'system' %}\n {%- set content = render_content(messages[0].content, false, true)|trim %}\n {%- if content %}\n {{- '\\n\\n' + content }}\n {%- endif %}\n {%- endif %}\n {{- '<|im_end|>\\n' }}\n{%- else %}\n {%- if messages[0].role == 'system' %}\n {%- set content = render_content(messages[0].content, false, true)|trim %}\n {%- if content %}\n {{- '<|im_start|>system\\n' + (reasoning_instructions + '\\n\\n' if reasoning_instructions else '') + content + '<|im_end|>\\n' }}\n {%- elif reasoning_instructions %}\n {{- '<|im_start|>system\\n' + reasoning_instructions + '<|im_end|>\\n' }}\n {%- endif %}\n {%- elif reasoning_instructions %}\n {{- '<|im_start|>system\\n' + reasoning_instructions + '<|im_end|>\\n' }}\n {%- endif %}\n{%- endif %}\n{%- set ns = namespace(multi_step_tool=true, last_query_index=messages|length - 1) %}\n{%- for message in messages[::-1] %}\n {%- set index = (messages|length - 1) - loop.index0 %}\n {%- if ns.multi_step_tool and message.role == \"user\" %}\n {%- set content = render_content(message.content, false)|trim %}\n {%- if not(content.startswith('') and content.endswith('')) %}\n {%- set ns.multi_step_tool = false %}\n {%- set ns.last_query_index = index %}\n {%- endif %}\n {%- endif %}\n{%- endfor %}\n{%- if ns.multi_step_tool %}\n {{- raise_exception('No user query found in messages.') }}\n{%- endif %}\n{%- for message in messages %}\n {%- set content = render_content(message.content, true)|trim %}\n {%- if message.role == \"system\" %}\n {%- if not loop.first %}\n {{- raise_exception('System message must be at the beginning.') }}\n {%- endif %}\n {%- elif message.role == \"user\" %}\n {{- '<|im_start|>' + message.role + '\\n' + content + '<|im_end|>' + '\\n' }}\n {%- elif message.role == \"assistant\" %}\n {%- set reasoning_content = '' %}\n {%- if message.reasoning_content is string %}\n {%- set reasoning_content = message.reasoning_content %}\n {%- endif %}\n {%- set reasoning_content = reasoning_content|trim %}\n {%- if preserve_thinking is undefined or preserve_thinking is true or loop.index0 > ns.last_query_index %}\n {{- '<|im_start|>' + message.role + '\\n\\n' + reasoning_content + '\\n\\n\\n' + content }}\n {%- else %}\n {{- '<|im_start|>' + message.role + '\\n' + content }}\n {%- endif %}\n {%- if message.tool_calls and message.tool_calls is iterable and message.tool_calls is not mapping %}\n {%- for tool_call in message.tool_calls %}\n {%- if tool_call.function is defined %}\n {%- set tool_call = tool_call.function %}\n {%- endif %}\n {%- if loop.first %}\n {%- if content|trim %}\n {{- '\\n\\n\\n\\n' }}\n {%- else %}\n {{- '\\n\\n' }}\n {%- endif %}\n {%- else %}\n {{- '\\n\\n\\n' }}\n {%- endif %}\n {%- if tool_call.arguments is defined and tool_call.arguments != '' %}\n {%- for args_name, args_value in tool_call.arguments|items %}\n {{- '\\n' }}\n {%- set args_value = args_value | string if args_value is string else args_value | tojson | safe %}\n {{- args_value }}\n {{- '\\n\\n' }}\n {%- endfor %}\n {%- endif %}\n {{- '\\n' }}\n {%- endfor %}\n {%- endif %}\n {{- '<|im_end|>\\n' }}\n {%- elif message.role == \"tool\" %}\n {%- if loop.previtem and loop.previtem.role != \"tool\" %}\n {{- '<|im_start|>user' }}\n {%- endif %}\n {{- '\\n\\n' }}\n {{- content }}\n {{- '\\n' }}\n {%- if not loop.last and loop.nextitem.role != \"tool\" %}\n {{- '<|im_end|>\\n' }}\n {%- elif loop.last %}\n {{- '<|im_end|>\\n' }}\n {%- endif %}\n {%- else %}\n {{- raise_exception('Unexpected message role.') }}\n {%- endif %}\n{%- endfor %}\n{%- if add_generation_prompt %}\n {{- '<|im_start|>assistant\\n' }}\n {%- if enable_thinking is defined and enable_thinking is false %}\n {{- '\\n\\n\\n\\n' }}\n {%- else %}\n {{- '\\n' }}\n {%- endif %}\n{%- endif %}", "chat_template_caps": { "supports_object_arguments": true, "supports_parallel_tool_calls": true, "supports_preserve_reasoning": true, "supports_reasoning_effort": true, "supports_string_content": true, "supports_system_role": true, "supports_tool_calls": true, "supports_tools": true, "supports_typed_content": false }, "bos_token": "<|endoftext|>", "eos_token": "<|im_end|>", "build_info": "b1-baba5e0", "is_sleeping": false, "cors_proxy_enabled": false }, "runtime": { "schema": "ciru.locked-runtime.v1", "candidate": "qwen38-ciru-rocm10-20260905-rc2", "locked_at": "2026-09-05T14:11:42.419157+00:00", "status": "LOCKED development candidate; Sozo ROCm10/gfx1151 qualified in stated scope; no production release or selector change", "base_commit": "baba5e0617ac40aa88b9ba96f4b90e584caec64e", "ancestry": [ "unified-next-20260904", "draft-shortlist", "retained-boundaries", "fixed6-graph-boundaries" ], "lane_checkpoints": [ { "id": "01a06e13-d2d8-7e83-82f3-8e8d1eb99868", "title": "sozo quality lane", "updated_at": "2026-09-05T07:15:56Z" }, { "id": "01a06e23-872a-7f52-8dbe-96b856f617e3", "title": "ciru speed lane", "updated_at": "2026-09-05T07:24:55Z" } ], "all_retained_lane_fixes_included": true, "missing_retained_fixes": [], "inclusions": [ { "fix": "H109C attention descriptor stride repair", "origin": "sozo quality lane", "included": "Inherited unified Next", "source": [ "ggml/src/ggml-cuda/fattn-tile.cuh" ], "evidence": "Six-source unified build identity; all 14 attention-header dependent objects verified in current HIP link" }, { "fix": "Long QSA TOP_K GPU admission", "origin": "sozo quality lane", "included": "Inherited and enabled: GGML_QSA_LONG_TOPK=1", "source": [ "ggml/src/ggml-cuda/ggml-cuda.cu", "ggml/src/ggml-cuda/top-k.cu" ], "evidence": "Qualified 35 selector cases / 17,920 rows; retained TOP_K recompiled unchanged during graph integration" }, { "fix": "Canonical restored-KV fast-path admission", "origin": "sozo quality lane", "included": "Inherited and enabled: GGML_QSA_RESTORE_FAST=1", "source": [ "src/llama-memory-hybrid-idx.cpp" ], "evidence": "Verified unified object in current libllama link; served recovery activation itself remains unproven" }, { "fix": "Benchmark PLE and context argument plumbing", "origin": "sozo quality lane", "included": "Inherited unified benchmark binary", "source": [ "tools/llama-bench/llama-bench.cpp" ], "evidence": "Same benchmark binary as retained shared ancestor" }, { "fix": "Seven-column Q5_K weight reuse", "origin": "ciru speed lane", "included": "Inherited unified HIP implementation", "source": [ "ggml/src/ggml-cuda/mmvq.cu" ], "evidence": "Source matches tested e0a39ef3...; verified unified mmvq object; historical 9/9 byte-identical GPU cases" }, { "fix": "GPU MTP top-10 selector and support-probe repair", "origin": "ciru speed lane", "included": "Inherited top10-v2, enabled: CIRU_MTP_TOPK10=1", "source": [ "ggml/src/ggml-cuda/ggml-cuda.cu", "ggml/src/ggml-cuda/top-k.cu", "ggml/src/ggml-cuda/top-k.cuh" ], "evidence": "189 direct cases / 1,323 rows and real sampler support/execution qualified; exact retained source preserved" }, { "fix": "Dynamic 32,768-row MTP output shortlist", "origin": "ciru speed lane", "included": "Inherited draft-shortlist, enabled: CIRU_MTP_SHORTLIST=32768", "source": [ "common/speculative.cpp", "src/ciru-mtp-shortlist.h", "src/models/qwen4exp.cpp" ], "evidence": "Original libllama-common unchanged; same source and retained model object. Target verification preserved" }, { "fix": "Decode-entry and internal-microbatch fences", "origin": "sozo quality lane; integrated and qualified by Profile Sozo tools", "included": "Added to direct shared shortlist ancestor", "source": [ "src/llama-context.cpp" ], "evidence": "22 production-shape numerical captures plus actual MTP and balanced 520-token serving; final tokens match; review proposals improve" }, { "fix": "Fresh HIP graph recapture and no redundant first update", "origin": "Profile Sozo tools", "included": "Added, fresh instantiation default; legacy update option cleared by launcher", "source": [ "ggml/src/ggml-cuda/ggml-cuda.cu" ], "evidence": "Same-policy MTP outputs/counters match; graph-only balanced serving; causal correlated profile. Retained for runtime-work reduction with modest TG" }, { "fix": "Set launcher working directory to its bundle", "origin": "Profile Sozo tools consolidation", "included": "Packaging fix in RC2", "source": [ "run.sh" ], "evidence": "Ciru version startup blocked on home/cirudata stat; same binary passed immediately with bundle cwd. Launcher now selects that cwd before exec." } ], "exclusions": [ { "candidate": "Adaptive confidence/draft-depth policy", "reason": "Corrected comparison is parity; fixed-six remains retained" }, { "candidate": "Broad Q4_1 expert reuse, route sorting and gate/up fusion", "reason": "No retained in-model win; broad reuse increases actual GPU execution despite isolated capture gains" }, { "candidate": "Q4_1 unroll, IQ4_NL expert representation, lookup/head replacements, depth eight", "reason": "No qualified retained upgrade in Ciru speed lane" }, { "candidate": "F32 alpha/beta quality experiment", "reason": "Failed its gate; no weight upgrade retained by Sozo quality lane" }, { "candidate": "DFlash/DFlash2 and feature-width prototype", "reason": "No compatible trained drafter or qualified speed gain; parked under same-day scope" }, { "candidate": "Cooperative expert-weight sharing across parallel token warps", "reason": "Future candidate only at latest Ciru checkpoint; no completed implementation to include" }, { "candidate": "Competitor factory tests and profiler installations", "reason": "Measurement/tooling work, not fixes to the retained model" } ], "launch": { "argv": [ "/bin/llama-server", "--model", "/srv/llm/models/Qwen3.8-Flash-CIRU-STRIX-IU4/Qwen3.8-Flash-CIRU-STRIX-IU4.gguf", "--alias", "qwen38-mtp-probe", "--host", "127.0.0.1", "--port", "18186", "--jinja", "--ple-sidecar", "/srv/llm/models/Qwen3.8-Flash-CIRU-STRIX-IU4/ple", "--ple-cache-mib", "4096", "-ngl", "all", "-sm", "none", "--fit", "off", "-c", "16384", "-b", "2048", "-ub", "512", "--parallel", "1", "-t", "8", "-tb", "8", "-ctk", "f16", "-ctv", "f16", "-fa", "on", "--cont-batching", "--cache-prompt", "--cache-ram", "8192", "--cache-idle-slots", "--ctx-checkpoints", "32", "--checkpoint-min-step", "8192", "--metrics", "--slots", "--spec-type", "draft-mtp", "--spec-draft-model", "/srv/llm/models/Qwen3.8-Flash-CIRU-STRIX-IU4/mtp/Qwen3.8-Flash-CIRU-STRIX-IU4-MTP-Q8_0.gguf", "--spec-draft-ngl", "all", "--spec-draft-device", "ROCm0", "--spec-draft-type-k", "q8_0", "--spec-draft-type-v", "q8_0", "--spec-draft-threads", "8", "--spec-draft-threads-batch", "8", "--spec-draft-n-max", "6", "--spec-draft-n-min", "0", "--spec-draft-p-min", "0", "--spec-draft-p-split", "0.10" ], "env": { "GGML_CUDA_Q41_MOE_FORCE_J": "32", "GGML_QWEN4EXP_PLE_WORKERS": "16", "GGML_QWEN4EXP_PLE_STRICT_SHA": "0", "ROCBLAS_USE_HIPBLASLT": "1", "GGML_QSA_LONG_TOPK": "1", "GGML_QSA_RESTORE_FAST": "1", "CIRU_MTP_TOPK10": "1", "CIRU_MTP_SHORTLIST": "32768" }, "unset": [ "GGML_HIP_GRAPH_EXEC_UPDATE", "CIRU_MTP_GPU_CONFIDENCE", "CIRU_MTP_GPU_ADAPTIVE", "CIRU_MTP_GPU_CONF_MIN", "CIRU_MOE_EXPERT_REUSE", "CIRU_MTP_TRACE", "CIRU_MTP_CONF_TRACE", "LD_PRELOAD" ], "library_path": "/bin plus installed ROCm10 libraries", "cwd": "" }, "models": "models.json; full current hashes verified at lock creation, no weight change", "runtime_hashes": { "bin/libgcc_s.so.1": "c2a120787046d833987189639ef707747f2840ec821b21456523c070030feb57", "bin/libggml-base.so.0.22.0": "277018e8a2c6653d5daa72a864d7b79f8cd82657fa28269689897a4b0082183a", "bin/libggml-cpu.so.0.22.0": "a5e48e3c50f174b25bf369f363db852dacd1f384a8e10dfaf18a392a2fd7aaaa", "bin/libggml-hip.so.0.22.0": "f60bfe62b7dfab548c5a30dcf7728f47b85f142c39ca9ae7377cc340b71c19d4", "bin/libggml.so.0.22.0": "0d80d5364e47fbf30304f754e0bc4dec514c8440e35e430f074a557344bd0df2", "bin/libgomp.so.1": "789f657765b7e6ea2bbc18408adf8bc123b1ea5b44ecb60f6673dd673effc953", "bin/libllama-bench-impl.so": "01feac4d93741273116ff4f5afc6e8398b3149983f0b68d94fa2511d60f7fd55", "bin/libllama-cli-impl.so": "6cad49e306929067c603477d53d56a5fcef13daef8ece5b30eeda804a11c3b17", "bin/libllama-common.so.0.3.0": "8961513d3c8d0d3df48e7197b4711a91c3f0da4da51db39b805a2825affcc3f4", "bin/libllama-server-impl.so": "c2d05f74db8da91d0e905778474d9bdfe554305441dece2117e12c4c617fddb7", "bin/libllama.so.0.3.0": "c29bec577633c8c0c0f0b08a2103b69135aff035aedd49eb70799310aef9576b", "bin/libmtmd.so.0.3.0": "fd0ce09eacd67eb8475facbe70d4cfe5d7a6e8593b3fe358f5857a2b43f65f3e", "bin/libstdc++.so.6": "91df7131df8e14c525527e436d7ea43f7d96ed880b5f3d650b09f8f9a4e53dd6", "bin/llama-bench": "0978408dcdd5d8a016d7072a6ea1f9466f2c5e637a6e6d61c45dc3ac6f9c3d84", "bin/llama-cli": "1276218095861368432ee7b8c9bbf8ddf58e74f16a204557747989ee4ca68b78", "bin/llama-server": "d522ca837e78a06347622b64d4e8f329c2bc1224675367331a20d561911aa5a8" }, "source_manifest": "evidence/inclusion-audit.json", "full_patch": "full.patch", "full_patch_sha256": "e4dd31d385dbe6e640be47cd185877d89cd3696272e80e2e38a58d6364377b3a", "incremental_patch": "evidence/retained-runtime.patch", "incremental_patch_sha256": "fad32d45623248c878bdeb7f470d8ab21839f44eabdd2a04694a318d697292df", "tested_performance": { "code_tg": [ 42.280053297306495, 42.314176153996854 ], "review_tg": [ 31.42178300879495, 31.314892421578826 ], "evidence": "evidence/CONTINUATION-RESULT.json", "milestone_50_plus_met": false }, "correctness": "22 full-vector production-shape captures plus same-policy MTP128 and balanced520 final-output equality. Graph-only acceptance counters equal; boundary correction improves review drafting.", "limits": [ "50+ TG remains unproven", "4K/8K prefixes within16K context and two fixed served prompts; not a full quality/long-session/multi-device release qualification", "Shared system RAM measured; separate VRAM counters unavailable", "Graph-only change +1.56% code/-0.34% review; do not advertise a large fixed6 speedup", "No new weight quantization or training in this lane" ], "lock_policy": "Read-only snapshot. Fork/copy for future changes; changed source, binary, environment or CLI settings must receive a new candidate/variant identity. No automatic updates from lane working trees.", "packaging_parent": "qwen38-ciru-rocm10-20260905-rc1", "launcher_sha256": "6178acd86bb5f9d1dc4300619f4dc798d2d71d3649c9f0e2232e4542c777eec5", "runtime_binary_delta_from_rc1": [] } } }, "laurent": { "protocol": { "classification": "local-custom HumanEval0-9 non-thinking MTP speed workload", "arm": "laurent", "host": "sozo", "command": [ "/srv/llm/work/agention-baseline-20260904/bin-native/llama-server", "-m", "/srv/llm/models/agentionai-Qwen3.8-Flash-Next-ROCmFP4-FAST-ad4c5717254a/Qwen3.8-Flash-Next-ROCmFP4-FAST-v2-ple16.gguf", "-md", "/srv/llm/models/agentionai-Qwen3.8-Flash-Next-MTP-ROCmFP4-FAST-5a2cf56c3e0f/Qwen3.8-Flash-Next-MTP-ROCmFP4-FAST.gguf", "--spec-type", "draft-mtp", "--spec-draft-adaptive", "--spec-draft-n-min", "2", "--spec-draft-n-max", "4", "-ngl", "99", "--n-gpu-layers-draft", "99", "-ctk", "q8_0", "-ctv", "q8_0", "-fa", "on", "--jinja", "--reasoning", "off", "--temp", "0", "--top-k", "1", "--top-p", "1", "--min-p", "0", "--host", "127.0.0.1", "--port", "18184", "--parallel", "1", "--metrics", "--slots", "--slot-save-path", "/srv/llm/work/qwen38-research-page-20260905/benchmarks/laurent/slots" ], "runtime": { "source": "https://github.com/LaurentZuijdwijk/llama.cpp", "branch": "vulkan/qwen4exp-rocmfpx", "commit": "5e085d123eead2e89b5c19f824fccb05727da6a2", "source_modified": false, "build": "Release; GGML_VULKAN=ON; GGML_NATIVE=ON; -march=native on identical Ryzen AI MAX+395; GCC15.2; glslc2026.1", "bin": "/srv/llm/work/agention-baseline-20260904/bin-native", "binary_sha256": { "agention-logit-capture": "e5e8fd13e0e30dd282c7e31c163a168ee5409779b95e57b2ffef3eb5d82db015", "libggml.so.0.22.0": "2bbc5d9d724471ce8ba2bd718f42540683937a516cbfd6e7a2db892142cae6c9", "libllama-bench-impl.so": "dd076a83339e15dd882511c4a572ea437769a1fbb9e0eac1431e1b79204ed375", "libggml-cpu.so.0.22.0": "3ef48c8227e760b0eef31a5cf8a9e304ba0e54da4c7bd25e1e85b8100170a287", "llama-perplexity": "18795d1a52d38afcbd98d9246e023a41254adaabb7bafa1432b9199c4e213f27", "libllama-common.so.0.3.0": "0ea75f4fbeddfb4f32df9551d351d2bd0070c3965386c41860de54abc803ce9c", "libllama.so.0.3.0": "ea96f618251bf27e91bbd01f6d88ec7c617b05bdb590696b33a19aed2fc039e4", "llama-server": "ee03ebdbfabd73ab96f4db975f64ee4c514ade22d7c91b5119a4c7298543cc23", "libllama-server-impl.so": "eaaf255d36c54f1792cceb9fc076139f3c7af082170dc464eb80b39585026e6a", "llama-bench": "0f86842c7c7d19be46715f92fd738b39457640d4a73f8564bb9e6e0bdf10db9f", "libggml-vulkan.so.0.22.0": "35027a5e4d4a16abbf03fb766179779801376d2706978f2802d5289b50302c5a", "libggml-base.so.0.22.0": "7c38040b141b7d992e5038fc41a0672a59e08f6da5100087d38f819db76efdaa", "libllama-perplexity-impl.so": "b4f4034405ea24ac288638dbd49958b250db9ad380243ae824ee399cc64c559d", "libmtmd.so.0.3.0": "77f181821ccb697b842e20ea17cbe373fe7a129c84a49c9bd24ed85230e94564" }, "model_receipt": { "state": "COMPLETE", "repo": "agentionai/Qwen3.8-Flash-Next-ROCmFP4-FAST-imatrix-GGUF", "revision": "ad4c5717254a630ee0c5a8db5208eb1f8476e56c", "files": [ { "file": ".gitattributes", "bytes": 3056, "sha256": "13abf0cb3eaf629e3431d353759d39dc05f6ee93be4e0be3659b998f3e83cd5f", "verified_utc": "2026-09-05T01:14:03.991766+00:00" }, { "file": "README.md", "bytes": 9020, "sha256": "2aa99e13ea2f3e4329692282ea6aed56f2e73e62364627d8c8209b0ca06d8bbf", "verified_utc": "2026-09-05T01:14:04.302070+00:00" }, { "file": "assets/throughput-vs-depth.png", "bytes": 185136, "sha256": "757755421061f21730f8251240e429e1ed2e1f683e277f81083c81146484c582", "verified_utc": "2026-09-05T01:14:04.930333+00:00" }, { "file": "mmproj/mmproj-Qwen3.8-Flash-Next-f16.gguf", "bytes": 904004128, "sha256": "f456cd796fdbdef0cadb22710b54e0071b6cdf22c07963365baa898267aec517", "verified_utc": "2026-09-05T01:14:12.918470+00:00" }, { "file": "Qwen3.8-Flash-Next-ROCmFP4-FAST-v2-ple16.gguf", "bytes": 93484237760, "sha256": "552a7a162f6a620c3aa0850d070086bc2b95094e0a4e8b860694c7f212cb59d8", "verified_utc": "2026-09-05T01:23:03.986191+00:00" } ], "completed_utc": "2026-09-05T01:23:03.986271+00:00" }, "draft_receipt": { "state": "COMPLETE", "repo": "agentionai/Qwen3.8-Flash-Next-MTP-ROCmFP4-FAST-GGUF", "revision": "5a2cf56c3e0f8bc8d395d53bec649ec8e358a993", "files": [ { "file": ".gitattributes", "bytes": 1596, "sha256": "2c8de7fbce4de3b2a94d3215ba4e3e33ebeb107c634b429d3b18e2713252d669", "verified_utc": "2026-09-05T03:33:09.932944+00:00" }, { "file": "README.md", "bytes": 2939, "sha256": "fa6de4146c9b31e3f96c2969d6a8c958681f36a92b28033f98f36c24bc908a18", "verified_utc": "2026-09-05T03:33:10.160120+00:00" }, { "file": "LICENSE", "bytes": 3235, "sha256": "a0dc422560841fd68e06d974907f8b4c709bca44a67daad2b528437bdf676c08", "verified_utc": "2026-09-05T03:33:10.377404+00:00" }, { "file": "Qwen3.8-Flash-Next-MTP-ROCmFP4-FAST.gguf", "bytes": 2444519296, "sha256": "5046d69571bb35c699e19c7a00b36633c2fdfdd06d70482551744ec523a6b590", "verified_utc": "2026-09-05T03:34:21.425457+00:00" } ], "completed_utc": "2026-09-05T03:34:21.425537+00:00" }, "environment": { "VK_ICD_FILENAMES": "/run/opengl-driver/share/vulkan/icd.d/radeon_icd.x86_64.json", "LD_LIBRARY_PATH": "/srv/llm/work/agention-baseline-20260904/bin-native:/nix/store/zs7y2aadk71bawprdcn000az9y05s8nf-vulkan-loader-1.4.341.0/lib" } }, "artifacts": { "state": "COMPLETE", "repo": "agentionai/Qwen3.8-Flash-Next-ROCmFP4-FAST-imatrix-GGUF", "revision": "ad4c5717254a630ee0c5a8db5208eb1f8476e56c", "files": [ { "file": ".gitattributes", "bytes": 3056, "sha256": "13abf0cb3eaf629e3431d353759d39dc05f6ee93be4e0be3659b998f3e83cd5f", "verified_utc": "2026-09-05T01:14:03.991766+00:00" }, { "file": "README.md", "bytes": 9020, "sha256": "2aa99e13ea2f3e4329692282ea6aed56f2e73e62364627d8c8209b0ca06d8bbf", "verified_utc": "2026-09-05T01:14:04.302070+00:00" }, { "file": "assets/throughput-vs-depth.png", "bytes": 185136, "sha256": "757755421061f21730f8251240e429e1ed2e1f683e277f81083c81146484c582", "verified_utc": "2026-09-05T01:14:04.930333+00:00" }, { "file": "mmproj/mmproj-Qwen3.8-Flash-Next-f16.gguf", "bytes": 904004128, "sha256": "f456cd796fdbdef0cadb22710b54e0071b6cdf22c07963365baa898267aec517", "verified_utc": "2026-09-05T01:14:12.918470+00:00" }, { "file": "Qwen3.8-Flash-Next-ROCmFP4-FAST-v2-ple16.gguf", "bytes": 93484237760, "sha256": "552a7a162f6a620c3aa0850d070086bc2b95094e0a4e8b860694c7f212cb59d8", "verified_utc": "2026-09-05T01:23:03.986191+00:00" } ], "completed_utc": "2026-09-05T01:23:03.986271+00:00" }, "verified_binary_hashes": { "agention-logit-capture": "e5e8fd13e0e30dd282c7e31c163a168ee5409779b95e57b2ffef3eb5d82db015", "libggml.so.0.22.0": "2bbc5d9d724471ce8ba2bd718f42540683937a516cbfd6e7a2db892142cae6c9", "libllama-bench-impl.so": "dd076a83339e15dd882511c4a572ea437769a1fbb9e0eac1431e1b79204ed375", "libggml-cpu.so.0.22.0": "3ef48c8227e760b0eef31a5cf8a9e304ba0e54da4c7bd25e1e85b8100170a287", "llama-perplexity": "18795d1a52d38afcbd98d9246e023a41254adaabb7bafa1432b9199c4e213f27", "libllama-common.so.0.3.0": "0ea75f4fbeddfb4f32df9551d351d2bd0070c3965386c41860de54abc803ce9c", "libllama.so.0.3.0": "ea96f618251bf27e91bbd01f6d88ec7c617b05bdb590696b33a19aed2fc039e4", "llama-server": "ee03ebdbfabd73ab96f4db975f64ee4c514ade22d7c91b5119a4c7298543cc23", "libllama-server-impl.so": "eaaf255d36c54f1792cceb9fc076139f3c7af082170dc464eb80b39585026e6a", "llama-bench": "0f86842c7c7d19be46715f92fd738b39457640d4a73f8564bb9e6e0bdf10db9f", "libggml-vulkan.so.0.22.0": "35027a5e4d4a16abbf03fb766179779801376d2706978f2802d5289b50302c5a", "libggml-base.so.0.22.0": "7c38040b141b7d992e5038fc41a0672a59e08f6da5100087d38f819db76efdaa", "libllama-perplexity-impl.so": "b4f4034405ea24ac288638dbd49958b250db9ad380243ae824ee399cc64c559d", "libmtmd.so.0.3.0": "77f181821ccb697b842e20ea17cbe373fe7a129c84a49c9bd24ed85230e94564" }, "dataset_sha256": "b796127e635a67f93fb35c04f4cb03cf06f38c8072ee7cee8833d7bee06979ef", "driver_sha256": "b9533cc94b2305d79153567443b895b806cecf24211b40d054a9a960c66df81c", "sampler": { "temperature": 0, "top_k": 1, "top_p": 1, "min_p": 0, "repeat_penalty": 1, "presence_penalty": 0, "frequency_penalty": 0, "seed": 123 }, "reasoning": false, "task_protocol": "Canonical HumanEval prompt in one user message, embedded chat template with enable_thinking=false; no added instructions. One first sample, no retries. Speed and acceptance only; no quality score.", "output_policy": "Natural EOS; n_predict=-1; no custom stops. 1800-second transport timeout. Preserve unsuccessful/incomplete requests.", "aggregate_policy": "TG=sum generated tokens / sum server decode seconds. Request wall=sum measured HTTP request wall seconds. Panel wall=elapsed loop time including recorder overhead, excluding model load.", "settings_basis": "CIRU locked retained d6 ROCm10 recipe at16K; Laurent publisher adaptive2\u20134 Q8 targetKV; Unsloth recommended fork sharedQ8 d2. Greedy non-thinking for MTP acceptance speed, supported by retained CIRU probe and published Unsloth greedy MTP recipe.", "governors": { "/sys/devices/system/cpu/cpu21/cpufreq/scaling_governor": "performance", "/sys/devices/system/cpu/cpu11/cpufreq/scaling_governor": "performance", "/sys/devices/system/cpu/cpu9/cpufreq/scaling_governor": "performance", "/sys/devices/system/cpu/cpu7/cpufreq/scaling_governor": "performance", "/sys/devices/system/cpu/cpu28/cpufreq/scaling_governor": "performance", "/sys/devices/system/cpu/cpu18/cpufreq/scaling_governor": "performance", "/sys/devices/system/cpu/cpu5/cpufreq/scaling_governor": "performance", "/sys/devices/system/cpu/cpu26/cpufreq/scaling_governor": "performance", "/sys/devices/system/cpu/cpu16/cpufreq/scaling_governor": "performance", "/sys/devices/system/cpu/cpu3/cpufreq/scaling_governor": "performance", "/sys/devices/system/cpu/cpu24/cpufreq/scaling_governor": "performance", "/sys/devices/system/cpu/cpu14/cpufreq/scaling_governor": "performance", "/sys/devices/system/cpu/cpu1/cpufreq/scaling_governor": "performance", "/sys/devices/system/cpu/cpu22/cpufreq/scaling_governor": "performance", "/sys/devices/system/cpu/cpu12/cpufreq/scaling_governor": "performance", "/sys/devices/system/cpu/cpu30/cpufreq/scaling_governor": "performance", "/sys/devices/system/cpu/cpu20/cpufreq/scaling_governor": "performance", "/sys/devices/system/cpu/cpu10/cpufreq/scaling_governor": "performance", "/sys/devices/system/cpu/cpu8/cpufreq/scaling_governor": "performance", "/sys/devices/system/cpu/cpu29/cpufreq/scaling_governor": "performance", "/sys/devices/system/cpu/cpu19/cpufreq/scaling_governor": "performance", "/sys/devices/system/cpu/cpu6/cpufreq/scaling_governor": "performance", "/sys/devices/system/cpu/cpu27/cpufreq/scaling_governor": "performance", "/sys/devices/system/cpu/cpu17/cpufreq/scaling_governor": "performance", "/sys/devices/system/cpu/cpu4/cpufreq/scaling_governor": "performance", "/sys/devices/system/cpu/cpu25/cpufreq/scaling_governor": "performance", "/sys/devices/system/cpu/cpu15/cpufreq/scaling_governor": "performance", "/sys/devices/system/cpu/cpu2/cpufreq/scaling_governor": "performance", "/sys/devices/system/cpu/cpu23/cpufreq/scaling_governor": "performance", "/sys/devices/system/cpu/cpu13/cpufreq/scaling_governor": "performance", "/sys/devices/system/cpu/cpu0/cpufreq/scaling_governor": "performance", "/sys/devices/system/cpu/cpu31/cpufreq/scaling_governor": "performance" }, "rendered_sha256": "2cc4c7910b4ad2e7f96e9f5f6ffda94e14ea458f110d302b2ff7a44358e92f2e", "template_sha256": "12827f24b742ea4e80cdc12dbcf9622227056b9f797252a3149263d4f9aaadce" }, "identity": { "arm": "laurent", "host": "sozo", "protocol_lock": "/srv/llm/work/qwen38-research-page-20260905/benchmarks/laurent/protocol.lock.json", "command": [ "/srv/llm/work/agention-baseline-20260904/bin-native/llama-server", "-m", "/srv/llm/models/agentionai-Qwen3.8-Flash-Next-ROCmFP4-FAST-ad4c5717254a/Qwen3.8-Flash-Next-ROCmFP4-FAST-v2-ple16.gguf", "-md", "/srv/llm/models/agentionai-Qwen3.8-Flash-Next-MTP-ROCmFP4-FAST-5a2cf56c3e0f/Qwen3.8-Flash-Next-MTP-ROCmFP4-FAST.gguf", "--spec-type", "draft-mtp", "--spec-draft-adaptive", "--spec-draft-n-min", "2", "--spec-draft-n-max", "4", "-ngl", "99", "--n-gpu-layers-draft", "99", "-ctk", "q8_0", "-ctv", "q8_0", "-fa", "on", "--jinja", "--reasoning", "off", "--temp", "0", "--top-k", "1", "--top-p", "1", "--min-p", "0", "--host", "127.0.0.1", "--port", "18184", "--parallel", "1", "--metrics", "--slots", "--slot-save-path", "/srv/llm/work/qwen38-research-page-20260905/benchmarks/laurent/slots" ], "maps": [ "78e61c600000-78e61c681000 r--p 00000000 103:04 30954711 /srv/llm/work/agention-baseline-20260904/bin-native/libggml-vulkan.so.0.22.0", "78e61c681000-78e61c95e000 r-xp 00081000 103:04 30954711 /srv/llm/work/agention-baseline-20260904/bin-native/libggml-vulkan.so.0.22.0", "78e61c95e000-78e6203af000 r--p 0035e000 103:04 30954711 /srv/llm/work/agention-baseline-20260904/bin-native/libggml-vulkan.so.0.22.0", "78e6203af000-78e6203b7000 r--p 03daf000 103:04 30954711 /srv/llm/work/agention-baseline-20260904/bin-native/libggml-vulkan.so.0.22.0", "78e6203b7000-78e6203b9000 rw-p 03db7000 103:04 30954711 /srv/llm/work/agention-baseline-20260904/bin-native/libggml-vulkan.so.0.22.0", "78e62043a000-78e62044d000 r--p 00000000 103:04 30954710 /srv/llm/work/agention-baseline-20260904/bin-native/libggml-cpu.so.0.22.0", "78e62044d000-78e620558000 r-xp 00013000 103:04 30954710 /srv/llm/work/agention-baseline-20260904/bin-native/libggml-cpu.so.0.22.0", "78e620558000-78e620578000 r--p 0011e000 103:04 30954710 /srv/llm/work/agention-baseline-20260904/bin-native/libggml-cpu.so.0.22.0", "78e620578000-78e62057a000 r--p 0013d000 103:04 30954710 /srv/llm/work/agention-baseline-20260904/bin-native/libggml-cpu.so.0.22.0", "78e62057a000-78e62057b000 rw-p 0013f000 103:04 30954710 /srv/llm/work/agention-baseline-20260904/bin-native/libggml-cpu.so.0.22.0", "78e620600000-78e6206ae000 r--p 00000000 103:04 30954717 /srv/llm/work/agention-baseline-20260904/bin-native/libllama.so.0.3.0", "78e6206ae000-78e620935000 r-xp 000ae000 103:04 30954717 /srv/llm/work/agention-baseline-20260904/bin-native/libllama.so.0.3.0", "78e620935000-78e6209b4000 r--p 00335000 103:04 30954717 /srv/llm/work/agention-baseline-20260904/bin-native/libllama.so.0.3.0", "78e6209b4000-78e6209c1000 r--p 003b4000 103:04 30954717 /srv/llm/work/agention-baseline-20260904/bin-native/libllama.so.0.3.0", "78e6209c1000-78e6209c6000 rw-p 003c1000 103:04 30954717 /srv/llm/work/agention-baseline-20260904/bin-native/libllama.so.0.3.0", "78e620c00000-78e620caa000 r--p 00000000 103:04 30954714 /srv/llm/work/agention-baseline-20260904/bin-native/libllama-common.so.0.3.0", "78e620caa000-78e620f8d000 r-xp 000aa000 103:04 30954714 /srv/llm/work/agention-baseline-20260904/bin-native/libllama-common.so.0.3.0", "78e620f8d000-78e62102b000 r--p 0038d000 103:04 30954714 /srv/llm/work/agention-baseline-20260904/bin-native/libllama-common.so.0.3.0", "78e62102b000-78e621033000 r--p 0042b000 103:04 30954714 /srv/llm/work/agention-baseline-20260904/bin-native/libllama-common.so.0.3.0", "78e621033000-78e621038000 rw-p 00433000 103:04 30954714 /srv/llm/work/agention-baseline-20260904/bin-native/libllama-common.so.0.3.0", "78e62111b000-78e621131000 r--p 00000000 103:04 30954709 /srv/llm/work/agention-baseline-20260904/bin-native/libggml-base.so.0.22.0", "78e621131000-78e6211d9000 r-xp 00016000 103:04 30954709 /srv/llm/work/agention-baseline-20260904/bin-native/libggml-base.so.0.22.0", "78e6211d9000-78e6211fb000 r--p 000be000 103:04 30954709 /srv/llm/work/agention-baseline-20260904/bin-native/libggml-base.so.0.22.0", "78e6211fb000-78e6211fe000 r--p 000df000 103:04 30954709 /srv/llm/work/agention-baseline-20260904/bin-native/libggml-base.so.0.22.0", "78e6211fe000-78e621200000 rw-p 000e2000 103:04 30954709 /srv/llm/work/agention-baseline-20260904/bin-native/libggml-base.so.0.22.0", "78e621200000-78e62127d000 r--p 00000000 103:04 30954716 /srv/llm/work/agention-baseline-20260904/bin-native/libllama-server-impl.so", "78e62127d000-78e62147a000 r-xp 0007d000 103:04 30954716 /srv/llm/work/agention-baseline-20260904/bin-native/libllama-server-impl.so", "78e62147a000-78e6217c9000 r--p 0027a000 103:04 30954716 /srv/llm/work/agention-baseline-20260904/bin-native/libllama-server-impl.so", "78e6217c9000-78e6217ce000 r--p 005c9000 103:04 30954716 /srv/llm/work/agention-baseline-20260904/bin-native/libllama-server-impl.so", "78e6217ce000-78e6217d2000 rw-p 005ce000 103:04 30954716 /srv/llm/work/agention-baseline-20260904/bin-native/libllama-server-impl.so" ], "pid": 321253, "slots": [ { "id": 0, "n_ctx": 262144, "speculative": true, "is_processing": false } ], "props": { "default_generation_settings": { "params": { "seed": 4294967295, "temperature": 0.0, "dynatemp_range": 0.0, "dynatemp_exponent": 1.0, "top_k": 1, "top_p": 1.0, "min_p": 0.0, "top_n_sigma": -1.0, "xtc_probability": 0.0, "xtc_threshold": 0.10000000149011612, "typical_p": 1.0, "repeat_last_n": 64, "repeat_penalty": 1.0, "presence_penalty": 0.0, "frequency_penalty": 0.0, "dry_multiplier": 0.0, "dry_base": 1.75, "dry_allowed_length": 2, "dry_penalty_last_n": 64, "mirostat": 0, "mirostat_tau": 5.0, "mirostat_eta": 0.10000000149011612, "adaptive_target": -1.0, "adaptive_decay": 0.8999999761581421, "max_tokens": -1, "n_predict": -1, "n_keep": 0, "n_discard": 0, "ignore_eos": false, "stream": false, "n_probs": 0, "min_keep": 0, "chat_format": "Content-only", "reasoning_format": "none", "reasoning_in_content": false, "generation_prompt": "", "samplers": [ "penalties", "dry", "top_n_sigma", "top_k", "typ_p", "top_p", "min_p", "xtc", "temperature" ], "speculative.types": "none", "timings_per_token": false, "post_sampling_probs": false, "backend_sampling": false, "lora": [] }, "n_ctx": 262144 }, "total_slots": 1, "model_alias": "/srv/llm/models/agentionai-Qwen3.8-Flash-Next-ROCmFP4-FAST-ad4c5717254a/Qwen3.8-Flash-Next-ROCmFP4-FAST-v2-ple16.gguf", "model_ftype": "Q4_0_ROCMFP4_FAST", "model_path": "/srv/llm/models/agentionai-Qwen3.8-Flash-Next-ROCmFP4-FAST-ad4c5717254a/Qwen3.8-Flash-Next-ROCmFP4-FAST-v2-ple16.gguf", "modalities": { "vision": false, "video": false, "audio": false }, "media_marker": "<__media_anicSXLOq1OdvsgwoV4pWJv8ZTVkl8LO__>", "endpoint_slots": true, "endpoint_props": false, "endpoint_metrics": true, "ui": true, "ui_settings": {}, "chat_template": "{%- set image_count = namespace(value=0) %}\n{%- set video_count = namespace(value=0) %}\n{%- macro render_content(content, do_vision_count, is_system_content=false) %}\n {%- if content is string %}\n {{- content }}\n {%- elif content is iterable and content is not mapping %}\n {%- for item in content %}\n {%- if 'image' in item or 'image_url' in item or item.type == 'image' %}\n {%- if is_system_content %}\n {{- raise_exception('System message cannot contain images.') }}\n {%- endif %}\n {%- if do_vision_count %}\n {%- set image_count.value = image_count.value + 1 %}\n {%- endif %}\n {%- if add_vision_id %}\n {{- 'Picture ' ~ image_count.value ~ ': ' }}\n {%- endif %}\n {{- '<|vision_start|><|image_pad|><|vision_end|>' }}\n {%- elif 'video' in item or item.type == 'video' %}\n {%- if is_system_content %}\n {{- raise_exception('System message cannot contain videos.') }}\n {%- endif %}\n {%- if do_vision_count %}\n {%- set video_count.value = video_count.value + 1 %}\n {%- endif %}\n {%- if add_vision_id %}\n {{- 'Video ' ~ video_count.value ~ ': ' }}\n {%- endif %}\n {{- '<|vision_start|><|video_pad|><|vision_end|>' }}\n {%- elif 'text' in item %}\n {{- item.text }}\n {%- else %}\n {{- raise_exception('Unexpected item type in content.') }}\n {%- endif %}\n {%- endfor %}\n {%- elif content is none or content is undefined %}\n {{- '' }}\n {%- else %}\n {{- raise_exception('Unexpected content type.') }}\n {%- endif %}\n{%- endmacro %}\n{%- if not messages %}\n {{- raise_exception('No messages provided.') }}\n{%- endif %}\n{%- set sysns = namespace(count=0, text='') %}\n{%- for message in messages %}\n {%- if sysns.count == loop.index0 and (message.role == 'system' or message.role == 'developer') %}\n {%- set sys_content = render_content(message.content, false, true)|trim %}\n {%- if sys_content %}\n {%- set sysns.text = sysns.text + ('\\n' if sysns.text else '') + sys_content %}\n {%- endif %}\n {%- set sysns.count = sysns.count + 1 %}\n {%- endif %}\n{%- endfor %}\n{%- set num_sys = sysns.count %}\n{%- set merged_system = sysns.text %}\n{%- set reasoning_instructions = '' %}\n{%- if enable_thinking is undefined or enable_thinking is true %}\n {%- set resolved_reasoning_effort = reasoning_effort|default('xhigh') %}\n {%- if resolved_reasoning_effort == 'high' %}\n {%- set resolved_reasoning_effort = 'xhigh' %}\n {%- endif %}\n {%- if resolved_reasoning_effort not in ('xhigh', 'medium', 'low') %}\n {{- raise_exception('Unexpected reasoning effort ' ~ reasoning_effort ~ '. Supported types are xhigh (default), medium, and low.') }}\n {%- endif %}\n {%- if resolved_reasoning_effort == 'xhigh' %}\n {%- set reasoning_instructions = 'Reasoning effort is set to xhigh. Please think carefully through the task, validate key assumptions, consider plausible alternatives, and prioritize correctness, consistency, and clarity in the final answer.' %}\n {%- elif resolved_reasoning_effort == 'low' %}\n {%- set reasoning_instructions = 'Reasoning effort is set to low. Keep your thinking brief and focused, moving directly to the conclusion without unnecessary elaboration.' %}\n {%- endif %}\n{%- endif %}\n{%- if tools and tools is iterable and tools is not mapping %}\n {{- '<|im_start|>system\\n' }}\n {%- if reasoning_instructions %}\n {{- reasoning_instructions + '\\n\\n' }}\n {%- endif %}\n {{- \"# Tools\\n\\nYou have access to the following functions:\\n\\n\" }}\n {%- for tool in tools %}\n {{- \"\\n\" }}\n {{- tool | tojson }}\n {%- endfor %}\n {{- \"\\n\" }}\n {{- '\\n\\nIf you choose to call a function ONLY reply in the following format with NO suffix:\\n\\n\\n\\n\\nvalue_1\\n\\n\\nThis is the value for the second parameter\\nthat can span\\nmultiple lines\\n\\n\\n\\n\\n\\nReminder:\\n- Function calls MUST follow the specified format: an inner block must be nested within XML tags\\n- Required parameters MUST be specified\\n- You may provide optional reasoning for your function call in natural language BEFORE the function call, but NOT after\\n- If there is no function call available, answer the question like normal with your current knowledge and do not tell the user about function calls\\n' }}\n {%- if merged_system %}\n {{- '\\n\\n' + merged_system }}\n {%- endif %}\n {{- '<|im_end|>\\n' }}\n{%- else %}\n {%- if merged_system %}\n {{- '<|im_start|>system\\n' + (reasoning_instructions + '\\n\\n' if reasoning_instructions else '') + merged_system + '<|im_end|>\\n' }}\n {%- elif reasoning_instructions %}\n {{- '<|im_start|>system\\n' + reasoning_instructions + '<|im_end|>\\n' }}\n {%- endif %}\n{%- endif %}\n{%- set ns = namespace(multi_step_tool=true, last_query_index=messages|length - 1) %}\n{%- for message in messages[::-1] %}\n {%- set index = (messages|length - 1) - loop.index0 %}\n {%- if ns.multi_step_tool and message.role == \"user\" %}\n {%- set content = render_content(message.content, false)|trim %}\n {%- if not(content.startswith('') and content.endswith('')) %}\n {%- set ns.multi_step_tool = false %}\n {%- set ns.last_query_index = index %}\n {%- endif %}\n {%- endif %}\n{%- endfor %}\n{%- for message in messages %}\n {%- if loop.index0 >= num_sys %}\n {%- set content = render_content(message.content, true)|trim %}\n {%- if message.role == \"system\" or message.role == \"developer\" %}\n {{- raise_exception('System message must be at the beginning.') }}\n {%- elif message.role == \"user\" %}\n {{- '<|im_start|>' + message.role + '\\n' + content + '<|im_end|>' + '\\n' }}\n {%- elif message.role == \"assistant\" %}\n {%- set reasoning_content = '' %}\n {%- if message.reasoning_content is string %}\n {%- set reasoning_content = message.reasoning_content %}\n {%- endif %}\n {%- set reasoning_content = reasoning_content|trim %}\n {%- if preserve_thinking is undefined or preserve_thinking is true or loop.index0 > ns.last_query_index %}\n {{- '<|im_start|>' + message.role + '\\n\\n' + reasoning_content + '\\n\\n\\n' + content }}\n {%- else %}\n {{- '<|im_start|>' + message.role + '\\n' + content }}\n {%- endif %}\n {%- if message.tool_calls and message.tool_calls is iterable and message.tool_calls is not mapping %}\n {%- for tool_call in message.tool_calls %}\n {%- if tool_call.function is defined %}\n {%- set tool_call = tool_call.function %}\n {%- endif %}\n {%- if tool_call.name is not defined or tool_call.name is none %}\n {{- raise_exception('Tool call is missing a function name.') }}\n {%- endif %}\n {%- if loop.first %}\n {%- if content|trim %}\n {{- '\\n\\n\\n\\n' }}\n {%- else %}\n {{- '\\n\\n' }}\n {%- endif %}\n {%- else %}\n {{- '\\n\\n\\n' }}\n {%- endif %}\n {%- if tool_call.arguments is mapping %}\n {%- for args_name, args_value in tool_call.arguments|items %}\n {{- '\\n' }}\n {%- set args_value = args_value | string if args_value is string else args_value | tojson | safe %}\n {{- args_value }}\n {{- '\\n\\n' }}\n {%- endfor %}\n {%- elif tool_call.arguments is string %}\n {%- if tool_call.arguments|trim %}\n {{- raise_exception('Tool call arguments for function \"' + (tool_call.name | string) + '\" were passed as a JSON string. Parse them into an object before calling apply_chat_template.') }}\n {%- endif %}\n {%- elif tool_call.arguments is defined and tool_call.arguments is not none %}\n {{- raise_exception('Tool call arguments for function \"' + (tool_call.name | string) + '\" must be an object/mapping or a JSON string.') }}\n {%- endif %}\n {{- '\\n' }}\n {%- endfor %}\n {%- endif %}\n {{- '<|im_end|>\\n' }}\n {%- elif message.role == \"tool\" %}\n {%- if loop.previtem and loop.previtem.role != \"tool\" %}\n {{- '<|im_start|>user' }}\n {%- endif %}\n {{- '\\n\\n' }}\n {{- content }}\n {{- '\\n' }}\n {%- if not loop.last and loop.nextitem.role != \"tool\" %}\n {{- '<|im_end|>\\n' }}\n {%- elif loop.last %}\n {{- '<|im_end|>\\n' }}\n {%- endif %}\n {%- else %}\n {{- raise_exception('Unexpected message role.') }}\n {%- endif %}\n {%- endif %}\n{%- endfor %}\n{%- if add_generation_prompt %}\n {{- '<|im_start|>assistant\\n' }}\n {%- if enable_thinking is defined and enable_thinking is false %}\n {{- '\\n\\n\\n\\n' }}\n {%- else %}\n {{- '\\n' }}\n {%- endif %}\n{%- endif %}\n{#- Unsloth fixes - developer role, merged system messages, tool calling #}", "chat_template_caps": { "supports_object_arguments": true, "supports_parallel_tool_calls": true, "supports_preserve_reasoning": true, "supports_reasoning_effort": true, "supports_string_content": true, "supports_system_role": true, "supports_tool_calls": true, "supports_tools": true, "supports_typed_content": false }, "bos_token": "<|endoftext|>", "eos_token": "<|im_end|>", "build_info": "b1-5e085d1", "is_sleeping": false, "cors_proxy_enabled": false }, "runtime": { "source": "https://github.com/LaurentZuijdwijk/llama.cpp", "branch": "vulkan/qwen4exp-rocmfpx", "commit": "5e085d123eead2e89b5c19f824fccb05727da6a2", "source_modified": false, "build": "Release; GGML_VULKAN=ON; GGML_NATIVE=ON; -march=native on identical Ryzen AI MAX+395; GCC15.2; glslc2026.1", "bin": "/srv/llm/work/agention-baseline-20260904/bin-native", "binary_sha256": { "agention-logit-capture": "e5e8fd13e0e30dd282c7e31c163a168ee5409779b95e57b2ffef3eb5d82db015", "libggml.so.0.22.0": "2bbc5d9d724471ce8ba2bd718f42540683937a516cbfd6e7a2db892142cae6c9", "libllama-bench-impl.so": "dd076a83339e15dd882511c4a572ea437769a1fbb9e0eac1431e1b79204ed375", "libggml-cpu.so.0.22.0": "3ef48c8227e760b0eef31a5cf8a9e304ba0e54da4c7bd25e1e85b8100170a287", "llama-perplexity": "18795d1a52d38afcbd98d9246e023a41254adaabb7bafa1432b9199c4e213f27", "libllama-common.so.0.3.0": "0ea75f4fbeddfb4f32df9551d351d2bd0070c3965386c41860de54abc803ce9c", "libllama.so.0.3.0": "ea96f618251bf27e91bbd01f6d88ec7c617b05bdb590696b33a19aed2fc039e4", "llama-server": "ee03ebdbfabd73ab96f4db975f64ee4c514ade22d7c91b5119a4c7298543cc23", "libllama-server-impl.so": "eaaf255d36c54f1792cceb9fc076139f3c7af082170dc464eb80b39585026e6a", "llama-bench": "0f86842c7c7d19be46715f92fd738b39457640d4a73f8564bb9e6e0bdf10db9f", "libggml-vulkan.so.0.22.0": "35027a5e4d4a16abbf03fb766179779801376d2706978f2802d5289b50302c5a", "libggml-base.so.0.22.0": "7c38040b141b7d992e5038fc41a0672a59e08f6da5100087d38f819db76efdaa", "libllama-perplexity-impl.so": "b4f4034405ea24ac288638dbd49958b250db9ad380243ae824ee399cc64c559d", "libmtmd.so.0.3.0": "77f181821ccb697b842e20ea17cbe373fe7a129c84a49c9bd24ed85230e94564" }, "model_receipt": { "state": "COMPLETE", "repo": "agentionai/Qwen3.8-Flash-Next-ROCmFP4-FAST-imatrix-GGUF", "revision": "ad4c5717254a630ee0c5a8db5208eb1f8476e56c", "files": [ { "file": ".gitattributes", "bytes": 3056, "sha256": "13abf0cb3eaf629e3431d353759d39dc05f6ee93be4e0be3659b998f3e83cd5f", "verified_utc": "2026-09-05T01:14:03.991766+00:00" }, { "file": "README.md", "bytes": 9020, "sha256": "2aa99e13ea2f3e4329692282ea6aed56f2e73e62364627d8c8209b0ca06d8bbf", "verified_utc": "2026-09-05T01:14:04.302070+00:00" }, { "file": "assets/throughput-vs-depth.png", "bytes": 185136, "sha256": "757755421061f21730f8251240e429e1ed2e1f683e277f81083c81146484c582", "verified_utc": "2026-09-05T01:14:04.930333+00:00" }, { "file": "mmproj/mmproj-Qwen3.8-Flash-Next-f16.gguf", "bytes": 904004128, "sha256": "f456cd796fdbdef0cadb22710b54e0071b6cdf22c07963365baa898267aec517", "verified_utc": "2026-09-05T01:14:12.918470+00:00" }, { "file": "Qwen3.8-Flash-Next-ROCmFP4-FAST-v2-ple16.gguf", "bytes": 93484237760, "sha256": "552a7a162f6a620c3aa0850d070086bc2b95094e0a4e8b860694c7f212cb59d8", "verified_utc": "2026-09-05T01:23:03.986191+00:00" } ], "completed_utc": "2026-09-05T01:23:03.986271+00:00" }, "draft_receipt": { "state": "COMPLETE", "repo": "agentionai/Qwen3.8-Flash-Next-MTP-ROCmFP4-FAST-GGUF", "revision": "5a2cf56c3e0f8bc8d395d53bec649ec8e358a993", "files": [ { "file": ".gitattributes", "bytes": 1596, "sha256": "2c8de7fbce4de3b2a94d3215ba4e3e33ebeb107c634b429d3b18e2713252d669", "verified_utc": "2026-09-05T03:33:09.932944+00:00" }, { "file": "README.md", "bytes": 2939, "sha256": "fa6de4146c9b31e3f96c2969d6a8c958681f36a92b28033f98f36c24bc908a18", "verified_utc": "2026-09-05T03:33:10.160120+00:00" }, { "file": "LICENSE", "bytes": 3235, "sha256": "a0dc422560841fd68e06d974907f8b4c709bca44a67daad2b528437bdf676c08", "verified_utc": "2026-09-05T03:33:10.377404+00:00" }, { "file": "Qwen3.8-Flash-Next-MTP-ROCmFP4-FAST.gguf", "bytes": 2444519296, "sha256": "5046d69571bb35c699e19c7a00b36633c2fdfdd06d70482551744ec523a6b590", "verified_utc": "2026-09-05T03:34:21.425457+00:00" } ], "completed_utc": "2026-09-05T03:34:21.425537+00:00" }, "environment": { "VK_ICD_FILENAMES": "/run/opengl-driver/share/vulkan/icd.d/radeon_icd.x86_64.json", "LD_LIBRARY_PATH": "/srv/llm/work/agention-baseline-20260904/bin-native:/nix/store/zs7y2aadk71bawprdcn000az9y05s8nf-vulkan-loader-1.4.341.0/lib" } } } }, "unsloth": { "protocol": { "classification": "local-custom HumanEval0-9 non-thinking MTP speed workload", "arm": "unsloth", "host": "sozo", "command": [ "/srv/llm/work/unsloth-runtime-compare-20260905/bin-unsloth/llama-server", "-m", "/srv/llm/models/unsloth-Qwen3.8-Flash-Next-38bb39ee9782/UD-IQ4_XS/Qwen3.8-Flash-Next-UD-IQ4_XS-00001-of-00003.gguf", "-md", "/srv/llm/models/unsloth-Qwen3.8-Flash-Next-38bb39ee9782/MTP/mtp-Qwen3.8-Flash-Next-shared-Q8_0.gguf", "--spec-type", "draft-mtp", "--spec-draft-n-max", "2", "-ngl", "999", "--jinja", "--reasoning", "off", "--temp", "0", "--top-k", "1", "--top-p", "1", "--min-p", "0", "--host", "127.0.0.1", "--port", "18184", "--parallel", "1", "--metrics", "--slots", "--slot-save-path", "/srv/llm/work/qwen38-research-page-20260905/benchmarks/unsloth/slots" ], "runtime": { "commit": "d1a92352cbd417fd840b4e765c0b82f5fe3d1d89", "path": "/srv/llm/work/unsloth-runtime-compare-20260905/bin-unsloth", "build": "native Release Vulkan GCC15.2", "sha256": { "libggml.so.0.22.0": "98a1577666dbc408ea7e0ca53ecf131c9fb2e0fdc5aa722d700f3f6d32de364f", "libllama-bench-impl.so": "007fd5258bcbe3df3bcd3a920e5ef0d751256b706a03edfb0c325bafa2a3f549", "libggml-cpu.so.0.22.0": "e9b2cc40e27e87ebaab290b67b69f7e6148483a7310b8efc5a477b45df985fa8", "libllama-common.so.0.3.0": "f5e60bf4529920ed0e58099e5540da46857f2c6b4bbcd4112c0ab6f801b6d1ae", "libllama.so.0.3.0": "2b772ffdfe672a28faaa00f313bf2a407c0a26a09222c64b6437349bf2874c59", "llama-server": "f16f7acc3bcbf035166df093470797e3ffcc7bd05fc09570994bbff8ba949f69", "libllama-server-impl.so": "c8058db9aec86b1d7a9326881e10d95b9d529f4476d19bb69635c33f13b97bde", "logit-capture": "9fdff72b74e326b802228fdf2b35be3516d71b3bdf6238e4a408caa65a45b6fc", "llama-bench": "b28863f4bffd5dc210745e8d154824bd121fc6c1daa075c3a99de9c82233d228", "libggml-vulkan.so.0.22.0": "43a9fd1608f68f9a8358b1e64911c1ebd39583e07c2ff6677d6673ae9da7393a", "libggml-base.so.0.22.0": "b18eb78d5e212593aa3ad509892abba5fc6fab77cbb85241b5742443ff47f82e", "libmtmd.so.0.3.0": "1846d70cc91fe4af957fc4562ab2e18993a7e8cbd17af7fbd2f4d5fcd8d39cc3" } }, "artifacts": { "state": "COMPLETE", "repo": "unsloth/Qwen3.8-Flash-Next-GGUF", "revision": "38bb39ee97821de2c9009abb7e93950eec396e66", "files": [ { "file": "README.md", "bytes": 59183, "sha256": "9538227ff778ba7d8ba61b42cde3927487bf4c9c647ce18ac31d08a0b42998e6", "verified_utc": "2026-09-05T05:04:33.439844+00:00" }, { "file": "MTP/README.md", "bytes": 5906, "sha256": "18a8c2d894bb0b63e31cf8aec0fceeee30564fa3e8115fee81e94ec73027fd63", "verified_utc": "2026-09-05T05:04:33.728653+00:00" }, { "file": "UD-IQ4_XS/Qwen3.8-Flash-Next-UD-IQ4_XS-00001-of-00003.gguf", "bytes": 10946624, "sha256": "5ce89370720f8bf90890f439361282104c1aa1482d4013bb9a50923e758e71a4", "verified_utc": "2026-09-05T05:04:34.952622+00:00" }, { "file": "UD-IQ4_XS/Qwen3.8-Flash-Next-UD-IQ4_XS-00002-of-00003.gguf", "bytes": 49835229856, "sha256": "577a38a2392b40ca2193cea502e1d92f60b8cd370675d308e0ec21885d9daaa7", "verified_utc": "2026-09-05T05:11:01.301186+00:00" }, { "file": "UD-IQ4_XS/Qwen3.8-Flash-Next-UD-IQ4_XS-00003-of-00003.gguf", "bytes": 43836407744, "sha256": "d4634e6d84f0ebb0940be15c90d3790bf6464e3dea3a1cddc567dc0e83ad8833", "verified_utc": "2026-09-05T05:19:54.395294+00:00" }, { "file": "MTP/mtp-Qwen3.8-Flash-Next-Q8_0.gguf", "bytes": 4137429120, "sha256": "cd87e5d1a4dadaeed63e35929f3b2f28d13e081b4cd32e00f2835095ec09351e", "verified_utc": "2026-09-05T05:20:59.165207+00:00" }, { "file": "MTP/mtp-Qwen3.8-Flash-Next-shared-Q8_0.gguf", "bytes": 2786568256, "sha256": "5ff54097406a905cf3a724c709124ceb0e3e10235ee862298969e91c96fa96e6", "verified_utc": "2026-09-05T05:21:22.474475+00:00" } ], "completed_utc": "2026-09-05T05:21:22.474550+00:00" }, "verified_binary_hashes": { "libggml.so.0.22.0": "98a1577666dbc408ea7e0ca53ecf131c9fb2e0fdc5aa722d700f3f6d32de364f", "libllama-bench-impl.so": "007fd5258bcbe3df3bcd3a920e5ef0d751256b706a03edfb0c325bafa2a3f549", "libggml-cpu.so.0.22.0": "e9b2cc40e27e87ebaab290b67b69f7e6148483a7310b8efc5a477b45df985fa8", "libllama-common.so.0.3.0": "f5e60bf4529920ed0e58099e5540da46857f2c6b4bbcd4112c0ab6f801b6d1ae", "libllama.so.0.3.0": "2b772ffdfe672a28faaa00f313bf2a407c0a26a09222c64b6437349bf2874c59", "llama-server": "f16f7acc3bcbf035166df093470797e3ffcc7bd05fc09570994bbff8ba949f69", "libllama-server-impl.so": "c8058db9aec86b1d7a9326881e10d95b9d529f4476d19bb69635c33f13b97bde", "logit-capture": "9fdff72b74e326b802228fdf2b35be3516d71b3bdf6238e4a408caa65a45b6fc", "llama-bench": "b28863f4bffd5dc210745e8d154824bd121fc6c1daa075c3a99de9c82233d228", "libggml-vulkan.so.0.22.0": "43a9fd1608f68f9a8358b1e64911c1ebd39583e07c2ff6677d6673ae9da7393a", "libggml-base.so.0.22.0": "b18eb78d5e212593aa3ad509892abba5fc6fab77cbb85241b5742443ff47f82e", "libmtmd.so.0.3.0": "1846d70cc91fe4af957fc4562ab2e18993a7e8cbd17af7fbd2f4d5fcd8d39cc3" }, "dataset_sha256": "b796127e635a67f93fb35c04f4cb03cf06f38c8072ee7cee8833d7bee06979ef", "driver_sha256": "b9533cc94b2305d79153567443b895b806cecf24211b40d054a9a960c66df81c", "sampler": { "temperature": 0, "top_k": 1, "top_p": 1, "min_p": 0, "repeat_penalty": 1, "presence_penalty": 0, "frequency_penalty": 0, "seed": 123 }, "reasoning": false, "task_protocol": "Canonical HumanEval prompt in one user message, embedded chat template with enable_thinking=false; no added instructions. One first sample, no retries. Speed and acceptance only; no quality score.", "output_policy": "Natural EOS; n_predict=-1; no custom stops. 1800-second transport timeout. Preserve unsuccessful/incomplete requests.", "aggregate_policy": "TG=sum generated tokens / sum server decode seconds. Request wall=sum measured HTTP request wall seconds. Panel wall=elapsed loop time including recorder overhead, excluding model load.", "settings_basis": "CIRU locked retained d6 ROCm10 recipe at16K; Laurent publisher adaptive2\u20134 Q8 targetKV; Unsloth recommended fork sharedQ8 d2. Greedy non-thinking for MTP acceptance speed, supported by retained CIRU probe and published Unsloth greedy MTP recipe.", "governors": { "/sys/devices/system/cpu/cpu21/cpufreq/scaling_governor": "performance", "/sys/devices/system/cpu/cpu11/cpufreq/scaling_governor": "performance", "/sys/devices/system/cpu/cpu9/cpufreq/scaling_governor": "performance", "/sys/devices/system/cpu/cpu7/cpufreq/scaling_governor": "performance", "/sys/devices/system/cpu/cpu28/cpufreq/scaling_governor": "performance", "/sys/devices/system/cpu/cpu18/cpufreq/scaling_governor": "performance", "/sys/devices/system/cpu/cpu5/cpufreq/scaling_governor": "performance", "/sys/devices/system/cpu/cpu26/cpufreq/scaling_governor": "performance", "/sys/devices/system/cpu/cpu16/cpufreq/scaling_governor": "performance", "/sys/devices/system/cpu/cpu3/cpufreq/scaling_governor": "performance", "/sys/devices/system/cpu/cpu24/cpufreq/scaling_governor": "performance", "/sys/devices/system/cpu/cpu14/cpufreq/scaling_governor": "performance", "/sys/devices/system/cpu/cpu1/cpufreq/scaling_governor": "performance", "/sys/devices/system/cpu/cpu22/cpufreq/scaling_governor": "performance", "/sys/devices/system/cpu/cpu12/cpufreq/scaling_governor": "performance", "/sys/devices/system/cpu/cpu30/cpufreq/scaling_governor": "performance", "/sys/devices/system/cpu/cpu20/cpufreq/scaling_governor": "performance", "/sys/devices/system/cpu/cpu10/cpufreq/scaling_governor": "performance", "/sys/devices/system/cpu/cpu8/cpufreq/scaling_governor": "performance", "/sys/devices/system/cpu/cpu29/cpufreq/scaling_governor": "performance", "/sys/devices/system/cpu/cpu19/cpufreq/scaling_governor": "performance", "/sys/devices/system/cpu/cpu6/cpufreq/scaling_governor": "performance", "/sys/devices/system/cpu/cpu27/cpufreq/scaling_governor": "performance", "/sys/devices/system/cpu/cpu17/cpufreq/scaling_governor": "performance", "/sys/devices/system/cpu/cpu4/cpufreq/scaling_governor": "performance", "/sys/devices/system/cpu/cpu25/cpufreq/scaling_governor": "performance", "/sys/devices/system/cpu/cpu15/cpufreq/scaling_governor": "performance", "/sys/devices/system/cpu/cpu2/cpufreq/scaling_governor": "performance", "/sys/devices/system/cpu/cpu23/cpufreq/scaling_governor": "performance", "/sys/devices/system/cpu/cpu13/cpufreq/scaling_governor": "performance", "/sys/devices/system/cpu/cpu0/cpufreq/scaling_governor": "performance", "/sys/devices/system/cpu/cpu31/cpufreq/scaling_governor": "performance" }, "rendered_sha256": "2cc4c7910b4ad2e7f96e9f5f6ffda94e14ea458f110d302b2ff7a44358e92f2e", "template_sha256": "12827f24b742ea4e80cdc12dbcf9622227056b9f797252a3149263d4f9aaadce" }, "identity": { "arm": "unsloth", "host": "sozo", "protocol_lock": "/srv/llm/work/qwen38-research-page-20260905/benchmarks/unsloth/protocol.lock.json", "command": [ "/srv/llm/work/unsloth-runtime-compare-20260905/bin-unsloth/llama-server", "-m", "/srv/llm/models/unsloth-Qwen3.8-Flash-Next-38bb39ee9782/UD-IQ4_XS/Qwen3.8-Flash-Next-UD-IQ4_XS-00001-of-00003.gguf", "-md", "/srv/llm/models/unsloth-Qwen3.8-Flash-Next-38bb39ee9782/MTP/mtp-Qwen3.8-Flash-Next-shared-Q8_0.gguf", "--spec-type", "draft-mtp", "--spec-draft-n-max", "2", "-ngl", "999", "--jinja", "--reasoning", "off", "--temp", "0", "--top-k", "1", "--top-p", "1", "--min-p", "0", "--host", "127.0.0.1", "--port", "18184", "--parallel", "1", "--metrics", "--slots", "--slot-save-path", "/srv/llm/work/qwen38-research-page-20260905/benchmarks/unsloth/slots" ], "maps": [ "74f493e00000-74f493e72000 r--p 00000000 103:04 30954756 /srv/llm/work/unsloth-runtime-compare-20260905/bin-unsloth/libggml-vulkan.so.0.22.0", "74f493e72000-74f4940dc000 r-xp 00072000 103:04 30954756 /srv/llm/work/unsloth-runtime-compare-20260905/bin-unsloth/libggml-vulkan.so.0.22.0", "74f4940dc000-74f497219000 r--p 002dc000 103:04 30954756 /srv/llm/work/unsloth-runtime-compare-20260905/bin-unsloth/libggml-vulkan.so.0.22.0", "74f497219000-74f497220000 r--p 03419000 103:04 30954756 /srv/llm/work/unsloth-runtime-compare-20260905/bin-unsloth/libggml-vulkan.so.0.22.0", "74f497220000-74f497222000 rw-p 03420000 103:04 30954756 /srv/llm/work/unsloth-runtime-compare-20260905/bin-unsloth/libggml-vulkan.so.0.22.0", "74f49722f000-74f497242000 r--p 00000000 103:04 30954755 /srv/llm/work/unsloth-runtime-compare-20260905/bin-unsloth/libggml-cpu.so.0.22.0", "74f497242000-74f497350000 r-xp 00013000 103:04 30954755 /srv/llm/work/unsloth-runtime-compare-20260905/bin-unsloth/libggml-cpu.so.0.22.0", "74f497350000-74f497378000 r--p 00121000 103:04 30954755 /srv/llm/work/unsloth-runtime-compare-20260905/bin-unsloth/libggml-cpu.so.0.22.0", "74f497378000-74f49737a000 r--p 00148000 103:04 30954755 /srv/llm/work/unsloth-runtime-compare-20260905/bin-unsloth/libggml-cpu.so.0.22.0", "74f49737a000-74f49737b000 rw-p 0014a000 103:04 30954755 /srv/llm/work/unsloth-runtime-compare-20260905/bin-unsloth/libggml-cpu.so.0.22.0", "74f497400000-74f4974ad000 r--p 00000000 103:04 30954761 /srv/llm/work/unsloth-runtime-compare-20260905/bin-unsloth/libllama.so.0.3.0", "74f4974ad000-74f497733000 r-xp 000ad000 103:04 30954761 /srv/llm/work/unsloth-runtime-compare-20260905/bin-unsloth/libllama.so.0.3.0", "74f497733000-74f4977b2000 r--p 00333000 103:04 30954761 /srv/llm/work/unsloth-runtime-compare-20260905/bin-unsloth/libllama.so.0.3.0", "74f4977b2000-74f4977bf000 r--p 003b2000 103:04 30954761 /srv/llm/work/unsloth-runtime-compare-20260905/bin-unsloth/libllama.so.0.3.0", "74f4977bf000-74f4977c4000 rw-p 003bf000 103:04 30954761 /srv/llm/work/unsloth-runtime-compare-20260905/bin-unsloth/libllama.so.0.3.0", "74f497a00000-74f497aa9000 r--p 00000000 103:04 30954759 /srv/llm/work/unsloth-runtime-compare-20260905/bin-unsloth/libllama-common.so.0.3.0", "74f497aa9000-74f497d89000 r-xp 000a9000 103:04 30954759 /srv/llm/work/unsloth-runtime-compare-20260905/bin-unsloth/libllama-common.so.0.3.0", "74f497d89000-74f497e26000 r--p 00389000 103:04 30954759 /srv/llm/work/unsloth-runtime-compare-20260905/bin-unsloth/libllama-common.so.0.3.0", "74f497e26000-74f497e2e000 r--p 00426000 103:04 30954759 /srv/llm/work/unsloth-runtime-compare-20260905/bin-unsloth/libllama-common.so.0.3.0", "74f497e2e000-74f497e33000 rw-p 0042e000 103:04 30954759 /srv/llm/work/unsloth-runtime-compare-20260905/bin-unsloth/libllama-common.so.0.3.0", "74f497f2e000-74f497f43000 r--p 00000000 103:04 30954754 /srv/llm/work/unsloth-runtime-compare-20260905/bin-unsloth/libggml-base.so.0.22.0", "74f497f43000-74f497fdc000 r-xp 00015000 103:04 30954754 /srv/llm/work/unsloth-runtime-compare-20260905/bin-unsloth/libggml-base.so.0.22.0", "74f497fdc000-74f497ffc000 r--p 000ae000 103:04 30954754 /srv/llm/work/unsloth-runtime-compare-20260905/bin-unsloth/libggml-base.so.0.22.0", "74f497ffc000-74f497ffe000 r--p 000ce000 103:04 30954754 /srv/llm/work/unsloth-runtime-compare-20260905/bin-unsloth/libggml-base.so.0.22.0", "74f497ffe000-74f498000000 rw-p 000d0000 103:04 30954754 /srv/llm/work/unsloth-runtime-compare-20260905/bin-unsloth/libggml-base.so.0.22.0", "74f498000000-74f49807d000 r--p 00000000 103:04 30954760 /srv/llm/work/unsloth-runtime-compare-20260905/bin-unsloth/libllama-server-impl.so", "74f49807d000-74f498279000 r-xp 0007d000 103:04 30954760 /srv/llm/work/unsloth-runtime-compare-20260905/bin-unsloth/libllama-server-impl.so", "74f498279000-74f4985c8000 r--p 00279000 103:04 30954760 /srv/llm/work/unsloth-runtime-compare-20260905/bin-unsloth/libllama-server-impl.so", "74f4985c8000-74f4985cd000 r--p 005c8000 103:04 30954760 /srv/llm/work/unsloth-runtime-compare-20260905/bin-unsloth/libllama-server-impl.so", "74f4985cd000-74f4985d1000 rw-p 005cd000 103:04 30954760 /srv/llm/work/unsloth-runtime-compare-20260905/bin-unsloth/libllama-server-impl.so" ], "pid": 321348, "slots": [ { "id": 0, "n_ctx": 262144, "speculative": true, "is_processing": false } ], "props": { "default_generation_settings": { "params": { "seed": 4294967295, "temperature": 0.0, "dynatemp_range": 0.0, "dynatemp_exponent": 1.0, "top_k": 1, "top_p": 1.0, "min_p": 0.0, "top_n_sigma": -1.0, "xtc_probability": 0.0, "xtc_threshold": 0.10000000149011612, "typical_p": 1.0, "repeat_last_n": 64, "repeat_penalty": 1.0, "presence_penalty": 0.0, "frequency_penalty": 0.0, "dry_multiplier": 0.0, "dry_base": 1.75, "dry_allowed_length": 2, "dry_penalty_last_n": 64, "mirostat": 0, "mirostat_tau": 5.0, "mirostat_eta": 0.10000000149011612, "adaptive_target": -1.0, "adaptive_decay": 0.8999999761581421, "max_tokens": -1, "n_predict": -1, "n_keep": 0, "n_discard": 0, "ignore_eos": false, "stream": false, "n_probs": 0, "min_keep": 0, "chat_format": "Content-only", "reasoning_format": "none", "reasoning_in_content": false, "generation_prompt": "", "samplers": [ "penalties", "dry", "top_n_sigma", "top_k", "typ_p", "top_p", "min_p", "xtc", "temperature" ], "speculative.types": "none", "timings_per_token": false, "post_sampling_probs": false, "backend_sampling": false, "lora": [] }, "n_ctx": 262144 }, "total_slots": 1, "model_alias": "/srv/llm/models/unsloth-Qwen3.8-Flash-Next-38bb39ee9782/UD-IQ4_XS/Qwen3.8-Flash-Next-UD-IQ4_XS-00001-of-00003.gguf", "model_ftype": "IQ4_XS - 4.25 bpw", "model_path": "/srv/llm/models/unsloth-Qwen3.8-Flash-Next-38bb39ee9782/UD-IQ4_XS/Qwen3.8-Flash-Next-UD-IQ4_XS-00001-of-00003.gguf", "modalities": { "vision": false, "video": false, "audio": false }, "media_marker": "<__media_D2vituKGgsFU3mPTqfvk5n6vMLiKUUlT__>", "endpoint_slots": true, "endpoint_props": false, "endpoint_metrics": true, "ui": true, "ui_settings": {}, "chat_template": "{%- set image_count = namespace(value=0) %}\n{%- set video_count = namespace(value=0) %}\n{%- macro render_content(content, do_vision_count, is_system_content=false) %}\n {%- if content is string %}\n {{- content }}\n {%- elif content is iterable and content is not mapping %}\n {%- for item in content %}\n {%- if 'image' in item or 'image_url' in item or item.type == 'image' %}\n {%- if is_system_content %}\n {{- raise_exception('System message cannot contain images.') }}\n {%- endif %}\n {%- if do_vision_count %}\n {%- set image_count.value = image_count.value + 1 %}\n {%- endif %}\n {%- if add_vision_id %}\n {{- 'Picture ' ~ image_count.value ~ ': ' }}\n {%- endif %}\n {{- '<|vision_start|><|image_pad|><|vision_end|>' }}\n {%- elif 'video' in item or item.type == 'video' %}\n {%- if is_system_content %}\n {{- raise_exception('System message cannot contain videos.') }}\n {%- endif %}\n {%- if do_vision_count %}\n {%- set video_count.value = video_count.value + 1 %}\n {%- endif %}\n {%- if add_vision_id %}\n {{- 'Video ' ~ video_count.value ~ ': ' }}\n {%- endif %}\n {{- '<|vision_start|><|video_pad|><|vision_end|>' }}\n {%- elif 'text' in item %}\n {{- item.text }}\n {%- else %}\n {{- raise_exception('Unexpected item type in content.') }}\n {%- endif %}\n {%- endfor %}\n {%- elif content is none or content is undefined %}\n {{- '' }}\n {%- else %}\n {{- raise_exception('Unexpected content type.') }}\n {%- endif %}\n{%- endmacro %}\n{%- if not messages %}\n {{- raise_exception('No messages provided.') }}\n{%- endif %}\n{%- set sysns = namespace(count=0, text='') %}\n{%- for message in messages %}\n {%- if sysns.count == loop.index0 and (message.role == 'system' or message.role == 'developer') %}\n {%- set sys_content = render_content(message.content, false, true)|trim %}\n {%- if sys_content %}\n {%- set sysns.text = sysns.text + ('\\n' if sysns.text else '') + sys_content %}\n {%- endif %}\n {%- set sysns.count = sysns.count + 1 %}\n {%- endif %}\n{%- endfor %}\n{%- set num_sys = sysns.count %}\n{%- set merged_system = sysns.text %}\n{%- set reasoning_instructions = '' %}\n{%- if enable_thinking is undefined or enable_thinking is true %}\n {%- set resolved_reasoning_effort = reasoning_effort|default('xhigh') %}\n {%- if resolved_reasoning_effort == 'high' %}\n {%- set resolved_reasoning_effort = 'xhigh' %}\n {%- endif %}\n {%- if resolved_reasoning_effort not in ('xhigh', 'medium', 'low') %}\n {{- raise_exception('Unexpected reasoning effort ' ~ reasoning_effort ~ '. Supported types are xhigh (default), medium, and low.') }}\n {%- endif %}\n {%- if resolved_reasoning_effort == 'xhigh' %}\n {%- set reasoning_instructions = 'Reasoning effort is set to xhigh. Please think carefully through the task, validate key assumptions, consider plausible alternatives, and prioritize correctness, consistency, and clarity in the final answer.' %}\n {%- elif resolved_reasoning_effort == 'low' %}\n {%- set reasoning_instructions = 'Reasoning effort is set to low. Keep your thinking brief and focused, moving directly to the conclusion without unnecessary elaboration.' %}\n {%- endif %}\n{%- endif %}\n{%- if tools and tools is iterable and tools is not mapping %}\n {{- '<|im_start|>system\\n' }}\n {%- if reasoning_instructions %}\n {{- reasoning_instructions + '\\n\\n' }}\n {%- endif %}\n {{- \"# Tools\\n\\nYou have access to the following functions:\\n\\n\" }}\n {%- for tool in tools %}\n {{- \"\\n\" }}\n {{- tool | tojson }}\n {%- endfor %}\n {{- \"\\n\" }}\n {{- '\\n\\nIf you choose to call a function ONLY reply in the following format with NO suffix:\\n\\n\\n\\n\\nvalue_1\\n\\n\\nThis is the value for the second parameter\\nthat can span\\nmultiple lines\\n\\n\\n\\n\\n\\nReminder:\\n- Function calls MUST follow the specified format: an inner block must be nested within XML tags\\n- Required parameters MUST be specified\\n- You may provide optional reasoning for your function call in natural language BEFORE the function call, but NOT after\\n- If there is no function call available, answer the question like normal with your current knowledge and do not tell the user about function calls\\n' }}\n {%- if merged_system %}\n {{- '\\n\\n' + merged_system }}\n {%- endif %}\n {{- '<|im_end|>\\n' }}\n{%- else %}\n {%- if merged_system %}\n {{- '<|im_start|>system\\n' + (reasoning_instructions + '\\n\\n' if reasoning_instructions else '') + merged_system + '<|im_end|>\\n' }}\n {%- elif reasoning_instructions %}\n {{- '<|im_start|>system\\n' + reasoning_instructions + '<|im_end|>\\n' }}\n {%- endif %}\n{%- endif %}\n{%- set ns = namespace(multi_step_tool=true, last_query_index=messages|length - 1) %}\n{%- for message in messages[::-1] %}\n {%- set index = (messages|length - 1) - loop.index0 %}\n {%- if ns.multi_step_tool and message.role == \"user\" %}\n {%- set content = render_content(message.content, false)|trim %}\n {%- if not(content.startswith('') and content.endswith('')) %}\n {%- set ns.multi_step_tool = false %}\n {%- set ns.last_query_index = index %}\n {%- endif %}\n {%- endif %}\n{%- endfor %}\n{%- for message in messages %}\n {%- if loop.index0 >= num_sys %}\n {%- set content = render_content(message.content, true)|trim %}\n {%- if message.role == \"system\" or message.role == \"developer\" %}\n {{- raise_exception('System message must be at the beginning.') }}\n {%- elif message.role == \"user\" %}\n {{- '<|im_start|>' + message.role + '\\n' + content + '<|im_end|>' + '\\n' }}\n {%- elif message.role == \"assistant\" %}\n {%- set reasoning_content = '' %}\n {%- if message.reasoning_content is string %}\n {%- set reasoning_content = message.reasoning_content %}\n {%- endif %}\n {%- set reasoning_content = reasoning_content|trim %}\n {%- if preserve_thinking is undefined or preserve_thinking is true or loop.index0 > ns.last_query_index %}\n {{- '<|im_start|>' + message.role + '\\n\\n' + reasoning_content + '\\n\\n\\n' + content }}\n {%- else %}\n {{- '<|im_start|>' + message.role + '\\n' + content }}\n {%- endif %}\n {%- if message.tool_calls and message.tool_calls is iterable and message.tool_calls is not mapping %}\n {%- for tool_call in message.tool_calls %}\n {%- if tool_call.function is defined %}\n {%- set tool_call = tool_call.function %}\n {%- endif %}\n {%- if tool_call.name is not defined or tool_call.name is none %}\n {{- raise_exception('Tool call is missing a function name.') }}\n {%- endif %}\n {%- if loop.first %}\n {%- if content|trim %}\n {{- '\\n\\n\\n\\n' }}\n {%- else %}\n {{- '\\n\\n' }}\n {%- endif %}\n {%- else %}\n {{- '\\n\\n\\n' }}\n {%- endif %}\n {%- if tool_call.arguments is mapping %}\n {%- for args_name, args_value in tool_call.arguments|items %}\n {{- '\\n' }}\n {%- set args_value = args_value | string if args_value is string else args_value | tojson | safe %}\n {{- args_value }}\n {{- '\\n\\n' }}\n {%- endfor %}\n {%- elif tool_call.arguments is string %}\n {%- if tool_call.arguments|trim %}\n {{- raise_exception('Tool call arguments for function \"' + (tool_call.name | string) + '\" were passed as a JSON string. Parse them into an object before calling apply_chat_template.') }}\n {%- endif %}\n {%- elif tool_call.arguments is defined and tool_call.arguments is not none %}\n {{- raise_exception('Tool call arguments for function \"' + (tool_call.name | string) + '\" must be an object/mapping or a JSON string.') }}\n {%- endif %}\n {{- '\\n' }}\n {%- endfor %}\n {%- endif %}\n {{- '<|im_end|>\\n' }}\n {%- elif message.role == \"tool\" %}\n {%- if loop.previtem and loop.previtem.role != \"tool\" %}\n {{- '<|im_start|>user' }}\n {%- endif %}\n {{- '\\n\\n' }}\n {{- content }}\n {{- '\\n' }}\n {%- if not loop.last and loop.nextitem.role != \"tool\" %}\n {{- '<|im_end|>\\n' }}\n {%- elif loop.last %}\n {{- '<|im_end|>\\n' }}\n {%- endif %}\n {%- else %}\n {{- raise_exception('Unexpected message role.') }}\n {%- endif %}\n {%- endif %}\n{%- endfor %}\n{%- if add_generation_prompt %}\n {{- '<|im_start|>assistant\\n' }}\n {%- if enable_thinking is defined and enable_thinking is false %}\n {{- '\\n\\n\\n\\n' }}\n {%- else %}\n {{- '\\n' }}\n {%- endif %}\n{%- endif %}\n{#- Unsloth fixes - developer role, merged system messages, tool calling #}", "chat_template_caps": { "supports_object_arguments": true, "supports_parallel_tool_calls": true, "supports_preserve_reasoning": true, "supports_reasoning_effort": true, "supports_string_content": true, "supports_system_role": true, "supports_tool_calls": true, "supports_tools": true, "supports_typed_content": false }, "bos_token": "<|endoftext|>", "eos_token": "<|im_end|>", "build_info": "b1-d1a9235", "is_sleeping": false, "cors_proxy_enabled": false }, "runtime": { "commit": "d1a92352cbd417fd840b4e765c0b82f5fe3d1d89", "path": "/srv/llm/work/unsloth-runtime-compare-20260905/bin-unsloth", "build": "native Release Vulkan GCC15.2", "sha256": { "libggml.so.0.22.0": "98a1577666dbc408ea7e0ca53ecf131c9fb2e0fdc5aa722d700f3f6d32de364f", "libllama-bench-impl.so": "007fd5258bcbe3df3bcd3a920e5ef0d751256b706a03edfb0c325bafa2a3f549", "libggml-cpu.so.0.22.0": "e9b2cc40e27e87ebaab290b67b69f7e6148483a7310b8efc5a477b45df985fa8", "libllama-common.so.0.3.0": "f5e60bf4529920ed0e58099e5540da46857f2c6b4bbcd4112c0ab6f801b6d1ae", "libllama.so.0.3.0": "2b772ffdfe672a28faaa00f313bf2a407c0a26a09222c64b6437349bf2874c59", "llama-server": "f16f7acc3bcbf035166df093470797e3ffcc7bd05fc09570994bbff8ba949f69", "libllama-server-impl.so": "c8058db9aec86b1d7a9326881e10d95b9d529f4476d19bb69635c33f13b97bde", "logit-capture": "9fdff72b74e326b802228fdf2b35be3516d71b3bdf6238e4a408caa65a45b6fc", "llama-bench": "b28863f4bffd5dc210745e8d154824bd121fc6c1daa075c3a99de9c82233d228", "libggml-vulkan.so.0.22.0": "43a9fd1608f68f9a8358b1e64911c1ebd39583e07c2ff6677d6673ae9da7393a", "libggml-base.so.0.22.0": "b18eb78d5e212593aa3ad509892abba5fc6fab77cbb85241b5742443ff47f82e", "libmtmd.so.0.3.0": "1846d70cc91fe4af957fc4562ab2e18993a7e8cbd17af7fbd2f4d5fcd8d39cc3" } } } }, "_analysis": { "reported_aggregate": "Sum(generated_tokens-1)/sum(predicted_ms/1000), matching native per-task TG. Raw driver summaries initially summed all generated tokens; collect_he.py derives the final report from preserved timing records and validates this convention for all30 requests.", "source_of_truth": "results.json and humaneval-totals.csv use the corrected native convention; no inference was repeated.", "context_correction": { "date": "2026-09-05", "scope": "CIRU HumanEval0\u20139 rerun at released262144-token context; supersedes earlier16384-token diagnostic configuration. Laurent/Unsloth and BF16/sweep data unchanged.", "new_source": "research-release256k-ciru-HumanEval-*" } } }