{"data":{"capabilities":{"chat":null,"reasoning":null,"tools":null,"vision":null},"description":"Observed LocalMaxxing leaderboard run. Evidence for compatibility, not an executable launch contract.","engine":{"graph_mode":null,"name":"llama.cpp","version":"db44417+sycl-async-cpy+qwen-split+single-kernel-allreduce"},"facts":{"capabilities.chat":{"provenance":{"captured_at":"2026-08-30T09:10:02Z","sources":[{"captured_at":"2026-08-30T09:10:02Z","kind":"registry-enrichment","url":"https://github.com/0xSero/local-ai-registry"}]},"reason":"capability-not-verified","state":"unknown"},"capabilities.reasoning":{"provenance":{"captured_at":"2026-08-30T09:10:02Z","sources":[{"captured_at":"2026-08-30T09:10:02Z","kind":"registry-enrichment","url":"https://github.com/0xSero/local-ai-registry"}]},"reason":"capability-not-verified","state":"unknown"},"capabilities.tools":{"provenance":{"captured_at":"2026-08-30T09:10:02Z","sources":[{"captured_at":"2026-08-30T09:10:02Z","kind":"registry-enrichment","url":"https://github.com/0xSero/local-ai-registry"}]},"reason":"capability-not-verified","state":"unknown"},"capabilities.vision":{"provenance":{"captured_at":"2026-08-30T09:10:02Z","sources":[{"captured_at":"2026-08-30T09:10:02Z","kind":"registry-enrichment","url":"https://github.com/0xSero/local-ai-registry"}]},"reason":"capability-not-verified","state":"unknown"},"engine.graph_mode":{"provenance":{"captured_at":"2026-08-30T09:10:02Z","sources":[{"captured_at":"2026-08-30T09:10:02Z","kind":"registry-enrichment","url":"https://github.com/0xSero/local-ai-registry"}]},"reason":"runtime-detail-not-published","state":"unknown"},"serving.kv_cache_tokens":{"provenance":{"captured_at":"2026-08-30T09:10:02Z","sources":[{"captured_at":"2026-08-30T09:10:02Z","kind":"registry-enrichment","url":"https://github.com/0xSero/local-ai-registry"}]},"reason":"kv-cache-capacity-not-published","state":"unknown"},"serving.max_concurrency":{"provenance":{"captured_at":"2026-08-31T23:03:15Z","sources":[{"captured_at":"2026-08-31T23:03:15Z","kind":"source-semantics","url":"https://www.localmaxxing.com/en/runs/cmor2e5r00004jl04o99d26p8"}]},"reason":"server-capacity-not-evidenced","state":"unknown"}},"hardware_count":4,"hardware_id":"intel-arc-pro-b70-32gb","id":"qwen3-6-27b-dflash-q4-0-intel-arc-pro-b70-32gb-llama-cpp-tp4","launch":{"container":{"captured_at":"2026-08-30T09:10:02Z","compose_file":null,"digest":null,"image":null,"reason":"reference-only-launch","runtime":null,"source":[{"captured_at":"2026-08-30T09:10:02Z","kind":"recipe-launch","url":"https://www.localmaxxing.com/en/runs/cmor2e5r00004jl04o99d26p8"}],"state":"none"},"kind":"reference","run_id":"cmor2e5r00004jl04o99d26p8","source":"localmaxxing","url":"https://www.localmaxxing.com/en/runs/cmor2e5r00004jl04o99d26p8"},"metadata":{"localmaxxing":{"backend":"xpu","batch_size":1,"hardware_label":"Intel Arc Pro B70","notes":"Quality-preserving Qwen3.6 27B Q4_0 GGUF tensor-split run on four Intel Arc Pro B70 32GB GPUs. Same Q4_0 weights and f16 KV cache, greedy decode, no speculative decode, no power-limit changes. Uses experimental direct 2-4 GPU single-kernel F32 allreduce. This is intentionally submitted as a negative scaling datapoint: 4-way tensor split regresses versus the best 3-GPU result because the root-kernel fanout must peer-read and write four partial tensors across 128 small F32 allreduces per generated token. 4-GPU root sweep stayed around 30.76-31.31 tok/s at 128 tokens, so the issue appears to be communication fanout/synchronization rather than a single bad root choice. Best TP3 result was 41.737 tok/s; best TP2 result was 39.849 tok/s; single-B70 baseline was 24.723 tok/s.","observed_command":"ONEAPI_DEVICE_SELECTOR=level_zero:0,1,2,3 GGML_SYCL_DISABLE_GRAPH=0 GGML_SYCL_DISABLE_DNN=1 GGML_SYCL_DISABLE_OPT=0 GGML_SYCL_PRIORITIZE_DMMV=0 GGML_SYCL_ASYNC_PEER_COPY=0 GGML_SYCL_ASYNC_CPY_TENSOR=1 GGML_SYCL_COMM_ALLREDUCE=1 GGML_SYCL_COMM_SINGLE_KERNEL=1 /home/steve/src/llama.cpp-q4-b70/build-sycl-2026-bmg-g31/bin/llama-bench -m /home/steve/models/qwen3.6-27b-q4_0-gguf/Qwen3.6-27B-Q4_0.gguf --prio 0 -dev SYCL0/SYCL1/SYCL2/SYCL3 -ngl 99 -p 0 -n 512 -sm tensor -ts 1/1/1/1 -b 512 -ub 32 -ctk f16 -ctv f16 -t 8 --poll 50 -fa 1 -r 1 -o jsonl","run_id":"cmor2e5r00004jl04o99d26p8","tokenized":{"arguments":["/home/steve/src/llama.cpp-q4-b70/build-sycl-2026-bmg-g31/bin/llama-bench","-m","/home/steve/models/qwen3.6-27b-q4_0-gguf/Qwen3.6-27B-Q4_0.gguf","--prio","0","-dev","SYCL0/SYCL1/SYCL2/SYCL3","-ngl","99","-p","0","-n","512","-sm","tensor","-ts","1/1/1/1","-b","512","-ub","32","-ctk","f16","-ctv","f16","-t","8","--poll","50","-fa","1","-r","1","-o","jsonl"],"environment":{"GGML_SYCL_ASYNC_CPY_TENSOR":"1","GGML_SYCL_ASYNC_PEER_COPY":"0","GGML_SYCL_COMM_ALLREDUCE":"1","GGML_SYCL_COMM_SINGLE_KERNEL":"1","GGML_SYCL_DISABLE_DNN":"1","GGML_SYCL_DISABLE_GRAPH":"0","GGML_SYCL_DISABLE_OPT":"0","GGML_SYCL_PRIORITIZE_DMMV":"0","ONEAPI_DEVICE_SELECTOR":"level_zero:0,1,2,3"},"fidelity":"faithful"}}},"model_instance_id":"z-lab-qwen3-6-27b-dflash--q4-0","provenance":{"captured_at":"2026-08-30T09:10:02Z","sources":[{"captured_at":"2026-08-30T09:10:02Z","kind":"normalized-recipe","url":"https://www.localmaxxing.com/en/runs/cmor2e5r00004jl04o99d26p8"}]},"recipe_source":"localmaxxing","schema_version":"local-ai-registry/v1","serving":{"kv_cache_tokens":null,"max_concurrency":null,"max_context_tokens":512,"tensor_parallel":4},"speed_sweep_ids":["qwen3-6-27b-dflash-q4-0-intel-arc-pro-b70-32gb-llama-cpp-tp4-sweep"],"status":"candidate","huggingface":{"link_type":"repository","provenance":{"captured_at":"2026-08-30T09:10:02Z","sources":[{"captured_at":"2026-08-30T09:10:02Z","kind":"huggingface-api","url":"https://huggingface.co/api/models/z-lab/Qwen3.6-27B-DFlash"}]},"reason":"hf-api-confirmed-public","repository":"z-lab/Qwen3.6-27B-DFlash","status":"known","url":"https://huggingface.co/z-lab/Qwen3.6-27B-DFlash"},"registry":{"launchable":false,"speed_evidence":{"available":true,"count":1,"speed_sweep_ids":["qwen3-6-27b-dflash-q4-0-intel-arc-pro-b70-32gb-llama-cpp-tp4-sweep"],"detail_urls":["/api/v1/speed-sweep/qwen3-6-27b-dflash-q4-0-intel-arc-pro-b70-32gb-llama-cpp-tp4-sweep"]},"runtime":"reference"},"relationships":{"hardware":{"api":"/api/v1/hardware/intel-arc-pro-b70-32gb","href":"/hardware/intel-arc-pro-b70-32gb","id":"intel-arc-pro-b70-32gb","name":"Intel Arc Pro B70"},"model":{"api":"/api/v1/models/qwen3-6-27b-dflash","href":"/models/qwen3-6-27b-dflash","id":"qwen3-6-27b-dflash","name":"Qwen3.6-27B-DFlash"},"model_instance":{"api":"/api/v1/model-instances/z-lab-qwen3-6-27b-dflash--q4-0","href":"/model-instances/z-lab-qwen3-6-27b-dflash--q4-0","id":"z-lab-qwen3-6-27b-dflash--q4-0","name":"z-lab/Qwen3.6-27B-DFlash"},"speed_sweep":[{"api":"/api/v1/speed-sweep/qwen3-6-27b-dflash-q4-0-intel-arc-pro-b70-32gb-llama-cpp-tp4-sweep","href":"/speed-sweep/qwen3-6-27b-dflash-q4-0-intel-arc-pro-b70-32gb-llama-cpp-tp4-sweep","id":"qwen3-6-27b-dflash-q4-0-intel-arc-pro-b70-32gb-llama-cpp-tp4-sweep"}]}},"meta":{"source":"registry"}}