╭────────────────────────────────────────────────────────────────────────╮
│                                                                        │
│  R I G M A R K   //   AGENT WORKLOAD RECEIPT                           │
│  BENCHMARKS LOCAL AI HOW CODING AGENTS ACTUALLY USE IT                 │
│  ●  9/9 BASIC OUTPUT GATES PASSED                                      │
│                                                                        │
├─ MODEL ────────────────────────────────────────────────────────────────┤
│                                                                        │
│  GLM-5.3-Flash-EXL3                                                    │
│                                                                        │
├─ SINGLE STREAM ────────────────────────────────────────────────────────┤
│                                                                        │
│  WORKLOAD         tok/s       Range tok/s   Last (s)    Checks         │
│  PROSE             48.6         46.8–49.7       60.6     ✓ 3/3         │
│  CODE              71.8         66.1–75.8      175.8     ✓ 3/3         │
│  STRUCTURED*      102.4       101.9–102.5       10.5     ✓ 3/3         │
│                                                                        │
│  * Predictable JSON ceiling; not general agent performance.            │
│  Decode medians are estimates; rates include streamed reasoning.       │
│                                                                        │
├─ PREFILL ──────────────────────────────────────────────────────────────┤
│                                                                        │
│  DEPTH       Cold tok/s      Cold TTFT (s)    Replay TTFT (s)          │
│  8K               1,996               4.10               0.03          │
│  32K              2,027              16.17               0.06          │
│  64K              2,003              32.72               0.06          │
│  Medians; replay is an immediate repeat, not a proven hit.             │
│                                                                        │
├─ CAPPED CONCURRENT GENERATION ─────────────────────────────────────────┤
│                                                                        │
│  Aggregate tok/s: C1 61.0 | C2 84.5 | C4 114.7                         │
│  C4: 0/12 normal stops; 0/12 with visible output.                      │
│  Capped throughput includes reasoning; not completed agent tasks.      │
│                                                                        │
├─ APPLIANCE ────────────────────────────────────────────────────────────┤
│                                                                        │
│  Hardware: 2x NVIDIA DGX Spark (GB10, 128 GB unified memory each)      │
│  Topology: TP2 across 2 Sparks, RoCE over ConnectX-7 via switch        │
│  Checkpoint: AxonZeta/GLM-5.3-Flash-AZ-Focus-EXL3 @ 13ba3ef94d88       │
│  Quantisation: EXL3 4 bpw routed experts (mcg codebook), BF16 elsewhere│
│  KV cache: fp8 (e4m3 DSA latent cache and indexer keys, Mia v1.10      │
│    KV=fp8 default); bf16 prompt activations                            │
│  Engine: TensorFold v0.6.0 + Mia GLM-5.3-Flash kit v1.10 (96 patches,  │
│    commit 549fdc5)                                                     │
│                                                                        │
├─ SETTINGS ─────────────────────────────────────────────────────────────┤
│                                                                        │
│  SUITE: CUSTOM SETTINGS                                                │
│  Changed: --runs=3                                                     │
│  Changed: --decode-tokens=32768                                        │
│  REQUEST: {"chat_template_kwargs": {"reasoning_effort": "max"}}        │
│  Temperature 0.0 | top_p 1.0 | seed 20260905 | protocol 1.3.0          │
│  Decode: 3 runs, 32768-token cap | Prefill: 8K/32K/64K, 3 pairs        │
│  Concurrency: C1/C2/C4, 3 rounds, 256-token cap; workload code         │
│  Cache isolation: random-salt-per-pair                                 │
│  Comparison ID: glm53flash-tp2-dflash2-max                             │
│                                                                        │
├─ RECEIPT ──────────────────────────────────────────────────────────────┤
│                                                                        │
│  SOURCE     git:52c29c3fc1d7  •  clean                                 │
│  JSON sha256:e068abb87faa777b…                                         │
│  SHARE THE CARD • LINK THE JSON RECEIPT • #RIGMARK                     │
│  github.com/alexellis/rigmark                                          │
│                                                                        │
╰────────────────────────────────────────────────────────────────────────╯