{
  "artifacts": [
    {
      "file": "nemotron3_nano_30b_a3b_mixed_int4_ekv4096.litertlm",
      "sha256": "8b9c106a975a079080cf0170a2babba182861922836b64797eb98fd82d53e38c",
      "size_mb": 19670.354
    }
  ],
  "benchmarks": [],
  "conversion": {
    "command": "TODO (the card gives no conversion command); source nvidia/NVIDIA-Nemotron-3-Nano-30B-A3B-BF16, quantized (card L13-L14, L27)",
    "quantization": "experts int4 in blocks of 32 values; the other weight matrices int8; router fp32; fp32 activations declared; a 4,096-token context; 20.6 GB (card L38, L40)",
    "tool": "TODO (the card names no converter). The graph, as the card states it: each layer's experts are one `moe` custom op; relu² is written as its `gelu_tanh` (the activation was checked through the logits); each `moe` output has a second consumer (`moe * 0`) so ML Drift does not fuse the next op into it (card L40)",
    "tool_version": "TODO (the converter's version is not stated); runtime LiteRT-LM 0.19.1: the CPU through the PyPI package, the GPU through the v0.19.1 command-line tool built from source (7a03724c) (card L27, L76-L80)"
  },
  "cross_runtime": [],
  "delegation": null,
  "device": {
    "records": [
      {
        "device": "mac-studio-m4-max",
        "run": {
          "accelerator": "cpu",
          "context_length": null,
          "date": "2026-11-10",
          "decode_tokens_per_s": null,
          "delegated_ops": null,
          "env": {
            "device": "Mac Studio (M4 Max, Mac16,9)",
            "machine_label": "mac-studio-m4-max",
            "os_build": "macOS 27.0 (26A428)",
            "runtime": "litert-lm",
            "runtime_version": "0.19.1",
            "soc": "Apple M4 Max",
            "vendor_sdk": null
          },
          "error": null,
          "evidence": [
            "bytes: these runs read as the bundle step wrote it (done 2026-11-10 03:02:39 JST), hashed by the pre-upload after them (done 2026-11-10 07:51:02 JST, sha256 8b9c106a975a079080cf0170a2babba182861922836b64797eb98fd82d53e38c, 20625860800 B) = the Hub's LFS oid and size at main 5792549c (anonymous) and at commit ccb09e82 (the commit's readback)",
            "runtime: the PyPI litert-lm 0.19.1 macOS arm64 wheel's Python API (tools/line3_rt.py; each record names litert-lm 0.19.1), CPU (XNNPACK) 8 threads, one Engine per process and one Conversation with the file's own template, SamplerConfig(top_k=1), ThinkingConfig(enable_thinking=False), max_output_tokens 2,048; the 3 turns in one conversation (the runtime keeps the history); the file's CPU runs used one shared engine cache directory (tools/line3_chain.sh)",
            "output_match = line 3 (a) and (b) as written before any output: (a) the file's verdict equals the reference's on 12 of 12 (at least 11 required; both answer 'violet' to the rhyme question, so 11 are right), (b) every question the reference ends with a stop token ends with one here (12 / 12). Reference = the float32 implementation of the model on the bf16 weights, thinking off, greedy, its replies built by draft-and-verify (every appended token its argmax), held to NVIDIA's modeling_nemotron_h.py by lines 0 and R",
            "the 3-turn conversation: 'Kenji', then '401', all three ending with a stop token; the reference's own turns: 'Kenji', then '401'. Line 3 (c) as written asks for 391 in turn 3 (not met); line 3' (written after these outputs were seen: a turn the reference gets wrong is not required of the file) is met; in metrics, not part of output_match",
            "(cpu_off: 2026-11-10 (thinking off 03:03-03:08 and its 3 turns at 03:4x, thinking on 03:41-04:04), after the bundle step (03:02:39) and before the pre-upload (done 07:51:02)): reply tokens up to 89; = MEASUREMENTS.md's Accuracy row for 'mixed_int4 ekv4096' and the Hub card's Accuracy table at main 5792549c"
          ],
          "failure_class": null,
          "full_delegation": null,
          "latency_p50_ms": null,
          "loads": true,
          "max_abs_diff": null,
          "max_rel_diff": null,
          "metrics": {
            "answers_right": 11.0,
            "conversation_c_as_written": 0.0,
            "conversation_c_line3_prime": 1.0,
            "conversation_turn2_name_right": 1.0,
            "conversation_turn3_answer": 401.0,
            "conversation_turns": 3.0,
            "conversation_turns_ended_with_stop_token": 3.0,
            "cpu_threads": 8.0,
            "ended_with_stop_token": 12.0,
            "max_new_tokens": 2048.0,
            "questions": 12.0,
            "reference_ended_with_stop_token": 12.0,
            "reference_turn3_answer": 401.0,
            "reply_tokens_max": 89.0,
            "verdicts_equal_reference": 12.0
          },
          "output_match": true,
          "peak_mem_mb": null,
          "prefill_tokens_per_s": null,
          "provenance": "measured",
          "runs": true,
          "signature": "the 12 check questions of line 3 (the 8 short questions of hf-to-litertlm's scripts/verify_quality.py, a price word problem, a syllogism, an average speed, a question in Japanese) and its 3-turn conversation, thinking off (enable_thinking false), greedy (top-k 1), at most 2,048 new tokens, one process per question; verdicts against the source model's replies (float32 on the bf16 weights, thinking off)",
          "total_ops": null,
          "ttft_ms": null
        },
        "source": "data/device_runs/0.19.1/2026-11-10/nemotron-3-nano-30b-a3b-mixed-int4-ekv4096__mac-studio-m4-max.json"
      },
      {
        "device": "mac-studio-m4-max",
        "run": {
          "accelerator": "cpu",
          "context_length": null,
          "date": "2026-11-10",
          "decode_tokens_per_s": null,
          "delegated_ops": null,
          "env": {
            "device": "Mac Studio (M4 Max, Mac16,9)",
            "machine_label": "mac-studio-m4-max",
            "os_build": "macOS 27.0 (26A428)",
            "runtime": "litert-lm",
            "runtime_version": "0.19.1",
            "soc": "Apple M4 Max",
            "vendor_sdk": null
          },
          "error": null,
          "evidence": [
            "bytes: these runs read as the bundle step wrote it (done 2026-11-10 03:02:39 JST), hashed by the pre-upload after them (done 2026-11-10 07:51:02 JST, sha256 8b9c106a975a079080cf0170a2babba182861922836b64797eb98fd82d53e38c, 20625860800 B) = the Hub's LFS oid and size at main 5792549c (anonymous) and at commit ccb09e82 (the commit's readback)",
            "runtime: the PyPI litert-lm 0.19.1 macOS arm64 wheel's Python API (tools/line3_rt.py; each record names litert-lm 0.19.1), CPU (XNNPACK) 8 threads, one Engine per process and one Conversation with the file's own template, SamplerConfig(top_k=1), ThinkingConfig(enable_thinking=True), max_output_tokens 2,048; the 3 turns in one conversation (the runtime keeps the history); the file's CPU runs used one shared engine cache directory (tools/line3_chain.sh)",
            "output_match = line 3's rule for thinking on, written before any output (the fallback taken on 2026-11-10 because the reference generator ran at 77.8 s per step, so no reference ran with thinking on): the line's regex right on at least 11 of the 12 and every reply ending with a stop token within 2,048 new tokens: 12 / 12 right, 12 / 12 stopped",
            "the 3-turn conversation ('My name is Kenji.' / 'What is my name? Answer with the name only.' / 'What is 17 times 23? Answer with the number only.'): 'Kenji', then '391', all three ending with a stop token = line 3 (c) as written; in metrics, not part of output_match",
            "(cpu_on: 2026-11-10 (thinking off 03:03-03:08 and its 3 turns at 03:4x, thinking on 03:41-04:04), after the bundle step (03:02:39) and before the pre-upload (done 07:51:02)): reply tokens (the thought included) up to 201; = MEASUREMENTS.md's Accuracy row for 'mixed_int4 ekv4096' and the Hub card's Accuracy table at main 5792549c"
          ],
          "failure_class": null,
          "full_delegation": null,
          "latency_p50_ms": null,
          "loads": true,
          "max_abs_diff": null,
          "max_rel_diff": null,
          "metrics": {
            "answers_right": 12.0,
            "conversation_c_as_written": 1.0,
            "conversation_turn2_name_right": 1.0,
            "conversation_turn3_answer": 391.0,
            "conversation_turns": 3.0,
            "conversation_turns_ended_with_stop_token": 3.0,
            "cpu_threads": 8.0,
            "ended_with_stop_token": 12.0,
            "max_new_tokens": 2048.0,
            "questions": 12.0,
            "reply_tokens_max": 201.0
          },
          "output_match": true,
          "peak_mem_mb": null,
          "prefill_tokens_per_s": null,
          "provenance": "measured",
          "runs": true,
          "signature": "the 12 check questions of line 3 (the 8 short questions of hf-to-litertlm's scripts/verify_quality.py, a price word problem, a syllogism, an average speed, a question in Japanese) and its 3-turn conversation, thinking on (the template's default), greedy (top-k 1), at most 2,048 new tokens, one process per question; verdicts by the line's regex on the text after the thought",
          "total_ops": null,
          "ttft_ms": null
        },
        "source": "data/device_runs/0.19.1/2026-11-10/nemotron-3-nano-30b-a3b-mixed-int4-ekv4096__mac-studio-m4-max.json"
      },
      {
        "device": "mac-studio-m4-max",
        "run": {
          "accelerator": "cpu",
          "context_length": 4096,
          "date": "2026-11-12",
          "decode_tokens_per_s": 7.98,
          "delegated_ops": null,
          "env": {
            "device": "Mac Studio (M4 Max, Mac16,9)",
            "machine_label": "mac-studio-m4-max",
            "os_build": "macOS 27.0 (26A428)",
            "runtime": "litert-lm",
            "runtime_version": "0.19.1",
            "soc": "Apple M4 Max",
            "vendor_sdk": null
          },
          "error": null,
          "evidence": [
            "bytes: windows W22 and W23 read, downloaded again from the Hub (logged back at 2026-11-11 18:22:28 JST) and checked before use (tools/unpark_bundles.sh compares size and sha256 before it logs the file back): sha256 8b9c106a975a079080cf0170a2babba182861922836b64797eb98fd82d53e38c, 20625860800 B, the value the work had checked against the Hub's before it removed the first copy (2026-11-11 15:03) = the Hub's LFS oid and size at main 5792549c (anonymous) and at commit ccb09e82 (the commit's readback); the copy was removed after the windows (2026-11-12 03:31:28, deleted_paths.txt)",
            "runtime: litert_lm_advanced_main built from the LiteRT-LM v0.19.1 tag (7a03724c; sha256 6f0eba84..., with the tag's prebuilt macos_arm64 libwebgpu_dawn.dylib and libGemmaModelConstraintProvider.dylib), one process at a time: `--backend=cpu --benchmark --benchmark_prefill_tokens=1024 --benchmark_decode_tokens=256 --num_iterations=2 --max_num_tokens=2048 --disable_cache=true --report_peak_memory_footprint --use_session=true --async=false --num_cpu_threads=8`; CPU = XNNPACK, number_of_threads 8 (each log prints it); the --use_session=true --async=false path gives one 1,024-token prefill per iteration (the CLI's conversation path prefilled the 1,024 tokens twice per iteration on this bundle, MEASUREMENTS.md); the logs print no partition or node count",
            "window W23 of 2026-11-12: this arm's processes ran 02:23:12-02:41:32 JST between the other arms' (one process at a time); every process's meta says 'overlap clean'; both exclusion lists of the window are empty; the cell (the benchmark rule): per process iteration 2 of 2 of one engine (iteration 1 = the warm-up), the cell = the median of the 3 processes, prefill and decode each; a process counts when it exits 0, its log has no Validation error, Shape mismatch or not-fully-delegated line, every iteration is one 1,024-token prefill turn and one 256-token decode turn, and the window's 1 s record shows no burst of other work at 300 % or more in its counted spans (every span ok); ttft_ms = the median of the processes' mean TTFT over the same iterations; peak_mem_mb = the largest of the processes' last 'Peak private footprint' print (--report_peak_memory_footprint; the block of the median of the iterations), the reading of the card and MEASUREMENTS.md, whose GB = MB x 2^20 / 10^9 (tools/speed_parse.py reads the print as MiB); each iteration's 'Peak system ram usage' print is in metrics",
            "context_length = the file's KV cache, 4,096 tokens (the card's Context column; the writer's --cache=4096); --max_num_tokens=2048 caps the run's tokens",
            "(started 2026-11-12 02:23:12 JST, CPU 78 % idle): 53.19 / 9.02 tok/s, TTFT 19.36 s; 113.15 / 7.66 tok/s, TTFT 9.18 s; Init Total 4945.88 ms; the last Peak private footprint print 6620.77 MB",
            "(started 2026-11-12 02:34:51 JST, CPU 79 % idle): 55.96 / 7.8 tok/s, TTFT 18.43 s; 115.24 / 8.09 tok/s, TTFT 9.01 s; Init Total 4856.35 ms; the last Peak private footprint print 6619.62 MB",
            "(started 2026-11-12 02:39:46 JST, CPU 77 % idle): 71.54 / 7.96 tok/s, TTFT 14.44 s; 114.42 / 7.98 tok/s, TTFT 9.08 s; Init Total 5002.68 ms; the last Peak private footprint print 6618.01 MB",
            "= MEASUREMENTS.md section 2026-11-12 ('| mixed_int4 ekv4096 | CPU (XNNPACK, 8 threads) | 114.42 tokens/s | 7.98 tokens/s | 3 | 6.94 GB |' and its per-process lists) and the Hub card's Performance row ('| mixed_int4 ekv4096 | CPU (8 threads) | 114 tokens/s | 7.98 tokens/s | 6.9 GB |') at main 5792549c"
          ],
          "failure_class": null,
          "full_delegation": null,
          "latency_p50_ms": null,
          "loads": true,
          "max_abs_diff": null,
          "max_rel_diff": null,
          "metrics": {
            "cpu_threads": 8.0,
            "decode_tokens": 256.0,
            "iterations": 2.0,
            "max_num_tokens": 2048.0,
            "peak_private_footprint_gb_max": 6.94,
            "prefill_tokens": 1024.0,
            "proc1_cpu_idle_at_start_pct": 78.0,
            "proc1_init_total_ms": 4945.88,
            "proc1_iter1_decode_tok_s": 9.02,
            "proc1_iter1_prefill_tok_s": 53.19,
            "proc1_iter1_ttft_s": 19.36,
            "proc1_iter2_decode_tok_s": 7.66,
            "proc1_iter2_prefill_tok_s": 113.15,
            "proc1_iter2_ttft_s": 9.18,
            "proc1_peak_private_footprint_mb": 6620.77,
            "proc1_peak_system_ram_mb_max": 6620.77,
            "proc2_cpu_idle_at_start_pct": 79.0,
            "proc2_init_total_ms": 4856.35,
            "proc2_iter1_decode_tok_s": 7.8,
            "proc2_iter1_prefill_tok_s": 55.96,
            "proc2_iter1_ttft_s": 18.43,
            "proc2_iter2_decode_tok_s": 8.09,
            "proc2_iter2_prefill_tok_s": 115.24,
            "proc2_iter2_ttft_s": 9.01,
            "proc2_peak_private_footprint_mb": 6619.62,
            "proc2_peak_system_ram_mb_max": 6619.64,
            "proc3_cpu_idle_at_start_pct": 77.0,
            "proc3_init_total_ms": 5002.68,
            "proc3_iter1_decode_tok_s": 7.96,
            "proc3_iter1_prefill_tok_s": 71.54,
            "proc3_iter1_ttft_s": 14.44,
            "proc3_iter2_decode_tok_s": 7.98,
            "proc3_iter2_prefill_tok_s": 114.42,
            "proc3_iter2_ttft_s": 9.08,
            "proc3_peak_private_footprint_mb": 6618.01,
            "proc3_peak_system_ram_mb_max": 6618.08,
            "processes": 3.0
          },
          "output_match": null,
          "peak_mem_mb": 6620.77,
          "prefill_tokens_per_s": 114.42,
          "provenance": "measured",
          "runs": true,
          "signature": null,
          "total_ops": null,
          "ttft_ms": 9080.0
        },
        "source": "data/device_runs/0.19.1/2026-11-12/nemotron-3-nano-30b-a3b-mixed-int4-ekv4096__mac-studio-m4-max.json"
      },
      {
        "device": "mac-studio-m4-max",
        "run": {
          "accelerator": "gpu",
          "context_length": null,
          "date": "2026-11-10",
          "decode_tokens_per_s": null,
          "delegated_ops": null,
          "env": {
            "device": "Mac Studio (M4 Max, Mac16,9)",
            "machine_label": "mac-studio-m4-max",
            "os_build": "macOS 27.0 (26A428)",
            "runtime": "litert-lm",
            "runtime_version": "0.19.1",
            "soc": "Apple M4 Max",
            "vendor_sdk": null
          },
          "error": null,
          "evidence": [
            "bytes: these runs read as the bundle step wrote it (done 2026-11-10 03:02:39 JST), hashed by the pre-upload after them (done 2026-11-10 07:51:02 JST, sha256 8b9c106a975a079080cf0170a2babba182861922836b64797eb98fd82d53e38c, 20625860800 B) = the Hub's LFS oid and size at main 5792549c (anonymous) and at commit ccb09e82 (the commit's readback)",
            "runtime: litert_lm_advanced_main built from the LiteRT-LM v0.19.1 tag (7a03724c; sha256 6f0eba84...), WebGPU on Metal, one process per question: `--backend=gpu --max_output_tokens=2048 --min_log_severity=2 --sampler_top_k=1 --sampler_handles_input=false --disable_cache=true --use_session=true --async=false --input_prompt_file=<the vendor template's render with enable_thinking false>` (the CLI has no thinking switch; the render equals the file's template for these prompts, line T); the 3 turns = 3 processes, each render carrying the file's own earlier answers",
            "output_match = line 3 (a) and (b) as written before any output: (a) the file's verdict equals the reference's on 12 of 12 (at least 11 required; both answer 'violet' to the rhyme question, so 11 are right), (b) every question the reference ends with a stop token ends with one here (12 / 12). Reference = the float32 implementation of the model on the bf16 weights, thinking off, greedy, its replies built by draft-and-verify (every appended token its argmax), held to NVIDIA's modeling_nemotron_h.py by lines 0 and R",
            "the 3-turn conversation: 'Kenji', then '401', all three ending with a stop token; the reference's own turns: 'Kenji', then '401'. Line 3 (c) as written asks for 391 in turn 3 (not met); line 3' (written after these outputs were seen: a turn the reference gets wrong is not required of the file) is met; in metrics, not part of output_match",
            "(gpu_off: 2026-11-10 04:08:10-04:17:26 JST (each run's end time in the jsonl)): reply tokens up to 286; = MEASUREMENTS.md's Accuracy row for 'mixed_int4 ekv4096' and the Hub card's Accuracy table at main 5792549c"
          ],
          "failure_class": null,
          "full_delegation": null,
          "latency_p50_ms": null,
          "loads": true,
          "max_abs_diff": null,
          "max_rel_diff": null,
          "metrics": {
            "answers_right": 11.0,
            "conversation_c_as_written": 0.0,
            "conversation_c_line3_prime": 1.0,
            "conversation_turn2_name_right": 1.0,
            "conversation_turn3_answer": 401.0,
            "conversation_turns": 3.0,
            "conversation_turns_ended_with_stop_token": 3.0,
            "ended_with_stop_token": 12.0,
            "max_new_tokens": 2048.0,
            "questions": 12.0,
            "reference_ended_with_stop_token": 12.0,
            "reference_turn3_answer": 401.0,
            "reply_tokens_max": 286.0,
            "verdicts_equal_reference": 12.0
          },
          "output_match": true,
          "peak_mem_mb": null,
          "prefill_tokens_per_s": null,
          "provenance": "measured",
          "runs": true,
          "signature": "the 12 check questions of line 3 (the 8 short questions of hf-to-litertlm's scripts/verify_quality.py, a price word problem, a syllogism, an average speed, a question in Japanese) and its 3-turn conversation, thinking off (enable_thinking false), greedy (top-k 1), at most 2,048 new tokens, one process per question; verdicts against the source model's replies (float32 on the bf16 weights, thinking off)",
          "total_ops": null,
          "ttft_ms": null
        },
        "source": "data/device_runs/0.19.1/2026-11-10/nemotron-3-nano-30b-a3b-mixed-int4-ekv4096__mac-studio-m4-max.json"
      },
      {
        "device": "mac-studio-m4-max",
        "run": {
          "accelerator": "gpu",
          "context_length": null,
          "date": "2026-11-10",
          "decode_tokens_per_s": null,
          "delegated_ops": null,
          "env": {
            "device": "Mac Studio (M4 Max, Mac16,9)",
            "machine_label": "mac-studio-m4-max",
            "os_build": "macOS 27.0 (26A428)",
            "runtime": "litert-lm",
            "runtime_version": "0.19.1",
            "soc": "Apple M4 Max",
            "vendor_sdk": null
          },
          "error": null,
          "evidence": [
            "bytes: these runs read as the bundle step wrote it (done 2026-11-10 03:02:39 JST), hashed by the pre-upload after them (done 2026-11-10 07:51:02 JST, sha256 8b9c106a975a079080cf0170a2babba182861922836b64797eb98fd82d53e38c, 20625860800 B) = the Hub's LFS oid and size at main 5792549c (anonymous) and at commit ccb09e82 (the commit's readback)",
            "runtime: litert_lm_advanced_main built from the LiteRT-LM v0.19.1 tag (7a03724c; sha256 6f0eba84...), WebGPU on Metal, one process per question: `--backend=gpu --max_output_tokens=2048 --min_log_severity=2 --sampler_top_k=1 --sampler_handles_input=false --disable_cache=true --input_prompt=<question>` (the file's template, thinking on by default); the 3 turns through `--multi_turns` in one process",
            "output_match = line 3's rule for thinking on, written before any output (the fallback taken on 2026-11-10 because the reference generator ran at 77.8 s per step, so no reference ran with thinking on): the line's regex right on at least 11 of the 12 and every reply ending with a stop token within 2,048 new tokens: 12 / 12 right, 12 / 12 stopped",
            "the 3-turn conversation ('My name is Kenji.' / 'What is my name? Answer with the name only.' / 'What is 17 times 23? Answer with the number only.'): 'Kenji', then '391', all three ending with a stop token = line 3 (c) as written; in metrics, not part of output_match",
            "(gpu_on: 2026-11-10 03:49:44-04:03:43 JST (each run's end time in the jsonl)): reply tokens (the thought included) up to 244; = MEASUREMENTS.md's Accuracy row for 'mixed_int4 ekv4096' and the Hub card's Accuracy table at main 5792549c"
          ],
          "failure_class": null,
          "full_delegation": null,
          "latency_p50_ms": null,
          "loads": true,
          "max_abs_diff": null,
          "max_rel_diff": null,
          "metrics": {
            "answers_right": 12.0,
            "conversation_c_as_written": 1.0,
            "conversation_turn2_name_right": 1.0,
            "conversation_turn3_answer": 391.0,
            "conversation_turns": 3.0,
            "conversation_turns_ended_with_stop_token": 3.0,
            "ended_with_stop_token": 12.0,
            "max_new_tokens": 2048.0,
            "questions": 12.0,
            "reply_tokens_max": 244.0
          },
          "output_match": true,
          "peak_mem_mb": null,
          "prefill_tokens_per_s": null,
          "provenance": "measured",
          "runs": true,
          "signature": "the 12 check questions of line 3 (the 8 short questions of hf-to-litertlm's scripts/verify_quality.py, a price word problem, a syllogism, an average speed, a question in Japanese) and its 3-turn conversation, thinking on (the template's default), greedy (top-k 1), at most 2,048 new tokens, one process per question; verdicts by the line's regex on the text after the thought",
          "total_ops": null,
          "ttft_ms": null
        },
        "source": "data/device_runs/0.19.1/2026-11-10/nemotron-3-nano-30b-a3b-mixed-int4-ekv4096__mac-studio-m4-max.json"
      },
      {
        "device": "mac-studio-m4-max",
        "run": {
          "accelerator": "gpu",
          "context_length": 4096,
          "date": "2026-11-12",
          "decode_tokens_per_s": 67.63,
          "delegated_ops": null,
          "env": {
            "device": "Mac Studio (M4 Max, Mac16,9)",
            "machine_label": "mac-studio-m4-max",
            "os_build": "macOS 27.0 (26A428)",
            "runtime": "litert-lm",
            "runtime_version": "0.19.1",
            "soc": "Apple M4 Max",
            "vendor_sdk": null
          },
          "error": null,
          "evidence": [
            "bytes: windows W22 and W23 read, downloaded again from the Hub (logged back at 2026-11-11 18:22:28 JST) and checked before use (tools/unpark_bundles.sh compares size and sha256 before it logs the file back): sha256 8b9c106a975a079080cf0170a2babba182861922836b64797eb98fd82d53e38c, 20625860800 B, the value the work had checked against the Hub's before it removed the first copy (2026-11-11 15:03) = the Hub's LFS oid and size at main 5792549c (anonymous) and at commit ccb09e82 (the commit's readback); the copy was removed after the windows (2026-11-12 03:31:28, deleted_paths.txt)",
            "runtime: litert_lm_advanced_main built from the LiteRT-LM v0.19.1 tag (7a03724c; sha256 6f0eba84..., with the tag's prebuilt macos_arm64 libwebgpu_dawn.dylib and libGemmaModelConstraintProvider.dylib), one process at a time: `--backend=gpu --benchmark --benchmark_prefill_tokens=1024 --benchmark_decode_tokens=256 --num_iterations=4 --max_num_tokens=2048 --disable_cache=true --report_peak_memory_footprint --sampler_handles_input=false --use_session=true --async=false`; GPU = WebGPU on Metal (each log selects 'Apple M4 Max, arch=metal-3, vendor=apple, backend=Metal'); the --use_session=true --async=false path gives one 1,024-token prefill per iteration (the CLI's conversation path prefilled the 1,024 tokens twice per iteration on this bundle, MEASUREMENTS.md); the logs print no partition or node count",
            "window W22 of 2026-11-12: this arm's processes ran 01:07:09-01:27:50 JST between the other arms' (one process at a time); every process's meta says 'overlap clean'; both exclusion lists of the window are empty; the cell (the benchmark rule): per process the mean of iterations 2-4 of one engine (iteration 1 = the warm-up), the cell = the median of the 5 processes, prefill and decode each; a process counts when it exits 0, its log has no Validation error, Shape mismatch or not-fully-delegated line, every iteration is one 1,024-token prefill turn and one 256-token decode turn, and the window's 1 s record shows no burst of other work at 300 % or more in its counted spans (every span ok); ttft_ms = the median of the processes' mean TTFT over the same iterations; peak_mem_mb = the largest of the processes' last 'Peak private footprint' print (--report_peak_memory_footprint; the block of the median of the iterations), the reading of the card and MEASUREMENTS.md, whose GB = MB x 2^20 / 10^9 (tools/speed_parse.py reads the print as MiB); each iteration's 'Peak system ram usage' print is in metrics",
            "context_length = the file's KV cache, 4,096 tokens (the card's Context column; the writer's --cache=4096); --max_num_tokens=2048 caps the run's tokens",
            "(started 2026-11-12 01:07:09 JST, CPU 65 % idle): 628.78 / 66.15 tok/s, TTFT 1.64 s; 808.56 / 67.29 tok/s, TTFT 1.28 s; 809.24 / 67.35 tok/s, TTFT 1.28 s; 798.02 / 67.13 tok/s, TTFT 1.3 s; Init Total 40589.55 ms; the last Peak private footprint print 22109.51 MB",
            "(started 2026-11-12 01:15:41 JST, CPU 81 % idle): 780.56 / 67.08 tok/s, TTFT 1.33 s; 810.26 / 67.66 tok/s, TTFT 1.28 s; 805.28 / 66.09 tok/s, TTFT 1.29 s; 811.31 / 67.1 tok/s, TTFT 1.28 s; Init Total 40211.74 ms; the last Peak private footprint print 22108.99 MB",
            "(started 2026-11-12 01:19:20 JST, CPU 80 % idle): 780.12 / 67.34 tok/s, TTFT 1.33 s; 810.45 / 67.82 tok/s, TTFT 1.28 s; 811.03 / 67.94 tok/s, TTFT 1.28 s; 809.32 / 67.8 tok/s, TTFT 1.28 s; Init Total 39847.59 ms; the last Peak private footprint print 22112.95 MB",
            "(started 2026-11-12 01:22:59 JST, CPU 78 % idle): 772.55 / 67.43 tok/s, TTFT 1.34 s; 809.15 / 67.7 tok/s, TTFT 1.28 s; 811.48 / 67.98 tok/s, TTFT 1.28 s; 807.5 / 67.21 tok/s, TTFT 1.28 s; Init Total 38955.56 ms; the last Peak private footprint print 22115.23 MB",
            "(started 2026-11-12 01:26:39 JST, CPU 87 % idle): 776.25 / 67.97 tok/s, TTFT 1.33 s; 808.17 / 67.97 tok/s, TTFT 1.28 s; 809.63 / 68.35 tok/s, TTFT 1.28 s; 806.97 / 67.87 tok/s, TTFT 1.28 s; Init Total 38812.02 ms; the last Peak private footprint print 22102.75 MB",
            "= MEASUREMENTS.md section 2026-11-12 ('| mixed_int4 ekv4096 | GPU (WebGPU) | 808.95 tokens/s | 67.63 tokens/s | 5 | 23.19 GB |' and its per-process lists) and the Hub card's Performance row ('| mixed_int4 ekv4096 | GPU (WebGPU) | 809 tokens/s | 67.6 tokens/s | 23.2 GB |') at main 5792549c"
          ],
          "failure_class": null,
          "full_delegation": null,
          "latency_p50_ms": null,
          "loads": true,
          "max_abs_diff": null,
          "max_rel_diff": null,
          "metrics": {
            "decode_tokens": 256.0,
            "iterations": 4.0,
            "max_num_tokens": 2048.0,
            "peak_private_footprint_gb_max": 23.19,
            "prefill_tokens": 1024.0,
            "proc1_cpu_idle_at_start_pct": 65.0,
            "proc1_init_total_ms": 40589.55,
            "proc1_iter1_decode_tok_s": 66.15,
            "proc1_iter1_prefill_tok_s": 628.78,
            "proc1_iter1_ttft_s": 1.64,
            "proc1_iter2_decode_tok_s": 67.29,
            "proc1_iter2_prefill_tok_s": 808.56,
            "proc1_iter2_ttft_s": 1.28,
            "proc1_iter3_decode_tok_s": 67.35,
            "proc1_iter3_prefill_tok_s": 809.24,
            "proc1_iter3_ttft_s": 1.28,
            "proc1_iter4_decode_tok_s": 67.13,
            "proc1_iter4_prefill_tok_s": 798.02,
            "proc1_iter4_ttft_s": 1.3,
            "proc1_peak_private_footprint_mb": 22109.51,
            "proc1_peak_system_ram_mb_max": 22111.8,
            "proc2_cpu_idle_at_start_pct": 81.0,
            "proc2_init_total_ms": 40211.74,
            "proc2_iter1_decode_tok_s": 67.08,
            "proc2_iter1_prefill_tok_s": 780.56,
            "proc2_iter1_ttft_s": 1.33,
            "proc2_iter2_decode_tok_s": 67.66,
            "proc2_iter2_prefill_tok_s": 810.26,
            "proc2_iter2_ttft_s": 1.28,
            "proc2_iter3_decode_tok_s": 66.09,
            "proc2_iter3_prefill_tok_s": 805.28,
            "proc2_iter3_ttft_s": 1.29,
            "proc2_iter4_decode_tok_s": 67.1,
            "proc2_iter4_prefill_tok_s": 811.31,
            "proc2_iter4_ttft_s": 1.28,
            "proc2_peak_private_footprint_mb": 22108.99,
            "proc2_peak_system_ram_mb_max": 22111.5,
            "proc3_cpu_idle_at_start_pct": 80.0,
            "proc3_init_total_ms": 39847.59,
            "proc3_iter1_decode_tok_s": 67.34,
            "proc3_iter1_prefill_tok_s": 780.12,
            "proc3_iter1_ttft_s": 1.33,
            "proc3_iter2_decode_tok_s": 67.82,
            "proc3_iter2_prefill_tok_s": 810.45,
            "proc3_iter2_ttft_s": 1.28,
            "proc3_iter3_decode_tok_s": 67.94,
            "proc3_iter3_prefill_tok_s": 811.03,
            "proc3_iter3_ttft_s": 1.28,
            "proc3_iter4_decode_tok_s": 67.8,
            "proc3_iter4_prefill_tok_s": 809.32,
            "proc3_iter4_ttft_s": 1.28,
            "proc3_peak_private_footprint_mb": 22112.95,
            "proc3_peak_system_ram_mb_max": 22127.0,
            "proc4_cpu_idle_at_start_pct": 78.0,
            "proc4_init_total_ms": 38955.56,
            "proc4_iter1_decode_tok_s": 67.43,
            "proc4_iter1_prefill_tok_s": 772.55,
            "proc4_iter1_ttft_s": 1.34,
            "proc4_iter2_decode_tok_s": 67.7,
            "proc4_iter2_prefill_tok_s": 809.15,
            "proc4_iter2_ttft_s": 1.28,
            "proc4_iter3_decode_tok_s": 67.98,
            "proc4_iter3_prefill_tok_s": 811.48,
            "proc4_iter3_ttft_s": 1.28,
            "proc4_iter4_decode_tok_s": 67.21,
            "proc4_iter4_prefill_tok_s": 807.5,
            "proc4_iter4_ttft_s": 1.28,
            "proc4_peak_private_footprint_mb": 22115.23,
            "proc4_peak_system_ram_mb_max": 22128.8,
            "proc5_cpu_idle_at_start_pct": 87.0,
            "proc5_init_total_ms": 38812.02,
            "proc5_iter1_decode_tok_s": 67.97,
            "proc5_iter1_prefill_tok_s": 776.25,
            "proc5_iter1_ttft_s": 1.33,
            "proc5_iter2_decode_tok_s": 67.97,
            "proc5_iter2_prefill_tok_s": 808.17,
            "proc5_iter2_ttft_s": 1.28,
            "proc5_iter3_decode_tok_s": 68.35,
            "proc5_iter3_prefill_tok_s": 809.63,
            "proc5_iter3_ttft_s": 1.28,
            "proc5_iter4_decode_tok_s": 67.87,
            "proc5_iter4_prefill_tok_s": 806.97,
            "proc5_iter4_ttft_s": 1.28,
            "proc5_peak_private_footprint_mb": 22102.75,
            "proc5_peak_system_ram_mb_max": 22116.9,
            "processes": 5.0
          },
          "output_match": null,
          "peak_mem_mb": 22115.23,
          "prefill_tokens_per_s": 808.95,
          "provenance": "measured",
          "runs": true,
          "signature": null,
          "total_ops": null,
          "ttft_ms": 1280.0
        },
        "source": "data/device_runs/0.19.1/2026-11-12/nemotron-3-nano-30b-a3b-mixed-int4-ekv4096__mac-studio-m4-max.json"
      }
    ]
  },
  "model": {
    "family": "nemotron-3-nano",
    "id": "nemotron-3-nano-30b-a3b-mixed-int4-ekv4096",
    "license": "other (nvidia-nemotron-open-model-license)",
    "source_url": "https://huggingface.co/litert-community/Nemotron-3-Nano-30B-A3B",
    "task": "text-generation"
  },
  "pitfalls": [
    "Keep the GPU cache off (cache_dir=\":nocache\" in Python, --disable_cache=true in the command-line tool): on a Mac, runs that loaded an earlier run's GPU program cache answered wrongly (card L68).",
    "One Engine per conversation: the Mamba-2 state is not reset between conversations (LiteRT-LM #3165) (card L69).",
    "Tool calls come back as text (the model's <tool_call> form) (card L70).",
    "Long replies: a blocking send_message still generating after 10 minutes fails with DEADLINE_EXCEEDED; use send_message_async or cap max_output_tokens (card L71).",
    "After a message overflows the context window, every later message in that conversation fails (Task failed with state: 7); start a new conversation (card L72).",
    "GPU: the card runs it with the v0.19.1 command-line tool built from source, which has --sampler_handles_input=false (the PyPI package does not); LiteRT-LM 0.18.0's GPU moe op only supports activation 'gelu'. CPU: the PyPI package, litert-lm 0.19.1 (card L27, L76-L93).",
    "Thinking is on by default; thinking_config=litert_lm.ThinkingConfig(enable_thinking=False) in create_conversation turns it off (card L29, L105).",
    "The card's checks (12 questions and a three-turn chat, a name then 17 × 23, greedy, against NVIDIA's bf16 weights): with thinking on, 12/12 right and 391 for 17 × 23 on all four files on the CPU and the GPU (the source model was not run with thinking on); with thinking off, 11/12 right, as the source model, and 17 × 23 comes back as 401 for the mixed_int4 files (this one among them) and for the source model, 391 on the CPU and 411 on the GPU for the int4 files (card L59-L64).",
    "Which file: this one gives faster CPU prefill than int4 ekv4096 (card L38). On an Apple M4 Max (128 GB) its benchmark peaked at 23.2 GB on the GPU and 6.9 GB on the CPU (2026-11-12); speed still being tuned (updates follow) (card L27, L52-L53, L55)."
  ],
  "schema_version": "1.2"
}
