deepseek-v4-tiny-random-bf16 / evidence-summary.json
malaiwah's picture
Complete native tiny CPU fixture and measured reproduction evidence
185723b verified
Raw History Blame Contribute Delete
42 kB
{
"captures": {
"first": {
"capture_content_digest": "4c091d00bcbfa482e7255b298fe8e5be16fe8318ff688ac8d25c13351ff82f4f",
"checkpoint_identity_sha256": "fc23543a91706d18ac96a8ebfb06e934c582f3ca7f4511b3d51a2cc8e83ecae9",
"cold_run": "cpu-first-a593e5769bd5482c857c2b089bbb33cc",
"dataset_sha256": "40b94307d66dcde0b264acd8f32e153e73dc8e23ac3a549cb53d0b5e42d8e86d",
"generation_sanity_probe": {
"degenerate": false,
"enforced": false,
"entropy_nats": 5.595019913847148,
"expect": null,
"logit_spread": 0.98046875,
"note": "one extra window through the same schedule; its hidden state is discarded and never enters the dataset.",
"prompt": "The capital of France is",
"schema": "malaiwah.generation-sanity-probe.v1",
"status": "recorded",
"top1_id": 10,
"top1_probability": 0.006240478005749957,
"top1_text": "'",
"top_k": [
{
"id": 10,
"probability": 0.006240478005749957,
"text": "'"
},
{
"id": 262,
"probability": 0.005443041658056204,
"text": "<reserved_262>"
},
{
"id": 202,
"probability": 0.005133277186172279,
"text": "\n"
},
{
"id": 195,
"probability": 0.005113264435015918,
"text": "\u0003"
},
{
"id": 269,
"probability": 0.005063573239716774,
"text": "<reserved_269>"
}
],
"uniform_entropy_nats": 5.605802066295998,
"vocab_size": 272
},
"head_tensor_content_sha256": "5c5d4b4405a5209284d8ce4512c873872a967aebb2c8c3be578d2381d4a7c701",
"memory_report": {
"context_length": 64,
"decoder_layers": null,
"device": "cpu",
"hidden_size": 64,
"layer_residency": null,
"model": "malaiwah/deepseek-v4-tiny-random-bf16",
"peak_resident_weight_bytes": 794644,
"peak_resident_weight_detail": {
"buffers": 13216,
"parameters": 781428,
"total": 794644
},
"peak_resident_weight_gb": 0.001,
"peak_rss_bytes": 371982336,
"peak_rss_gb": 0.372,
"resident_weight_note": "the maximum, over the run, of the materialised parameter+buffer bytes. On the CPU path peak_rss_bytes ALSO counts the safetensors mmap page cache, which is evictable and is not held by the schedule; this figure is not confounded by it. On CUDA, peak_cuda_allocated_bytes is the authoritative number because it includes activations and workspace and has no page cache.",
"rss_units_source": "Linux ru_maxrss",
"schedule": "window-outer",
"vocab_size": 272,
"windows": 4
},
"path": "first",
"pid": 1678251,
"resources": {
"bytes": {
"checkpoint_read": 816683,
"hidden_d2h": 36640,
"weights_h2d": null
},
"checkpoint_bytes": 816683,
"checkpoint_files": 2,
"device_name": null,
"forward_timing": "wall-clock",
"note": "provenance, not identity: peaks and stage seconds this run measured; forward_sum is device time under cuda-events, wall time otherwise; seal covers head serialisation up to this block's assembly",
"peak_cuda_allocated_bytes": null,
"peak_cuda_reserved_bytes": null,
"peak_resident_weight_bytes": 794644,
"peak_rss_bytes": 371982336,
"rss_units_source": "Linux ru_maxrss",
"seconds": {
"decode_sum": null,
"elapsed": 2.979,
"fill_sum": null,
"forward_sum": 0.296,
"identity": 0.005,
"layer_load_max": null,
"layer_load_sum": null,
"layer_loads": null,
"resident_load": 2.642,
"seal": 0.009
}
},
"runtime_versions": {
"cuda_runtime": null,
"executable": "/tmp/qfs-tiny-fixture-venv/bin/python",
"huggingface_hub": "1.30.0",
"machine": "x86_64",
"numpy": "2.5.3",
"pid": 1678251,
"platform": "Linux-7.0.0-30-generic-x86_64-with-glibc2.43",
"python": "3.12.14 (main, Aug 14 2026, 15:34:45) [Clang 22.1.3 ]",
"safetensors": "0.8.0",
"stage": "fixture_child_runtime",
"thread_environment": {
"MKL_NUM_THREADS": "2",
"NUMEXPR_NUM_THREADS": "2",
"OMP_NUM_THREADS": "2",
"OPENBLAS_NUM_THREADS": "2",
"VECLIB_MAXIMUM_THREADS": "2"
},
"tokenizers": "0.23.2",
"torch": "2.11.0+cpu",
"torch_num_interop_threads": 2,
"torch_num_threads": 2,
"transformers": "5.16.1"
},
"stack_fingerprint_sha256": "d4de1239e0d02e893fc23558cf4807b395712dc4d3eefa8d267a36a31fa5e468"
},
"repeat": {
"capture_content_digest": "4c091d00bcbfa482e7255b298fe8e5be16fe8318ff688ac8d25c13351ff82f4f",
"checkpoint_identity_sha256": "fc23543a91706d18ac96a8ebfb06e934c582f3ca7f4511b3d51a2cc8e83ecae9",
"cold_run": "cpu-repeat-54ae984a11ce4dd981b59cdce9e52596",
"dataset_sha256": "86d9d51dcb8f435855ef195a309935ab92a737d916665ead464ed804878bc2e4",
"generation_sanity_probe": {
"degenerate": false,
"enforced": false,
"entropy_nats": 5.595019913847148,
"expect": null,
"logit_spread": 0.98046875,
"note": "one extra window through the same schedule; its hidden state is discarded and never enters the dataset.",
"prompt": "The capital of France is",
"schema": "malaiwah.generation-sanity-probe.v1",
"status": "recorded",
"top1_id": 10,
"top1_probability": 0.006240478005749957,
"top1_text": "'",
"top_k": [
{
"id": 10,
"probability": 0.006240478005749957,
"text": "'"
},
{
"id": 262,
"probability": 0.005443041658056204,
"text": "<reserved_262>"
},
{
"id": 202,
"probability": 0.005133277186172279,
"text": "\n"
},
{
"id": 195,
"probability": 0.005113264435015918,
"text": "\u0003"
},
{
"id": 269,
"probability": 0.005063573239716774,
"text": "<reserved_269>"
}
],
"uniform_entropy_nats": 5.605802066295998,
"vocab_size": 272
},
"head_tensor_content_sha256": "5c5d4b4405a5209284d8ce4512c873872a967aebb2c8c3be578d2381d4a7c701",
"memory_report": {
"context_length": 64,
"decoder_layers": null,
"device": "cpu",
"hidden_size": 64,
"layer_residency": null,
"model": "malaiwah/deepseek-v4-tiny-random-bf16",
"peak_resident_weight_bytes": 794644,
"peak_resident_weight_detail": {
"buffers": 13216,
"parameters": 781428,
"total": 794644
},
"peak_resident_weight_gb": 0.001,
"peak_rss_bytes": 372019200,
"peak_rss_gb": 0.372,
"resident_weight_note": "the maximum, over the run, of the materialised parameter+buffer bytes. On the CPU path peak_rss_bytes ALSO counts the safetensors mmap page cache, which is evictable and is not held by the schedule; this figure is not confounded by it. On CUDA, peak_cuda_allocated_bytes is the authoritative number because it includes activations and workspace and has no page cache.",
"rss_units_source": "Linux ru_maxrss",
"schedule": "window-outer",
"vocab_size": 272,
"windows": 4
},
"path": "repeat",
"pid": 1679873,
"resources": {
"bytes": {
"checkpoint_read": 816683,
"hidden_d2h": 36640,
"weights_h2d": null
},
"checkpoint_bytes": 816683,
"checkpoint_files": 2,
"device_name": null,
"forward_timing": "wall-clock",
"note": "provenance, not identity: peaks and stage seconds this run measured; forward_sum is device time under cuda-events, wall time otherwise; seal covers head serialisation up to this block's assembly",
"peak_cuda_allocated_bytes": null,
"peak_cuda_reserved_bytes": null,
"peak_resident_weight_bytes": 794644,
"peak_rss_bytes": 372019200,
"rss_units_source": "Linux ru_maxrss",
"seconds": {
"decode_sum": null,
"elapsed": 2.721,
"fill_sum": null,
"forward_sum": 0.301,
"identity": 0.007,
"layer_load_max": null,
"layer_load_sum": null,
"layer_loads": null,
"resident_load": 2.377,
"seal": 0.009
}
},
"runtime_versions": {
"cuda_runtime": null,
"executable": "/tmp/qfs-tiny-fixture-venv/bin/python",
"huggingface_hub": "1.30.0",
"machine": "x86_64",
"numpy": "2.5.3",
"pid": 1679873,
"platform": "Linux-7.0.0-30-generic-x86_64-with-glibc2.43",
"python": "3.12.14 (main, Aug 14 2026, 15:34:45) [Clang 22.1.3 ]",
"safetensors": "0.8.0",
"stage": "fixture_child_runtime",
"thread_environment": {
"MKL_NUM_THREADS": "2",
"NUMEXPR_NUM_THREADS": "2",
"OMP_NUM_THREADS": "2",
"OPENBLAS_NUM_THREADS": "2",
"VECLIB_MAXIMUM_THREADS": "2"
},
"tokenizers": "0.23.2",
"torch": "2.11.0+cpu",
"torch_num_interop_threads": 2,
"torch_num_threads": 2,
"transformers": "5.16.1"
},
"stack_fingerprint_sha256": "d4de1239e0d02e893fc23558cf4807b395712dc4d3eefa8d267a36a31fa5e468"
}
},
"code_repository": "malaiwah/deepseek-v4-tiny-random-bf16",
"code_revision": "4f9dcfcbabdbe336f5f83e836dd8fac7138dda23",
"commands": [
{
"argv": [
"/tmp/qfs-tiny-fixture-venv/bin/python",
"-c",
"\nimport json, os, platform, runpy, sys\nimport numpy, torch, transformers, safetensors, tokenizers\nimport huggingface_hub\ntorch.set_num_threads(2)\ntorch.set_num_interop_threads(2)\nprint(json.dumps({\"stage\": \"fixture_child_runtime\", \"pid\": os.getpid(),\n \"python\": sys.version, \"executable\": sys.executable,\n \"platform\": platform.platform(), \"machine\": platform.machine(),\n \"torch\": torch.__version__, \"transformers\": transformers.__version__,\n \"numpy\": numpy.__version__, \"safetensors\": safetensors.__version__,\n \"tokenizers\": tokenizers.__version__, \"huggingface_hub\": huggingface_hub.__version__,\n \"torch_num_threads\": torch.get_num_threads(),\n \"torch_num_interop_threads\": torch.get_num_interop_threads(),\n \"cuda_runtime\": torch.version.cuda,\n \"thread_environment\": {k: os.environ.get(k) for k in\n (\"OMP_NUM_THREADS\", \"MKL_NUM_THREADS\", \"OPENBLAS_NUM_THREADS\",\n \"NUMEXPR_NUM_THREADS\", \"VECLIB_MAXIMUM_THREADS\")}}, sort_keys=True), flush=True)\nsys.argv = sys.argv[1:]\nrunpy.run_path(sys.argv[0], run_name=\"__main__\")\n",
"/tmp/qfs-capture-pin-snapshot/engines/tools/hf_capture.py",
"--model",
"/tmp/qfs-deepseek-v4-tiny-release/stage-corrected/model",
"--weights-repository",
"malaiwah/deepseek-v4-tiny-random-bf16",
"--model-revision",
"4f9dcfcbabdbe336f5f83e836dd8fac7138dda23",
"--panel",
"/tmp/qfs-deepseek-v4-tiny-release/stage-corrected/panel",
"--panel-role",
"final",
"--panel-binding",
"/tmp/qfs-deepseek-v4-tiny-release/stage-corrected/reproduction/panel-binding.json",
"--panel-binding-sha256",
"434f18780ab22ea57d86d6551ccb35c32a7ce5107ac89d66c3181b80d2e63ffd",
"--panel-tokenizer-root",
"/tmp/qfs-deepseek-v4-tiny-release/stage-corrected/model",
"--role",
"root",
"--lane",
"other",
"--device",
"cpu",
"--dtype",
"bfloat16",
"--drop-parallel-plan",
"--schedule",
"window-outer",
"--trust-remote-code",
"--code-repository",
"malaiwah/deepseek-v4-tiny-random-bf16",
"--code-revision",
"4f9dcfcbabdbe336f5f83e836dd8fac7138dda23",
"--dataset-id",
"fidelity--deepseek-v4-tiny.malaiwah.cpu-root-v1",
"--dataset-name",
"DeepSeek-V4 text tiny random CPU root",
"--repository",
"malaiwah/deepseek-v4-tiny-cpu-repro-v1",
"--author",
"malaiwah",
"--disclosures",
"[{\"code\":\"random_init_cpu_fixture\",\"severity\":\"info\",\"affects_comparability\":false,\"detail\":\"Complete randomly initialized tiny deepseek_v4 native-primitives checkpoint with explicit pinned per-instance RMSNorm FP32-weighted/final-input-dtype correction; CPU smoke fixture only, not trained quality evidence, unmodified-native arithmetic or a registry measurement. Lane other denotes CPU, not local-cuda-budget.\"}]",
"--out",
"/tmp/qfs-deepseek-v4-tiny-release/stage-corrected/reproduction/first",
"--run-name",
"cpu-first-a593e5769bd5482c857c2b089bbb33cc",
"--cold-run",
"cpu-first-a593e5769bd5482c857c2b089bbb33cc",
"--memory-report",
"/tmp/qfs-deepseek-v4-tiny-release/stage-corrected/reproduction/first.memory.json"
],
"cwd": "/tmp/qfs-capture-pin-snapshot",
"elapsed_seconds": 7.03227502387017,
"name": "capture-first",
"pid": 1678251,
"returncode": 0,
"runtime": {
"cuda_runtime": null,
"executable": "/tmp/qfs-tiny-fixture-venv/bin/python",
"huggingface_hub": "1.30.0",
"machine": "x86_64",
"numpy": "2.5.3",
"pid": 1678251,
"platform": "Linux-7.0.0-30-generic-x86_64-with-glibc2.43",
"python": "3.12.14 (main, Aug 14 2026, 15:34:45) [Clang 22.1.3 ]",
"safetensors": "0.8.0",
"stage": "fixture_child_runtime",
"thread_environment": {
"MKL_NUM_THREADS": "2",
"NUMEXPR_NUM_THREADS": "2",
"OMP_NUM_THREADS": "2",
"OPENBLAS_NUM_THREADS": "2",
"VECLIB_MAXIMUM_THREADS": "2"
},
"tokenizers": "0.23.2",
"torch": "2.11.0+cpu",
"torch_num_interop_threads": 2,
"torch_num_threads": 2,
"transformers": "5.16.1"
},
"stderr": "capture-first.stderr.log",
"stdout": "capture-first.stdout.log",
"timed_out": false,
"timeout_seconds": 1200
},
{
"argv": [
"/tmp/qfs-tiny-fixture-venv/bin/python",
"-c",
"\nimport json, os, platform, runpy, sys\nimport numpy, torch, transformers, safetensors, tokenizers\nimport huggingface_hub\ntorch.set_num_threads(2)\ntorch.set_num_interop_threads(2)\nprint(json.dumps({\"stage\": \"fixture_child_runtime\", \"pid\": os.getpid(),\n \"python\": sys.version, \"executable\": sys.executable,\n \"platform\": platform.platform(), \"machine\": platform.machine(),\n \"torch\": torch.__version__, \"transformers\": transformers.__version__,\n \"numpy\": numpy.__version__, \"safetensors\": safetensors.__version__,\n \"tokenizers\": tokenizers.__version__, \"huggingface_hub\": huggingface_hub.__version__,\n \"torch_num_threads\": torch.get_num_threads(),\n \"torch_num_interop_threads\": torch.get_num_interop_threads(),\n \"cuda_runtime\": torch.version.cuda,\n \"thread_environment\": {k: os.environ.get(k) for k in\n (\"OMP_NUM_THREADS\", \"MKL_NUM_THREADS\", \"OPENBLAS_NUM_THREADS\",\n \"NUMEXPR_NUM_THREADS\", \"VECLIB_MAXIMUM_THREADS\")}}, sort_keys=True), flush=True)\nsys.argv = sys.argv[1:]\nrunpy.run_path(sys.argv[0], run_name=\"__main__\")\n",
"/tmp/qfs-capture-pin-snapshot/bin/fidelity_dataset.py",
"verify",
"/tmp/qfs-deepseek-v4-tiny-release/stage-corrected/reproduction/first",
"--verify-tensors",
"--json",
"/tmp/qfs-deepseek-v4-tiny-release/stage-corrected/reproduction/first.verify.json"
],
"cwd": "/tmp/qfs-capture-pin-snapshot",
"elapsed_seconds": 3.8242294369265437,
"name": "verify-first",
"pid": 1679043,
"returncode": 0,
"runtime": {
"cuda_runtime": null,
"executable": "/tmp/qfs-tiny-fixture-venv/bin/python",
"huggingface_hub": "1.30.0",
"machine": "x86_64",
"numpy": "2.5.3",
"pid": 1679043,
"platform": "Linux-7.0.0-30-generic-x86_64-with-glibc2.43",
"python": "3.12.14 (main, Aug 14 2026, 15:34:45) [Clang 22.1.3 ]",
"safetensors": "0.8.0",
"stage": "fixture_child_runtime",
"thread_environment": {
"MKL_NUM_THREADS": "2",
"NUMEXPR_NUM_THREADS": "2",
"OMP_NUM_THREADS": "2",
"OPENBLAS_NUM_THREADS": "2",
"VECLIB_MAXIMUM_THREADS": "2"
},
"tokenizers": "0.23.2",
"torch": "2.11.0+cpu",
"torch_num_interop_threads": 2,
"torch_num_threads": 2,
"transformers": "5.16.1"
},
"stderr": "verify-first.stderr.log",
"stdout": "verify-first.stdout.log",
"timed_out": false,
"timeout_seconds": 1200
},
{
"argv": [
"/tmp/qfs-tiny-fixture-venv/bin/python",
"-c",
"\nimport json, os, platform, runpy, sys\nimport numpy, torch, transformers, safetensors, tokenizers\nimport huggingface_hub\ntorch.set_num_threads(2)\ntorch.set_num_interop_threads(2)\nprint(json.dumps({\"stage\": \"fixture_child_runtime\", \"pid\": os.getpid(),\n \"python\": sys.version, \"executable\": sys.executable,\n \"platform\": platform.platform(), \"machine\": platform.machine(),\n \"torch\": torch.__version__, \"transformers\": transformers.__version__,\n \"numpy\": numpy.__version__, \"safetensors\": safetensors.__version__,\n \"tokenizers\": tokenizers.__version__, \"huggingface_hub\": huggingface_hub.__version__,\n \"torch_num_threads\": torch.get_num_threads(),\n \"torch_num_interop_threads\": torch.get_num_interop_threads(),\n \"cuda_runtime\": torch.version.cuda,\n \"thread_environment\": {k: os.environ.get(k) for k in\n (\"OMP_NUM_THREADS\", \"MKL_NUM_THREADS\", \"OPENBLAS_NUM_THREADS\",\n \"NUMEXPR_NUM_THREADS\", \"VECLIB_MAXIMUM_THREADS\")}}, sort_keys=True), flush=True)\nsys.argv = sys.argv[1:]\nrunpy.run_path(sys.argv[0], run_name=\"__main__\")\n",
"/tmp/qfs-capture-pin-snapshot/bin/fidelity_dataset.py",
"validate",
"/tmp/qfs-deepseek-v4-tiny-release/stage-corrected/reproduction/first",
"--strict",
"--verify-tensors",
"--json",
"/tmp/qfs-deepseek-v4-tiny-release/stage-corrected/reproduction/first.strict.json"
],
"cwd": "/tmp/qfs-capture-pin-snapshot",
"elapsed_seconds": 3.572876560036093,
"name": "strict-first",
"pid": 1679397,
"returncode": 0,
"runtime": {
"cuda_runtime": null,
"executable": "/tmp/qfs-tiny-fixture-venv/bin/python",
"huggingface_hub": "1.30.0",
"machine": "x86_64",
"numpy": "2.5.3",
"pid": 1679397,
"platform": "Linux-7.0.0-30-generic-x86_64-with-glibc2.43",
"python": "3.12.14 (main, Aug 14 2026, 15:34:45) [Clang 22.1.3 ]",
"safetensors": "0.8.0",
"stage": "fixture_child_runtime",
"thread_environment": {
"MKL_NUM_THREADS": "2",
"NUMEXPR_NUM_THREADS": "2",
"OMP_NUM_THREADS": "2",
"OPENBLAS_NUM_THREADS": "2",
"VECLIB_MAXIMUM_THREADS": "2"
},
"tokenizers": "0.23.2",
"torch": "2.11.0+cpu",
"torch_num_interop_threads": 2,
"torch_num_threads": 2,
"transformers": "5.16.1"
},
"stderr": "strict-first.stderr.log",
"stdout": "strict-first.stdout.log",
"timed_out": false,
"timeout_seconds": 1200
},
{
"argv": [
"/tmp/qfs-tiny-fixture-venv/bin/python",
"-c",
"\nimport json, os, platform, runpy, sys\nimport numpy, torch, transformers, safetensors, tokenizers\nimport huggingface_hub\ntorch.set_num_threads(2)\ntorch.set_num_interop_threads(2)\nprint(json.dumps({\"stage\": \"fixture_child_runtime\", \"pid\": os.getpid(),\n \"python\": sys.version, \"executable\": sys.executable,\n \"platform\": platform.platform(), \"machine\": platform.machine(),\n \"torch\": torch.__version__, \"transformers\": transformers.__version__,\n \"numpy\": numpy.__version__, \"safetensors\": safetensors.__version__,\n \"tokenizers\": tokenizers.__version__, \"huggingface_hub\": huggingface_hub.__version__,\n \"torch_num_threads\": torch.get_num_threads(),\n \"torch_num_interop_threads\": torch.get_num_interop_threads(),\n \"cuda_runtime\": torch.version.cuda,\n \"thread_environment\": {k: os.environ.get(k) for k in\n (\"OMP_NUM_THREADS\", \"MKL_NUM_THREADS\", \"OPENBLAS_NUM_THREADS\",\n \"NUMEXPR_NUM_THREADS\", \"VECLIB_MAXIMUM_THREADS\")}}, sort_keys=True), flush=True)\nsys.argv = sys.argv[1:]\nrunpy.run_path(sys.argv[0], run_name=\"__main__\")\n",
"/tmp/qfs-capture-pin-snapshot/engines/tools/hf_capture.py",
"--model",
"/tmp/qfs-deepseek-v4-tiny-release/stage-corrected/model",
"--weights-repository",
"malaiwah/deepseek-v4-tiny-random-bf16",
"--model-revision",
"4f9dcfcbabdbe336f5f83e836dd8fac7138dda23",
"--panel",
"/tmp/qfs-deepseek-v4-tiny-release/stage-corrected/panel",
"--panel-role",
"final",
"--panel-binding",
"/tmp/qfs-deepseek-v4-tiny-release/stage-corrected/reproduction/panel-binding.json",
"--panel-binding-sha256",
"434f18780ab22ea57d86d6551ccb35c32a7ce5107ac89d66c3181b80d2e63ffd",
"--panel-tokenizer-root",
"/tmp/qfs-deepseek-v4-tiny-release/stage-corrected/model",
"--role",
"root",
"--lane",
"other",
"--device",
"cpu",
"--dtype",
"bfloat16",
"--drop-parallel-plan",
"--schedule",
"window-outer",
"--trust-remote-code",
"--code-repository",
"malaiwah/deepseek-v4-tiny-random-bf16",
"--code-revision",
"4f9dcfcbabdbe336f5f83e836dd8fac7138dda23",
"--dataset-id",
"fidelity--deepseek-v4-tiny.malaiwah.cpu-root-v1",
"--dataset-name",
"DeepSeek-V4 text tiny random CPU root",
"--repository",
"malaiwah/deepseek-v4-tiny-cpu-repro-v1",
"--author",
"malaiwah",
"--disclosures",
"[{\"code\":\"random_init_cpu_fixture\",\"severity\":\"info\",\"affects_comparability\":false,\"detail\":\"Complete randomly initialized tiny deepseek_v4 native-primitives checkpoint with explicit pinned per-instance RMSNorm FP32-weighted/final-input-dtype correction; CPU smoke fixture only, not trained quality evidence, unmodified-native arithmetic or a registry measurement. Lane other denotes CPU, not local-cuda-budget.\"}]",
"--out",
"/tmp/qfs-deepseek-v4-tiny-release/stage-corrected/reproduction/repeat",
"--run-name",
"cpu-repeat-54ae984a11ce4dd981b59cdce9e52596",
"--cold-run",
"cpu-repeat-54ae984a11ce4dd981b59cdce9e52596",
"--memory-report",
"/tmp/qfs-deepseek-v4-tiny-release/stage-corrected/reproduction/repeat.memory.json"
],
"cwd": "/tmp/qfs-capture-pin-snapshot",
"elapsed_seconds": 6.431408139877021,
"name": "capture-repeat",
"pid": 1679873,
"returncode": 0,
"runtime": {
"cuda_runtime": null,
"executable": "/tmp/qfs-tiny-fixture-venv/bin/python",
"huggingface_hub": "1.30.0",
"machine": "x86_64",
"numpy": "2.5.3",
"pid": 1679873,
"platform": "Linux-7.0.0-30-generic-x86_64-with-glibc2.43",
"python": "3.12.14 (main, Aug 14 2026, 15:34:45) [Clang 22.1.3 ]",
"safetensors": "0.8.0",
"stage": "fixture_child_runtime",
"thread_environment": {
"MKL_NUM_THREADS": "2",
"NUMEXPR_NUM_THREADS": "2",
"OMP_NUM_THREADS": "2",
"OPENBLAS_NUM_THREADS": "2",
"VECLIB_MAXIMUM_THREADS": "2"
},
"tokenizers": "0.23.2",
"torch": "2.11.0+cpu",
"torch_num_interop_threads": 2,
"torch_num_threads": 2,
"transformers": "5.16.1"
},
"stderr": "capture-repeat.stderr.log",
"stdout": "capture-repeat.stdout.log",
"timed_out": false,
"timeout_seconds": 1200
},
{
"argv": [
"/tmp/qfs-tiny-fixture-venv/bin/python",
"-c",
"\nimport json, os, platform, runpy, sys\nimport numpy, torch, transformers, safetensors, tokenizers\nimport huggingface_hub\ntorch.set_num_threads(2)\ntorch.set_num_interop_threads(2)\nprint(json.dumps({\"stage\": \"fixture_child_runtime\", \"pid\": os.getpid(),\n \"python\": sys.version, \"executable\": sys.executable,\n \"platform\": platform.platform(), \"machine\": platform.machine(),\n \"torch\": torch.__version__, \"transformers\": transformers.__version__,\n \"numpy\": numpy.__version__, \"safetensors\": safetensors.__version__,\n \"tokenizers\": tokenizers.__version__, \"huggingface_hub\": huggingface_hub.__version__,\n \"torch_num_threads\": torch.get_num_threads(),\n \"torch_num_interop_threads\": torch.get_num_interop_threads(),\n \"cuda_runtime\": torch.version.cuda,\n \"thread_environment\": {k: os.environ.get(k) for k in\n (\"OMP_NUM_THREADS\", \"MKL_NUM_THREADS\", \"OPENBLAS_NUM_THREADS\",\n \"NUMEXPR_NUM_THREADS\", \"VECLIB_MAXIMUM_THREADS\")}}, sort_keys=True), flush=True)\nsys.argv = sys.argv[1:]\nrunpy.run_path(sys.argv[0], run_name=\"__main__\")\n",
"/tmp/qfs-capture-pin-snapshot/bin/fidelity_dataset.py",
"verify",
"/tmp/qfs-deepseek-v4-tiny-release/stage-corrected/reproduction/repeat",
"--verify-tensors",
"--json",
"/tmp/qfs-deepseek-v4-tiny-release/stage-corrected/reproduction/repeat.verify.json"
],
"cwd": "/tmp/qfs-capture-pin-snapshot",
"elapsed_seconds": 3.6247158378828317,
"name": "verify-repeat",
"pid": 1680594,
"returncode": 0,
"runtime": {
"cuda_runtime": null,
"executable": "/tmp/qfs-tiny-fixture-venv/bin/python",
"huggingface_hub": "1.30.0",
"machine": "x86_64",
"numpy": "2.5.3",
"pid": 1680594,
"platform": "Linux-7.0.0-30-generic-x86_64-with-glibc2.43",
"python": "3.12.14 (main, Aug 14 2026, 15:34:45) [Clang 22.1.3 ]",
"safetensors": "0.8.0",
"stage": "fixture_child_runtime",
"thread_environment": {
"MKL_NUM_THREADS": "2",
"NUMEXPR_NUM_THREADS": "2",
"OMP_NUM_THREADS": "2",
"OPENBLAS_NUM_THREADS": "2",
"VECLIB_MAXIMUM_THREADS": "2"
},
"tokenizers": "0.23.2",
"torch": "2.11.0+cpu",
"torch_num_interop_threads": 2,
"torch_num_threads": 2,
"transformers": "5.16.1"
},
"stderr": "verify-repeat.stderr.log",
"stdout": "verify-repeat.stdout.log",
"timed_out": false,
"timeout_seconds": 1200
},
{
"argv": [
"/tmp/qfs-tiny-fixture-venv/bin/python",
"-c",
"\nimport json, os, platform, runpy, sys\nimport numpy, torch, transformers, safetensors, tokenizers\nimport huggingface_hub\ntorch.set_num_threads(2)\ntorch.set_num_interop_threads(2)\nprint(json.dumps({\"stage\": \"fixture_child_runtime\", \"pid\": os.getpid(),\n \"python\": sys.version, \"executable\": sys.executable,\n \"platform\": platform.platform(), \"machine\": platform.machine(),\n \"torch\": torch.__version__, \"transformers\": transformers.__version__,\n \"numpy\": numpy.__version__, \"safetensors\": safetensors.__version__,\n \"tokenizers\": tokenizers.__version__, \"huggingface_hub\": huggingface_hub.__version__,\n \"torch_num_threads\": torch.get_num_threads(),\n \"torch_num_interop_threads\": torch.get_num_interop_threads(),\n \"cuda_runtime\": torch.version.cuda,\n \"thread_environment\": {k: os.environ.get(k) for k in\n (\"OMP_NUM_THREADS\", \"MKL_NUM_THREADS\", \"OPENBLAS_NUM_THREADS\",\n \"NUMEXPR_NUM_THREADS\", \"VECLIB_MAXIMUM_THREADS\")}}, sort_keys=True), flush=True)\nsys.argv = sys.argv[1:]\nrunpy.run_path(sys.argv[0], run_name=\"__main__\")\n",
"/tmp/qfs-capture-pin-snapshot/bin/fidelity_dataset.py",
"validate",
"/tmp/qfs-deepseek-v4-tiny-release/stage-corrected/reproduction/repeat",
"--strict",
"--verify-tensors",
"--json",
"/tmp/qfs-deepseek-v4-tiny-release/stage-corrected/reproduction/repeat.strict.json"
],
"cwd": "/tmp/qfs-capture-pin-snapshot",
"elapsed_seconds": 3.7755501018837094,
"name": "strict-repeat",
"pid": 1681025,
"returncode": 0,
"runtime": {
"cuda_runtime": null,
"executable": "/tmp/qfs-tiny-fixture-venv/bin/python",
"huggingface_hub": "1.30.0",
"machine": "x86_64",
"numpy": "2.5.3",
"pid": 1681025,
"platform": "Linux-7.0.0-30-generic-x86_64-with-glibc2.43",
"python": "3.12.14 (main, Aug 14 2026, 15:34:45) [Clang 22.1.3 ]",
"safetensors": "0.8.0",
"stage": "fixture_child_runtime",
"thread_environment": {
"MKL_NUM_THREADS": "2",
"NUMEXPR_NUM_THREADS": "2",
"OMP_NUM_THREADS": "2",
"OPENBLAS_NUM_THREADS": "2",
"VECLIB_MAXIMUM_THREADS": "2"
},
"tokenizers": "0.23.2",
"torch": "2.11.0+cpu",
"torch_num_interop_threads": 2,
"torch_num_threads": 2,
"transformers": "5.16.1"
},
"stderr": "strict-repeat.stderr.log",
"stdout": "strict-repeat.stdout.log",
"timed_out": false,
"timeout_seconds": 1200
},
{
"argv": [
"/tmp/qfs-tiny-fixture-venv/bin/python",
"-c",
"\nimport json, os, platform, runpy, sys\nimport numpy, torch, transformers, safetensors, tokenizers\nimport huggingface_hub\ntorch.set_num_threads(2)\ntorch.set_num_interop_threads(2)\nprint(json.dumps({\"stage\": \"fixture_child_runtime\", \"pid\": os.getpid(),\n \"python\": sys.version, \"executable\": sys.executable,\n \"platform\": platform.platform(), \"machine\": platform.machine(),\n \"torch\": torch.__version__, \"transformers\": transformers.__version__,\n \"numpy\": numpy.__version__, \"safetensors\": safetensors.__version__,\n \"tokenizers\": tokenizers.__version__, \"huggingface_hub\": huggingface_hub.__version__,\n \"torch_num_threads\": torch.get_num_threads(),\n \"torch_num_interop_threads\": torch.get_num_interop_threads(),\n \"cuda_runtime\": torch.version.cuda,\n \"thread_environment\": {k: os.environ.get(k) for k in\n (\"OMP_NUM_THREADS\", \"MKL_NUM_THREADS\", \"OPENBLAS_NUM_THREADS\",\n \"NUMEXPR_NUM_THREADS\", \"VECLIB_MAXIMUM_THREADS\")}}, sort_keys=True), flush=True)\nsys.argv = sys.argv[1:]\nrunpy.run_path(sys.argv[0], run_name=\"__main__\")\n",
"/tmp/qfs-capture-pin-snapshot/bin/fidelity_dataset.py",
"compare",
"--reference",
"/tmp/qfs-deepseek-v4-tiny-release/stage-corrected/reproduction/first",
"--candidate",
"/tmp/qfs-deepseek-v4-tiny-release/stage-corrected/reproduction/repeat",
"--out",
"/tmp/qfs-deepseek-v4-tiny-release/stage-corrected/reproduction/comparison",
"--self-compare",
"--force-compute",
"--device",
"cpu",
"--replay-device",
"numpy",
"--replay-dtype",
"float32",
"--vocab-chunk",
"8192",
"--chunk-positions",
"128",
"--verify-tensors",
"--reference-label",
"cpu-first-a593e5769bd5482c857c2b089bbb33cc",
"--candidate-label",
"cpu-repeat-54ae984a11ce4dd981b59cdce9e52596"
],
"cwd": "/tmp/qfs-capture-pin-snapshot",
"elapsed_seconds": 3.5752729210071266,
"name": "compare",
"pid": 1681479,
"returncode": 0,
"runtime": {
"cuda_runtime": null,
"executable": "/tmp/qfs-tiny-fixture-venv/bin/python",
"huggingface_hub": "1.30.0",
"machine": "x86_64",
"numpy": "2.5.3",
"pid": 1681479,
"platform": "Linux-7.0.0-30-generic-x86_64-with-glibc2.43",
"python": "3.12.14 (main, Aug 14 2026, 15:34:45) [Clang 22.1.3 ]",
"safetensors": "0.8.0",
"stage": "fixture_child_runtime",
"thread_environment": {
"MKL_NUM_THREADS": "2",
"NUMEXPR_NUM_THREADS": "2",
"OMP_NUM_THREADS": "2",
"OPENBLAS_NUM_THREADS": "2",
"VECLIB_MAXIMUM_THREADS": "2"
},
"tokenizers": "0.23.2",
"torch": "2.11.0+cpu",
"torch_num_interop_threads": 2,
"torch_num_threads": 2,
"transformers": "5.16.1"
},
"stderr": "compare.stderr.log",
"stdout": "compare.stdout.log",
"timed_out": false,
"timeout_seconds": 1200
},
{
"argv": [
"/tmp/qfs-tiny-fixture-venv/bin/python",
"-c",
"\nimport json, os, platform, runpy, sys\nimport numpy, torch, transformers, safetensors, tokenizers\nimport huggingface_hub\ntorch.set_num_threads(2)\ntorch.set_num_interop_threads(2)\nprint(json.dumps({\"stage\": \"fixture_child_runtime\", \"pid\": os.getpid(),\n \"python\": sys.version, \"executable\": sys.executable,\n \"platform\": platform.platform(), \"machine\": platform.machine(),\n \"torch\": torch.__version__, \"transformers\": transformers.__version__,\n \"numpy\": numpy.__version__, \"safetensors\": safetensors.__version__,\n \"tokenizers\": tokenizers.__version__, \"huggingface_hub\": huggingface_hub.__version__,\n \"torch_num_threads\": torch.get_num_threads(),\n \"torch_num_interop_threads\": torch.get_num_interop_threads(),\n \"cuda_runtime\": torch.version.cuda,\n \"thread_environment\": {k: os.environ.get(k) for k in\n (\"OMP_NUM_THREADS\", \"MKL_NUM_THREADS\", \"OPENBLAS_NUM_THREADS\",\n \"NUMEXPR_NUM_THREADS\", \"VECLIB_MAXIMUM_THREADS\")}}, sort_keys=True), flush=True)\nsys.argv = sys.argv[1:]\nrunpy.run_path(sys.argv[0], run_name=\"__main__\")\n",
"/tmp/qfs-capture-pin-snapshot/bin/fidelity_dataset.py",
"validate",
"--receipt",
"/tmp/qfs-deepseek-v4-tiny-release/stage-corrected/reproduction/comparison/comparison-receipt.json",
"--strict",
"--json",
"/tmp/qfs-deepseek-v4-tiny-release/stage-corrected/reproduction/comparison.validate.json"
],
"cwd": "/tmp/qfs-capture-pin-snapshot",
"elapsed_seconds": 3.5747926908079535,
"name": "validate-comparison",
"pid": 1681835,
"returncode": 0,
"runtime": {
"cuda_runtime": null,
"executable": "/tmp/qfs-tiny-fixture-venv/bin/python",
"huggingface_hub": "1.30.0",
"machine": "x86_64",
"numpy": "2.5.3",
"pid": 1681835,
"platform": "Linux-7.0.0-30-generic-x86_64-with-glibc2.43",
"python": "3.12.14 (main, Aug 14 2026, 15:34:45) [Clang 22.1.3 ]",
"safetensors": "0.8.0",
"stage": "fixture_child_runtime",
"thread_environment": {
"MKL_NUM_THREADS": "2",
"NUMEXPR_NUM_THREADS": "2",
"OMP_NUM_THREADS": "2",
"OPENBLAS_NUM_THREADS": "2",
"VECLIB_MAXIMUM_THREADS": "2"
},
"tokenizers": "0.23.2",
"torch": "2.11.0+cpu",
"torch_num_interop_threads": 2,
"torch_num_threads": 2,
"transformers": "5.16.1"
},
"stderr": "validate-comparison.stderr.log",
"stdout": "validate-comparison.stdout.log",
"timed_out": false,
"timeout_seconds": 1200
}
],
"comparison": {
"comparator": {
"accumulation_dtype": "float64",
"bf16_reduced_precision_reduction": false,
"cublas_workspace_config": null,
"deterministic_algorithms": true,
"device": "cpu",
"estimator_backend": "torch:k6_kld_report._token_kld",
"head_applied_candidate_tensor_content_sha256": "5c5d4b4405a5209284d8ce4512c873872a967aebb2c8c3be578d2381d4a7c701",
"head_applied_reference_tensor_content_sha256": "5c5d4b4405a5209284d8ce4512c873872a967aebb2c8c3be578d2381d4a7c701",
"head_applied_tensor_content_sha256": "5c5d4b4405a5209284d8ce4512c873872a967aebb2c8c3be578d2381d4a7c701",
"logprob_dtype": "float64",
"position_block": 128,
"replay_backend": "numpy:cpu:float32",
"replay_env": {
"blas_configuration": "OpenBLAS 0.3.34.106.0 USE64BITINT DYNAMIC_ARCH NO_AFFINITY Haswell MAX_THREADS=64",
"blas_name": "scipy-openblas",
"blas_threads": 2,
"blas_threads_source": "ctypes:scipy_openblas_get_num_threads64_",
"blas_version": "0.3.34.106.0",
"cpu_model": "Intel(R) Xeon(R) CPU X5570 @ 2.93GHz",
"library": "numpy",
"numpy_version": "2.5.3",
"replay_dtype": "float32"
},
"replay_peak_device_bytes": null,
"short_circuited": true,
"source_file_hashes_verified": true,
"tensor_content_digests_verified": true,
"tf32": false,
"tool": {
"entrypoint": "bin/fidelity-dataset compare",
"note": "the fp64 estimator is k6_kld_report._token_kld, imported not copied, whenever torch is importable; estimator_backend records which path ran."
},
"two_pass": true,
"vocab_chunk": 8192
},
"comparison_kind": "reproduction_confirmation",
"kl": {
"max": 0.0,
"mean": 0.0,
"median": 0.0,
"p95": 0.0,
"p99": 0.0,
"p99_9": 0.0
},
"measurement_scope": {
"contexts": 4,
"covers_full_panel": true,
"position_filter": "all",
"positions_per_context": 63,
"scored_positions": 252,
"subset_detail": null
},
"metric": {
"direction": "reference_to_candidate",
"direction_label": "KL(reference || candidate)",
"higher_is_better": false,
"name": "mean_tokenwise_kld",
"units": "nats",
"value": 0.0
},
"note": "QFS retains its hash-proof result and top-1 value for reproduction confirmations; --force-compute independently runs the estimator and requires its tokenwise KL array to agree exactly. short_circuited=true therefore does not mean replay was skipped.",
"receipt": "comparison/comparison-receipt.json",
"receipt_sha256": "133f49204e58c4530590d46b7c95e70a4499e257dad1fbbdccb4e99133996d20",
"self_compare": {
"asserted_exact_zero": true,
"capture_content_digest_equal": true,
"expected_tokenwise_bytes": 2144,
"expected_tokenwise_sha256": "896ad5b6b732dce1588636c3af3a8a93af8eb1250df72809f6e26fc29914f65d",
"force_compute_agreed": true,
"head_digest_equal": true,
"weights_identity_equal": true
},
"tokenwise": {
"bytes": 2144,
"path": "tokenwise-kld.npy",
"sha256": "896ad5b6b732dce1588636c3af3a8a93af8eb1250df72809f6e26fc29914f65d"
},
"top1_agreement": 1.0
},
"cpu_threads": 2,
"dataset_id": "fidelity--deepseek-v4-tiny.malaiwah.cpu-root-v1",
"dataset_repository": "malaiwah/deepseek-v4-tiny-cpu-repro-v1",
"device": "cpu",
"driver_argv": [
"/tmp/qfs-tiny-fixture-venv/bin/python",
"/tmp/qfs-deepseek-v4-tiny-release/verify_cpu.py",
"--model-dir",
"/tmp/qfs-deepseek-v4-tiny-release/stage-corrected/model",
"--model-repository",
"malaiwah/deepseek-v4-tiny-random-bf16",
"--model-revision",
"4f9dcfcbabdbe336f5f83e836dd8fac7138dda23",
"--code-repository",
"malaiwah/deepseek-v4-tiny-random-bf16",
"--code-revision",
"4f9dcfcbabdbe336f5f83e836dd8fac7138dda23",
"--panel-dir",
"/tmp/qfs-deepseek-v4-tiny-release/stage-corrected/panel",
"--qfs-root",
"/tmp/qfs-capture-pin-snapshot",
"--out",
"/tmp/qfs-deepseek-v4-tiny-release/stage-corrected/reproduction"
],
"driver_sha256": "966abd10fd9e7dc178520cc1075b71ca06483cd048033f4c65fee5fefcd073d5",
"dtype": "bfloat16",
"elapsed_seconds": 35.589186880970374,
"lane": "other",
"lane_description": "local CPU random-init fixture",
"model_repository": "malaiwah/deepseek-v4-tiny-random-bf16",
"model_revision": "4f9dcfcbabdbe336f5f83e836dd8fac7138dda23",
"network_scope": "Capture children use local weights/tokenizer; HF network permitted for immutable reviewed code verification. Comparison children remain offline.",
"panel_binding": "panel-binding.json",
"panel_binding_sha256": "434f18780ab22ea57d86d6551ccb35c32a7ce5107ac89d66c3181b80d2e63ffd",
"provenance_note": "Local QFS source bytes are hashed; no source commit is asserted. The caller supplies an immutable model pin; QFS hashes the local checkpoint. This script does not independently prove Hub-to-local weight equality.",
"qfs_source_files": {
"bin/fidelity/dscompare.py": "4a53f735e1e3fb8bd739854451b57f34e8369bffa9fd4f7b7fd935fb2e7d52df",
"bin/fidelity/dsvalidate.py": "a9d6cd57e801e3ef044a37635e24518c83dbc6d1ec8a4c0b4b5d1b24483ca8d1",
"bin/fidelity/panel.py": "0bf78fca76289920e0dc10d58082f42f24f4f2db6e0ecde61daefa4c22de286d",
"bin/fidelity_dataset.py": "81940c51f59a6ffea3db2edad8a501a663d25b1e7fb1d96a3af2d875d18bd150",
"engines/tools/hf_capture.py": "ee98f9d13d28aacd1c07f0b4c42a8ba4106a631863fa7ebd7725283bb44cf855"
},
"schema": "malaiwah.deepseek-v4-tiny-cpu-reproduction.v1",
"scope": "Random-init CPU reproduction only; no quality claim or registry submission.",
"status": "passed"
}