datasets
Training and evaluation data, with the modality, task and licence stated up front. Listed live from the Hugging Face Hub.
qfs-capture-a1ec5246f3ef
HF workflow a1ec5246f3ef11d323425ae8a9ea47ef
A root fidelity dataset in hidden form, produced by engines/tools/hf_capture.py from Qwen/Qwen3.8-27B.
The cut
the final hidden state handed to lm_head -- after the text model's final norm and immediately before the head matmul -- captured as the head module's input via torch.nn.Module.register_forward_pre_hook; replay applies the head ONLY (no final norm at replay time: the capture already sits after it). Same cut as… See the full description on the dataset page: https://huggingface.co/datasets/malaiwah/qfs-capture-a1ec5246f3ef.qfs-capture-6c6b45736e86
HF workflow 6c6b45736e86851500161bfeb9f44dd8
A root fidelity dataset in hidden form, produced by engines/tools/hf_capture.py from Qwen/Qwen3.8-27B.
The cut
the final hidden state handed to lm_head -- after the text model's final norm and immediately before the head matmul -- captured as the head module's input via torch.nn.Module.register_forward_pre_hook; replay applies the head ONLY (no final norm at replay time: the capture already sits after it). Same cut as… See the full description on the dataset page: https://huggingface.co/datasets/malaiwah/qfs-capture-6c6b45736e86.qfs-capture-02e41a4b6bc6
HF workflow 02e41a4b6bc63a9c8a9ffb68124ae5f7
A root fidelity dataset in hidden form, produced by engines/tools/hf_capture.py from Qwen/Qwen3.8-27B.
The cut
the final hidden state handed to lm_head -- after the text model's final norm and immediately before the head matmul -- captured as the head module's input via torch.nn.Module.register_forward_pre_hook; replay applies the head ONLY (no final norm at replay time: the capture already sits after it). Same cut as… See the full description on the dataset page: https://huggingface.co/datasets/malaiwah/qfs-capture-02e41a4b6bc6.qfs-fixture-root-captures-v1
QFS community fixture roots and captures
Twelve complete native random-initialized model fixtures, each with two independent CPU captures, forced full-vocabulary NumPy FP64 comparison, and a canonical published root dataset. These are pipeline tests, not trained assistants or language-quality benchmarks.
Start here
Use root-publications.json for exact model/root repositories and immutable revisions. Each canonical root is directly usable as… See the full description on the dataset page: https://huggingface.co/datasets/malaiwah/qfs-fixture-root-captures-v1.qfs-evidence-6c6b45736e86851500161bfeb9f44dd8-public
QFS workflow evidence
Recovered and receipt-checked root workflow. Provider verification and registry acceptance are not independent model reproduction.
Original HF Job
Exact producing source
Immutable execution plan · Worker result inventory · Provider readback
First capture · Second cold capture · Numerical reproduction control
Original raw panel · Panel build receipt · Original checkpoint license
The raw panel's referenced arrays are retained under input-panel/. Use the… See the full description on the dataset page: https://huggingface.co/datasets/malaiwah/qfs-evidence-6c6b45736e86851500161bfeb9f44dd8-public.qfs-smollm2-135m-wikitext2-campaign-v1
SmolLM2-135M QFS calibration and evaluation campaign
A small stored-weight fidelity study, not a broad model-quality benchmark.
Evaluation: 16 complete WikiText2 raw test articles, one 256-token window each, 4080 prediction positions.
Calibration: 32 disjoint train articles, 256 tokens each, 8192 calibration tokens.
Complete article title, normalized content and exact 13-token-ngram separation were checked. Validation is unused. Pretraining overlap remains unknown.
Original… See the full description on the dataset page: https://huggingface.co/datasets/malaiwah/qfs-smollm2-135m-wikitext2-campaign-v1.qfs-evidence-9bafb145820e0eba1117bbfcefff8abc-public
QFS workflow evidence
Recovered and receipt-checked candidate workflow. Provider verification and registry acceptance are not independent model reproduction.
Original HF Job
Exact producing source
Immutable execution plan · Worker result inventory · Provider readback
First capture · Second cold capture · Numerical reproduction control
Original raw panel · Panel build receipt · Original checkpoint license
The raw panel's referenced arrays are retained under input-panel/. Use… See the full description on the dataset page: https://huggingface.co/datasets/malaiwah/qfs-evidence-9bafb145820e0eba1117bbfcefff8abc-public.qfs-hf-jobs-campaign-evidence-20260909qfs-evidence-9ac7f94b2ab1bab8f4d9d65fc9af9003-public
QFS workflow evidence
Recovered and receipt-checked candidate workflow. Provider verification and registry acceptance are not independent model reproduction.
Original HF Job
Exact producing source
Immutable execution plan · Worker result inventory · Provider readback
First capture · Second cold capture · Numerical reproduction control
Original raw panel · Panel build receipt · Original checkpoint license
The raw panel's referenced arrays are retained under input-panel/. Use… See the full description on the dataset page: https://huggingface.co/datasets/malaiwah/qfs-evidence-9ac7f94b2ab1bab8f4d9d65fc9af9003-public.qfs-affine-tiny-cpu-format-v1
affine tiny CPU format fixture reproducibility
Complete tiny random FORMAT fixture evidence. Round-to-nearest (RTN) storage/reader exercise only; optimizer-not-run. No GPTQ/AWQ/AutoRound optimization, calibrated ModelOpt/CT/QAT quality, trained-model quality ranking, GPU parity, or native serving-kernel correctness claim.
Reconstructed weights are evaluated by the captured native forward. KL is own-head, full-vocabulary on the recorded panel, not a benchmark of training quality.… See the full description on the dataset page: https://huggingface.co/datasets/malaiwah/qfs-affine-tiny-cpu-format-v1.qfs-evidence-02e41a4b6bc63a9c8a9ffb68124ae5f7-public
QFS workflow evidence
Recovered and receipt-checked root workflow. Provider verification and registry acceptance are not independent model reproduction.
Original HF Job
Exact producing source
Immutable execution plan · Worker result inventory · Provider readback
First capture · Second cold capture · Numerical reproduction control
Original raw panel · Panel build receipt · Original checkpoint license
The raw panel's referenced arrays are retained under input-panel/. Use the… See the full description on the dataset page: https://huggingface.co/datasets/malaiwah/qfs-evidence-02e41a4b6bc63a9c8a9ffb68124ae5f7-public.qfs-evidence-90737fc68a3c2f65761b7211ae03b63f-public
QFS workflow evidence
Recovered and receipt-checked root workflow. Provider verification and registry acceptance are not independent model reproduction.
Original HF Job
Exact producing source
Immutable execution plan · Worker result inventory · Provider readback
First capture · Second cold capture · Numerical reproduction control
Original raw panel · Panel build receipt · Original checkpoint license
The raw panel's referenced arrays are retained under input-panel/. Use the… See the full description on the dataset page: https://huggingface.co/datasets/malaiwah/qfs-evidence-90737fc68a3c2f65761b7211ae03b63f-public.qfs-smollm2-135m-wikitext2-native-v1
HF workflow d3dc69602aeb981f06bd9f4c726937f9
A root fidelity dataset in hidden form, produced by engines/tools/hf_capture.py from malaiwah/SmolLM2-135M-QFS-native-bf16.
The cut
the final hidden state handed to lm_head -- after the text model's final norm and immediately before the head matmul -- captured as the head module's input via torch.nn.Module.register_forward_pre_hook; replay applies the head ONLY (no final norm at replay time: the capture already sits… See the full description on the dataset page: https://huggingface.co/datasets/malaiwah/qfs-smollm2-135m-wikitext2-native-v1.qfs-microfloat-tiny-cpu-format-v1
microfloat tiny CPU format fixture reproducibility
Complete tiny random FORMAT fixture evidence. Round-to-nearest (RTN) storage/reader exercise only; optimizer-not-run. No GPTQ/AWQ/AutoRound optimization, calibrated ModelOpt/CT/QAT quality, trained-model quality ranking, GPU parity, or native serving-kernel correctness claim.
Reconstructed weights are evaluated by the captured native forward. KL is own-head, full-vocabulary on the recorded panel, not a benchmark of training… See the full description on the dataset page: https://huggingface.co/datasets/malaiwah/qfs-microfloat-tiny-cpu-format-v1.qfs-capture-90737fc68a3c
HF workflow 90737fc68a3c2f65761b7211ae03b63f
A root fidelity dataset in hidden form, produced by engines/tools/hf_capture.py from malaiwah/glm-moe-dsa-tiny-random-bf16.
The cut
the final hidden state handed to lm_head -- after the text model's final norm and immediately before the head matmul -- captured as the head module's input via torch.nn.Module.register_forward_pre_hook; replay applies the head ONLY (no final norm at replay time: the capture already sits… See the full description on the dataset page: https://huggingface.co/datasets/malaiwah/qfs-capture-90737fc68a3c.qfs-existing-tiny-cpu-format-v1
existing tiny CPU format fixture reproducibility
Complete tiny random FORMAT fixture evidence. Round-to-nearest (RTN) storage/reader exercise only; optimizer-not-run. No GPTQ/AWQ/AutoRound optimization, calibrated ModelOpt/CT/QAT quality, trained-model quality ranking, GPU parity, or native serving-kernel correctness claim.
Reconstructed weights are evaluated by the captured native forward. KL is own-head, full-vocabulary on the recorded panel, not a benchmark of training… See the full description on the dataset page: https://huggingface.co/datasets/malaiwah/qfs-existing-tiny-cpu-format-v1.qfs-qwen-gguf-tiny-cpu-format-v1
qwen-gguf tiny CPU format fixture reproducibility
Complete tiny random FORMAT fixture evidence. Round-to-nearest (RTN) storage/reader exercise only; optimizer-not-run. No GPTQ/AWQ/AutoRound optimization, calibrated ModelOpt/CT/QAT quality, trained-model quality ranking, GPU parity, or native serving-kernel correctness claim.
Reconstructed weights are evaluated by the captured native forward. KL is own-head, full-vocabulary on the recorded panel, not a benchmark of training… See the full description on the dataset page: https://huggingface.co/datasets/malaiwah/qfs-qwen-gguf-tiny-cpu-format-v1.qfs-evidence-32c6ab05b0ceab1cecdceda838846388-public
QFS workflow evidence
Recovered and receipt-checked candidate workflow. Provider verification and registry acceptance are not independent model reproduction.
Original HF Job
Exact producing source
Immutable execution plan · Worker result inventory · Provider readback
First capture · Second cold capture · Numerical reproduction control
Original raw panel · Panel build receipt · Original checkpoint license
The raw panel's referenced arrays are retained under input-panel/. Use… See the full description on the dataset page: https://huggingface.co/datasets/malaiwah/qfs-evidence-32c6ab05b0ceab1cecdceda838846388-public.qfs-capture-ddb0a2bd89f1
HF workflow ddb0a2bd89f1b76a3740f9f307d11372
A root fidelity dataset in hidden form, produced by engines/tools/hf_capture.py from malaiwah/glm-moe-dsa-tiny-random-bf16.
The cut
the final hidden state handed to lm_head -- after the text model's final norm and immediately before the head matmul -- captured as the head module's input via torch.nn.Module.register_forward_pre_hook; replay applies the head ONLY (no final norm at replay time: the capture already sits… See the full description on the dataset page: https://huggingface.co/datasets/malaiwah/qfs-capture-ddb0a2bd89f1.qfs-evidence-ddb0a2bd89f1b76a3740f9f307d11372-public
QFS workflow evidence
Recovered and receipt-checked root workflow. Provider verification and registry acceptance are not independent model reproduction.
Original HF Job
Exact producing source
Immutable execution plan · Worker result inventory · Provider readback
First capture · Second cold capture · Numerical reproduction control
Original raw panel · Panel build receipt · Original checkpoint license
The raw panel's referenced arrays are retained under input-panel/. Use the… See the full description on the dataset page: https://huggingface.co/datasets/malaiwah/qfs-evidence-ddb0a2bd89f1b76a3740f9f307d11372-public.qfs-evidence-6856ba5826bf2036dc98aaf97365af12-public
QFS workflow evidence
Recovered and receipt-checked candidate workflow. Provider verification and registry acceptance are not independent model reproduction.
Original HF Job
Exact producing source
Immutable execution plan · Worker result inventory · Provider readback
First capture · Second cold capture · Numerical reproduction control
Original raw panel · Panel build receipt · Original checkpoint license
The raw panel's referenced arrays are retained under input-panel/. Use… See the full description on the dataset page: https://huggingface.co/datasets/malaiwah/qfs-evidence-6856ba5826bf2036dc98aaf97365af12-public.qfs-evidence-ee13c03b2f256042a17c9817e17a2f20-public
QFS workflow evidence
Recovered and receipt-checked candidate workflow. Provider verification and registry acceptance are not independent model reproduction.
Original HF Job
Exact producing source
Immutable execution plan · Worker result inventory · Provider readback
First capture · Second cold capture · Numerical reproduction control
Original raw panel · Panel build receipt · Original checkpoint license
The raw panel's referenced arrays are retained under input-panel/. Use… See the full description on the dataset page: https://huggingface.co/datasets/malaiwah/qfs-evidence-ee13c03b2f256042a17c9817e17a2f20-public.qfs-smollm2-135m-wikitext2-gptq-g32-v1
HF workflow 32c6ab05b0ceab1cecdceda838846388
A quant fidelity dataset in hidden form, produced by engines/tools/hf_capture.py from malaiwah/SmolLM2-135M-QFS-gptq-int4-g32.
The cut
the final hidden state handed to lm_head -- after the text model's final norm and immediately before the head matmul -- captured as the head module's input via torch.nn.Module.register_forward_pre_hook; replay applies the head ONLY (no final norm at replay time: the capture already sits… See the full description on the dataset page: https://huggingface.co/datasets/malaiwah/qfs-smollm2-135m-wikitext2-gptq-g32-v1.qfs-evidence-a1ec5246f3ef11d323425ae8a9ea47ef-public
QFS workflow evidence
Recovered and receipt-checked root workflow. Provider verification and registry acceptance are not independent model reproduction.
Original HF Job
Exact producing source
Immutable execution plan · Worker result inventory · Provider readback
First capture · Second cold capture · Numerical reproduction control
Original raw panel · Panel build receipt · Original checkpoint license
The raw panel's referenced arrays are retained under input-panel/. Use the… See the full description on the dataset page: https://huggingface.co/datasets/malaiwah/qfs-evidence-a1ec5246f3ef11d323425ae8a9ea47ef-public.qfs-capture-9bafb145820e
HF workflow 9bafb145820e0eba1117bbfcefff8abc
A quant fidelity dataset in hidden form, produced by engines/tools/hf_capture.py from malaiwah/qwen3-5-tiny-random-gptq-v1-g32-rtn-format.
The cut
the final hidden state handed to lm_head -- after the text model's final norm and immediately before the head matmul -- captured as the head module's input via torch.nn.Module.register_forward_pre_hook; replay applies the head ONLY (no final norm at replay time: the capture… See the full description on the dataset page: https://huggingface.co/datasets/malaiwah/qfs-capture-9bafb145820e.qfs-capture-571d4d7defe5
HF workflow 571d4d7defe5266c928e7dcbf77c2791
A root fidelity dataset in hidden form, produced by engines/tools/hf_capture.py from malaiwah/glm-moe-dsa-tiny-random-bf16.
The cut
the final hidden state handed to lm_head -- after the text model's final norm and immediately before the head matmul -- captured as the head module's input via torch.nn.Module.register_forward_pre_hook; replay applies the head ONLY (no final norm at replay time: the capture already sits… See the full description on the dataset page: https://huggingface.co/datasets/malaiwah/qfs-capture-571d4d7defe5.qfs-smollm2-135m-wikitext2-gptq-g64-v1
HF workflow 73f0a12a901c7368794a3a886f55b675
A quant fidelity dataset in hidden form, produced by engines/tools/hf_capture.py from malaiwah/SmolLM2-135M-QFS-gptq-int4-g64.
The cut
the final hidden state handed to lm_head -- after the text model's final norm and immediately before the head matmul -- captured as the head module's input via torch.nn.Module.register_forward_pre_hook; replay applies the head ONLY (no final norm at replay time: the capture already sits… See the full description on the dataset page: https://huggingface.co/datasets/malaiwah/qfs-smollm2-135m-wikitext2-gptq-g64-v1.qfs-evidence-73f0a12a901c7368794a3a886f55b675-public
QFS workflow evidence
Recovered and receipt-checked candidate workflow. Provider verification and registry acceptance are not independent model reproduction.
Original HF Job
Exact producing source
Immutable execution plan · Worker result inventory · Provider readback
First capture · Second cold capture · Numerical reproduction control
Original raw panel · Panel build receipt · Original checkpoint license
The raw panel's referenced arrays are retained under input-panel/. Use… See the full description on the dataset page: https://huggingface.co/datasets/malaiwah/qfs-evidence-73f0a12a901c7368794a3a886f55b675-public.qfs-evidence-8135e1ebdb07afa3754d21f43e991b80-public
QFS workflow evidence
Recovered and receipt-checked root workflow. Provider verification and registry acceptance are not independent model reproduction.
Original HF Job
Exact producing source
Immutable execution plan · Worker result inventory · Provider readback
First capture · Second cold capture · Numerical reproduction control
Original raw panel · Panel build receipt · Original checkpoint license
The raw panel's referenced arrays are retained under input-panel/. Use the… See the full description on the dataset page: https://huggingface.co/datasets/malaiwah/qfs-evidence-8135e1ebdb07afa3754d21f43e991b80-public.qfs-capture-6856ba5826bf
HF workflow 6856ba5826bf2036dc98aaf97365af12
A quant fidelity dataset in hidden form, produced by engines/tools/hf_capture.py from malaiwah/Qwen3.8-27B-EXL3-K5K6-hydrated.
The cut
the final hidden state handed to lm_head -- after the text model's final norm and immediately before the head matmul -- captured as the head module's input via torch.nn.Module.register_forward_pre_hook; replay applies the head ONLY (no final norm at replay time: the capture already sits… See the full description on the dataset page: https://huggingface.co/datasets/malaiwah/qfs-capture-6856ba5826bf.