flint-FP8-NVFP4 / export-validation.json
HawkBearPig's picture
Publish 35.74 GB FP8/NVFP4 policy with held-out quality, exact export replay and generation checks
95e7465 verified
Raw History Blame Contribute Delete
11.2 kB
{
"status": "passed",
"policy": "budget36-swap1",
"purpose": "Export integrity replay, not new independent quality evidence",
"all_per_window_metrics_exactly_match_frozen_evaluation": true,
"quantization_manifest_sha256": "9c22fa2989edd9f9e46acf8d76e35243b6aec4643182ba2b4195fd2ed2c3d4a7",
"expected_sha256": "0198b1989b7109125bb3116394a3c414751e95559e1e48a9970d8ceae13418d1",
"corpus_sha256": "ad58c37420891d4c5b08181aace2e75f00ac3f8a3253269b89a0f8ee88418923",
"script_sha256": "7f1496529c034d77a944ee80cf3cb47ad98cf988b7614ac3b608afee5b20ae8e",
"reference_script_sha256": "2534d51af8e99699692f1a2cf5fc022bd7beaa7de07770dbffb9690aa4f62ff6",
"normalization_script_sha256": "f06c3d51f9018814648a1fe42c39274dd103528b30185d914f76adbd68fef3ac",
"policy_script_sha256": "e0e0aa655a97d9c78837196a94b280c54944ab5542686242dbaa7604d33dca1c",
"provenance": {
"source": {
"repo": "RecursiveSI/flint",
"revision": "2a247be0c0aec895ccb0ac7ca92229ec6df37288"
},
"scripts": {
"flint_quantize.py": "3f87c9cf1bed5a2346ebc085579a82e14a236325b0b2277171dfae1d07d7b692",
"flint_reference.py": "2534d51af8e99699692f1a2cf5fc022bd7beaa7de07770dbffb9690aa4f62ff6",
"flint_nvfp4.py": "31aec4a310b7bddaa90da083b80352e5607ce754c10052458859f7306b5cc566"
},
"packages": {
"torch": "2.8.0+cu128",
"triton": "3.4.0",
"safetensors": "0.8.0",
"transformers": "5.20.2",
"huggingface_hub": "2.2.0",
"datasets": "5.1.0"
},
"gpu": "NVIDIA RTX PRO 6000 Blackwell Server Edition",
"cuda": "12.8",
"created_utc": "2026-10-22T00:06:38Z"
},
"results": {
"bf16": {
"aggregate": {
"all": {
"tokens": 49140,
"nll": 2.214729754197971,
"teacher_nll": 2.214729754197971,
"kl": 0.0,
"top1_agree": 1.0,
"perplexity": 9.158933613767237,
"teacher_perplexity": 9.158933613767237,
"delta_nll": 0.0
},
"code": {
"tokens": 16380,
"nll": 1.585819314280127,
"teacher_nll": 1.585819314280127,
"kl": 0.0,
"top1_agree": 1.0,
"perplexity": 4.883290689738726,
"teacher_perplexity": 4.883290689738726,
"delta_nll": 0.0
},
"math": {
"tokens": 16380,
"nll": 1.5066873193514945,
"teacher_nll": 1.5066873193514945,
"kl": 0.0,
"top1_agree": 1.0,
"perplexity": 4.511759991246355,
"teacher_perplexity": 4.511759991246355,
"delta_nll": 0.0
},
"prose": {
"tokens": 16380,
"nll": 3.551682628962292,
"teacher_nll": 3.551682628962292,
"kl": 0.0,
"top1_agree": 1.0,
"perplexity": 34.871944693965524,
"teacher_perplexity": 34.871944693965524,
"delta_nll": 0.0
}
},
"sequences": [
{
"id": "prose-qualification-holdout-0",
"domain": "prose",
"tokens": 4095,
"nll": 3.570623514736674,
"teacher_nll": 3.570623514736674,
"kl": 0.0,
"top1_agree": 1.0
},
{
"id": "prose-qualification-holdout-1",
"domain": "prose",
"tokens": 4095,
"nll": 3.4886688269683876,
"teacher_nll": 3.4886688269683876,
"kl": 0.0,
"top1_agree": 1.0
},
{
"id": "prose-qualification-holdout-2",
"domain": "prose",
"tokens": 4095,
"nll": 3.586854711586157,
"teacher_nll": 3.586854711586157,
"kl": 0.0,
"top1_agree": 1.0
},
{
"id": "prose-qualification-holdout-3",
"domain": "prose",
"tokens": 4095,
"nll": 3.56058346255795,
"teacher_nll": 3.56058346255795,
"kl": 0.0,
"top1_agree": 1.0
},
{
"id": "code-qualification-holdout-0",
"domain": "code",
"tokens": 4095,
"nll": 1.612040780315469,
"teacher_nll": 1.612040780315469,
"kl": 0.0,
"top1_agree": 1.0
},
{
"id": "code-qualification-holdout-1",
"domain": "code",
"tokens": 4095,
"nll": 1.4962344744848826,
"teacher_nll": 1.4962344744848826,
"kl": 0.0,
"top1_agree": 1.0
},
{
"id": "code-qualification-holdout-2",
"domain": "code",
"tokens": 4095,
"nll": 1.4386393615759918,
"teacher_nll": 1.4386393615759918,
"kl": 0.0,
"top1_agree": 1.0
},
{
"id": "code-qualification-holdout-3",
"domain": "code",
"tokens": 4095,
"nll": 1.796362640744164,
"teacher_nll": 1.796362640744164,
"kl": 0.0,
"top1_agree": 1.0
},
{
"id": "math-qualification-holdout-0",
"domain": "math",
"tokens": 4095,
"nll": 1.4128543704711765,
"teacher_nll": 1.4128543704711765,
"kl": 0.0,
"top1_agree": 1.0
},
{
"id": "math-qualification-holdout-1",
"domain": "math",
"tokens": 4095,
"nll": 1.545063196724879,
"teacher_nll": 1.545063196724879,
"kl": 0.0,
"top1_agree": 1.0
},
{
"id": "math-qualification-holdout-2",
"domain": "math",
"tokens": 4095,
"nll": 1.5522358872223831,
"teacher_nll": 1.5522358872223831,
"kl": 0.0,
"top1_agree": 1.0
},
{
"id": "math-qualification-holdout-3",
"domain": "math",
"tokens": 4095,
"nll": 1.516595822987539,
"teacher_nll": 1.516595822987539,
"kl": 0.0,
"top1_agree": 1.0
}
],
"memory": {
"swap_kib": 0,
"cuda_peak_bytes": 6379677184
}
},
"budget36-swap1": {
"aggregate": {
"all": {
"tokens": 49140,
"nll": 2.219416871241161,
"teacher_nll": 2.214729754197971,
"kl": 0.09010865478740386,
"top1_agree": 0.9188848188848189,
"perplexity": 9.201963371584448,
"teacher_perplexity": 9.158933613767237,
"delta_nll": 0.0046871170431899145
},
"code": {
"tokens": 16380,
"nll": 1.5947170897863432,
"teacher_nll": 1.585819314280127,
"kl": 0.10190384458427537,
"top1_agree": 0.9322344322344323,
"perplexity": 4.92693499469703,
"teacher_perplexity": 4.883290689738726,
"delta_nll": 0.008897775506216288
},
"math": {
"tokens": 16380,
"nll": 1.5039469321350476,
"teacher_nll": 1.5066873193514945,
"kl": 0.03564856964480746,
"top1_agree": 0.9473748473748473,
"perplexity": 4.499412947410116,
"teacher_perplexity": 4.511759991246355,
"delta_nll": -0.0027403872164468712
},
"prose": {
"tokens": 16380,
"nll": 3.5595865918020917,
"teacher_nll": 3.551682628962292,
"kl": 0.13277355013312875,
"top1_agree": 0.877045177045177,
"perplexity": 35.14866339553476,
"teacher_perplexity": 34.871944693965524,
"delta_nll": 0.00790396283979966
}
},
"sequences": [
{
"id": "prose-qualification-holdout-0",
"domain": "prose",
"tokens": 4095,
"nll": 3.60210935525114,
"teacher_nll": 3.570623514736674,
"kl": 0.1251914977445125,
"top1_agree": 0.873992673992674
},
{
"id": "prose-qualification-holdout-1",
"domain": "prose",
"tokens": 4095,
"nll": 3.4887589190323567,
"teacher_nll": 3.4886688269683876,
"kl": 0.1547056760543432,
"top1_agree": 0.8717948717948718
},
{
"id": "prose-qualification-holdout-2",
"domain": "prose",
"tokens": 4095,
"nll": 3.587463203774849,
"teacher_nll": 3.586854711586157,
"kl": 0.11346375461899753,
"top1_agree": 0.8842490842490842
},
{
"id": "prose-qualification-holdout-3",
"domain": "prose",
"tokens": 4095,
"nll": 3.560014889150021,
"teacher_nll": 3.56058346255795,
"kl": 0.13773327211466174,
"top1_agree": 0.8781440781440781
},
{
"id": "code-qualification-holdout-0",
"domain": "code",
"tokens": 4095,
"nll": 1.634640592358488,
"teacher_nll": 1.612040780315469,
"kl": 0.11310800134917319,
"top1_agree": 0.9255189255189256
},
{
"id": "code-qualification-holdout-1",
"domain": "code",
"tokens": 4095,
"nll": 1.4733869531070212,
"teacher_nll": 1.4962344744848826,
"kl": 0.10195107398343174,
"top1_agree": 0.9423687423687424
},
{
"id": "code-qualification-holdout-2",
"domain": "code",
"tokens": 4095,
"nll": 1.4622409024081386,
"teacher_nll": 1.4386393615759918,
"kl": 0.0776867737409105,
"top1_agree": 0.9374847374847375
},
{
"id": "code-qualification-holdout-3",
"domain": "code",
"tokens": 4095,
"nll": 1.8085999112717253,
"teacher_nll": 1.796362640744164,
"kl": 0.11486952926358605,
"top1_agree": 0.9235653235653236
},
{
"id": "math-qualification-holdout-0",
"domain": "math",
"tokens": 4095,
"nll": 1.4042997626012352,
"teacher_nll": 1.4128543704711765,
"kl": 0.03463405605419215,
"top1_agree": 0.9472527472527472
},
{
"id": "math-qualification-holdout-1",
"domain": "math",
"tokens": 4095,
"nll": 1.5480228545173766,
"teacher_nll": 1.545063196724879,
"kl": 0.03576130380123963,
"top1_agree": 0.9457875457875458
},
{
"id": "math-qualification-holdout-2",
"domain": "math",
"tokens": 4095,
"nll": 1.5291209340823293,
"teacher_nll": 1.5522358872223831,
"kl": 0.03457320837965815,
"top1_agree": 0.9474969474969475
},
{
"id": "math-qualification-holdout-3",
"domain": "math",
"tokens": 4095,
"nll": 1.5343441773392488,
"teacher_nll": 1.516595822987539,
"kl": 0.03762571034413991,
"top1_agree": 0.9489621489621489
}
],
"memory": {
"swap_kib": 0,
"cuda_peak_bytes": 6379677184
}
}
},
"memory": {
"swap_kib": 0,
"cuda_peak_bytes": 6379677184
}
}