wtf.petrichor.drinkme.measurement
Samples
52 randomly sampled records from the AT Protocol firehose
wtf.petrichor.drinkme.measurement (52 samples)
{
"$type": "wtf.petrichor.drinkme.measurement",
"compression": {
"bitsPerWeight": "11.25",
"meanTensorBitsPerWeight": "11.264",
"profile": "sip"
},
"createdAt": "2026-09-30T01:15:00.714139+00:00",
"environment": {
"cpuInfo": "unknown",
"deviceClass": "NVIDIA H100 80GB HBM3",
"engine": "drinkme 1.0.0 / torch 2.13.0+cu130 / deltanet fla 0.5.2 / conv torch / narrow-gemv triton / stock-gemv linear / raw-gemv stock / decode cudagraph",
"gpuInfo": "NVIDIA H100 80GB HBM3, 81559",
"memoryBytes": 85520809984,
"memoryKind": "vram",
"os": "Debian GNU/Linux 12 (bookworm)",
"platform": "cuda"
},
"metrics": [
{
"name": "stock_decode_tok_s",
"samples": [
"35.35",
"35.41",
"35.39"
],
"unit": "tok/s",
"value": "35.39"
},
{
"name": "stock_prefill_tok_s",
"samples": [
"5535.59",
"5647.33",
"5690.9"
],
"unit": "tok/s",
"value": "5647.33"
},
{
"name": "stock_ttft_s",
"samples": [
"0.154",
"0.0923",
"0.0919"
],
"unit": "s",
"value": "0.0923"
},
{
"name": "stock_weights_gb",
"unit": "GB",
"value": "54.729"
},
{
"name": "stock_decode_read_gb",
"unit": "GB",
"value": "51.244"
},
{
"name": "twin_decode_tok_s",
"samples": [
"37.01",
"37.4",
"37.28"
],
"unit": "tok/s",
"value": "37.28"
},
{
"name": "twin_weights_gb",
"unit": "GB",
"value": "54.75"
},
{
"name": "twin_decode_read_gb",
"unit": "GB",
"value": "51.244"
},
{
"name": "compressed_decode_tok_s",
"samples": [
"24.05",
"24.06",
"24.06"
],
"unit": "tok/s",
"value": "24.06"
},
{
"name": "compressed_prefill_tok_s",
"samples": [
"3794.3",
"3797.54",
"3780.14"
],
"unit": "tok/s",
"value": "3794.3"
},
{
"name": "compressed_ttft_s",
"samples": [
"0.1591",
"0.1415",
"0.1399"
],
"unit": "s",
"value": "0.1415"
},
{
"name": "read_gb_s",
"unit": "GB/s",
"value": "3089.312"
},
{
"name": "copy_gb_s",
"unit": "GB/s",
"value": "3025.444"
},
{
"name": "compressed_weights_gb",
"unit": "GB",
"value": "39.406"
},
{
"name": "compressed_decode_read_gb",
"unit": "GB",
"value": "36.044"
},
{
"name": "stock_spec_decode_tok_s_agent",
"samples": [
"71.02",
"70.96",
"70.92"
],
"unit": "tok/s",
"value": "70.96"
},
{
"name": "stock_spec_decode_tok_s_chat",
"samples": [
"80.67",
"80.8",
"80.6"
],
"unit": "tok/s",
"value": "80.67"
},
{
"name": "stock_spec_decode_tok_s_code",
"samples": [
"97.64",
"97.58",
"97.65"
],
"unit": "tok/s",
"value": "97.64"
},
{
"name": "compressed_spec_decode_tok_s_agent",
"samples": [
"46.86",
"46.14",
"46.72"
],
"unit": "tok/s",
"value": "46.72"
},
{
"name": "compressed_spec_decode_tok_s_chat",
"samples": [
"51.64",
"51.57",
"51.83"
],
"unit": "tok/s",
"value": "51.64"
},
{
"name": "compressed_spec_decode_tok_s_code",
"samples": [
"63.1",
"63.13",
"63.26"
],
"unit": "tok/s",
"value": "63.13"
}
],
"model": {
"hfRepo": "Qwen/Qwen3.8-27B",
"name": "Qwen3.8-27B",
"revision": "1d4bf0f2ff6012fd82039f2fa52739d0dd7c60c0"
},
"raw": {
"bandwidth": {
"ceiling_tok_s_bf16": "56.46",
"copy_bytes_s": 3025443546594,
"device": "cuda",
"gpu_count": 1,
"gpus": [
"NVIDIA H100 80GB HBM3"
],
"probe_bytes": 4294967296,
"read_bytes_s": 3089312407306,
"total_param_bytes_bf16": 54713457120
},
"compressed_bytes_per_token": {
"counts": {
"bf16": 96,
"drinkme_codec": 401
},
"embedding_row_bytes": 10240,
"packed_bytes": 35996836384,
"raw_linear_bytes": 47185920,
"skipped": [
"model.visual"
],
"total_bytes": 36044032544
},
"compressed_decode_samples": [
"24.05",
"24.06",
"24.06"
],
"compressed_decode_step": "cudagraph",
"compressed_decode_tok_s": "24.06",
"compressed_linear_kinds": {
"RadixCompressedLinear": 511
},
"compressed_prefill_samples": [
"3794.3",
"3797.54",
"3780.14"
],
"compressed_prefill_tok_s": "3794.3",
"compressed_routing": {
"deltanet_conv": "torch",
"deltanet_kernel": "fla",
"narrow_count": 96,
"narrow_gemv": true,
"raw_gemv": "stock",
"stock_gemv": "linear"
},
"compressed_ttft_s": "0.1415",
"compressed_ttft_samples": [
"0.1591",
"0.1415",
"0.1399"
],
"compression_profile": "sip",
"detector": {
"budgetBytes": 85520809984,
"evidence": [
"nvidia-smi: NVIDIA H100 80GB HBM3, 81559"
],
"heuristic": false,
"version": 2
},
"gpu": "NVIDIA H100 80GB HBM3",
"host_load": {
"cpu_count": 32,
"load1": {
"compressed": [
"0.0",
"0.0"
],
"stock": [
"0.0",
"0.0"
],
"twin": [
"0.0",
"0.0"
]
}
},
"mean_bpw": "11.264",
"model": "Qwen/Qwen3.8-27B",
"prefill_prompt_len": 512,
"resolved_revision": "1d4bf0f2ff6012fd82039f2fa52739d0dd7c60c0",
"revision": "1d4bf0f2ff6012fd82039f2fa52739d0dd7c60c0",
"spec": {
"arms": {
"compressed": {
"agent": {
"accepted_per_step": "1.898",
"samples": [
"46.86",
"46.14",
"46.72"
]
},
"chat": {
"accepted_per_step": "2.188",
"samples": [
"51.64",
"51.57",
"51.83"
]
},
"code": {
"accepted_per_step": "2.923",
"samples": [
"63.1",
"63.13",
"63.26"
]
}
},
"stock": {
"agent": {
"accepted_per_step": "1.898",
"samples": [
"71.02",
"70.96",
"70.92"
]
},
"chat": {
"accepted_per_step": "2.228",
"samples": [
"80.67",
"80.8",
"80.6"
]
},
"code": {
"accepted_per_step": "2.923",
"samples": [
"97.64",
"97.58",
"97.65"
]
}
}
},
"ctx": 8192,
"mode": "mtp",
"prompts": {
"agent": {
"sha256": "9adaaba9c824271e813cf8817a3cde0823867d727ef2cd2c6951fd3f8c279c8e",
"tokens": 1434
},
"chat": {
"sha256": "694da0dd4dcfc7319974ade7e4456d4f1fd21ffcd788d9632ab32c052b91bc61",
"tokens": 27
},
"code": {
"sha256": "047bc057054850f650f43790dfba9bf08f08fe4739ae9ec962b91a5aa5753cdd",
"tokens": 42
}
},
"reps": 3,
"tokens": 256
},
"stock_bytes_per_token": {
"counts": {
"bf16": 497
},
"embedding_row_bytes": 10240,
"packed_bytes": 0,
"raw_linear_bytes": 51243909120,
"skipped": [
"model.visual"
],
"total_bytes": 51243919360
},
"stock_decode_samples": [
"35.35",
"35.41",
"35.39"
],
"stock_decode_step": "cudagraph",
"stock_decode_tok_s": "35.39",
"stock_loader": "stream",
"stock_outcome": "measured",
"stock_prefill_samples": [
"5535.59",
"5647.33",
"5690.9"
],
"stock_prefill_tok_s": "5647.33",
"stock_routing": {
"deltanet_conv": "torch",
"deltanet_kernel": "fla",
"narrow_count": 96,
"narrow_gemv": true,
"raw_gemv": "stock",
"stock_gemv": "linear"
},
"stock_ttft_s": "0.0923",
"stock_ttft_samples": [
"0.154",
"0.0923",
"0.0919"
],
"swapped_linears": 511,
"twin_bytes_per_token": {
"counts": {
"bf16": 96,
"drinkme_twin": 401
},
"embedding_row_bytes": 10240,
"packed_bytes": 51196723200,
"raw_linear_bytes": 47185920,
"skipped": [
"model.visual"
],
"total_bytes": 51243919360
},
"twin_decode_samples": [
"37.01",
"37.4",
"37.28"
],
"twin_decode_step": "cudagraph",
"twin_decode_tok_s": "37.28",
"twin_outcome": "measured",
"twin_routing": {
"deltanet_conv": "torch",
"deltanet_kernel": "fla",
"narrow_count": 96,
"narrow_gemv": true,
"raw_gemv": "stock",
"stock_gemv": "linear"
},
"verification": "roundtrip",
"verified_tensors": 511,
"vision_tower_included": true,
"vram_bf16_bytes": 54728546304,
"vram_compressed_bytes": 39405735936,
"vram_twin_bytes": 54750267392,
"warmup_rep": true,
"weighted_bpw": "11.25"
},
"stock": {
"outcome": "measured"
},
"version": "1.0.0"
}
did:plc:k7th57d6hylncxd3qu47yfi3 | at://did:plc:k7th57d6hylncxd3qu47yfi3/wtf.petrichor.drinkme.measurement/3mwuqpatu2225