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": "10.89",
"meanTensorBitsPerWeight": "10.884",
"packedBytes": 4946199508,
"profile": "gulp",
"residentBytes": 5733001536
},
"createdAt": "2026-10-01T07:39:14.519600+00:00",
"environment": {
"cpuInfo": "AMD RYZEN AI MAX+ 395 w/ Radeon 8060S",
"deviceClass": "AMD RYZEN AI MAX+ 395 w/ Radeon 8060S",
"engine": "drinkme 1.0.0 / torch 2.12.0a0+rocm7.13.0a20260411 / deltanet none / conv none / narrow-gemv triton / stock-gemv triton / raw-gemv twin / decode eager",
"gpuInfo": "amdgpu 0x1586 vram=0.54GB gtt=133.14GB",
"memoryBytes": 133680857088,
"memoryKind": "unified",
"os": "Ubuntu 26.04.1 LTS",
"platform": "rocm"
},
"metrics": [
{
"name": "stock_decode_tok_s",
"samples": [
"24.53",
"24.49",
"24.49"
],
"unit": "tok/s",
"value": "24.49"
},
{
"name": "stock_prefill_tok_s",
"samples": [
"3088.1",
"3104.29",
"3113.98"
],
"unit": "tok/s",
"value": "3104.29"
},
{
"name": "stock_ttft_s",
"samples": [
"0.1727",
"0.171",
"0.1714"
],
"unit": "s",
"value": "0.1714"
},
{
"name": "stock_weights_gb",
"unit": "GB",
"value": "8.102"
},
{
"name": "stock_decode_read_gb",
"unit": "GB",
"value": "8.045"
},
{
"name": "twin_decode_tok_s",
"samples": [
"24.49",
"24.49",
"24.5"
],
"unit": "tok/s",
"value": "24.49"
},
{
"name": "twin_weights_gb",
"unit": "GB",
"value": "8.135"
},
{
"name": "twin_decode_read_gb",
"unit": "GB",
"value": "8.045"
},
{
"name": "compressed_decode_tok_s",
"samples": [
"17.82",
"17.8",
"17.8"
],
"unit": "tok/s",
"value": "17.8"
},
{
"name": "compressed_prefill_tok_s",
"samples": [
"2354.51",
"2345.47",
"2352.14"
],
"unit": "tok/s",
"value": "2352.14"
},
{
"name": "compressed_ttft_s",
"samples": [
"0.2253",
"0.224",
"0.2237"
],
"unit": "s",
"value": "0.224"
},
{
"name": "read_gb_s",
"unit": "GB/s",
"value": "239.807"
},
{
"name": "copy_gb_s",
"unit": "GB/s",
"value": "207.427"
},
{
"name": "compressed_weights_gb",
"unit": "GB",
"value": "5.8"
},
{
"name": "compressed_decode_read_gb",
"unit": "GB",
"value": "5.733"
}
],
"model": {
"hfRepo": "Qwen/Qwen3-4B",
"name": "Qwen3-4B",
"revision": "1cfa9a7208912126459214e8b04321603b3df60c"
},
"raw": {
"bandwidth": {
"ceiling_tok_s_bf16": "29.81",
"copy_bytes_s": 207427075802,
"device": "cuda",
"gpu_count": 1,
"gpus": [
"Radeon 8060S Graphics"
],
"probe_bytes": 4294967296,
"read_bytes_s": 239807009940,
"total_param_bytes_bf16": 8044936192
},
"compressed_bytes_per_token": {
"counts": {
"bf16": 1,
"drinkme_codec": 252
},
"embedding_row_bytes": 5120,
"packed_bytes": 4954697024,
"raw_linear_bytes": 777912320,
"total_bytes": 5732614464
},
"compressed_decode_samples": [
"17.82",
"17.8",
"17.8"
],
"compressed_decode_step": "eager",
"compressed_decode_tok_s": "17.8",
"compressed_linear_kinds": {
"RadixCompressedLinear": 252
},
"compressed_prefill_samples": [
"2354.51",
"2345.47",
"2352.14"
],
"compressed_prefill_tok_s": "2352.14",
"compressed_routing": {
"deltanet_conv": "none",
"deltanet_kernel": "none",
"narrow_count": 0,
"narrow_gemv": true,
"raw_gemv": "twin",
"stock_gemv": "triton"
},
"compressed_ttft_s": "0.224",
"compressed_ttft_samples": [
"0.2253",
"0.224",
"0.2237"
],
"compression_profile": "gulp",
"detector": {
"budgetBytes": 133143986176,
"evidence": [
"vram_total=0.54GB",
"gtt_total=133.14GB",
"gtt >> vram carveout -> unified APU"
],
"heuristic": true,
"version": 2
},
"device_source": "cpu-brand",
"gpu": "Radeon 8060S Graphics",
"host_load": {
"cpu_count": 32,
"load1": {
"compressed": [
"1.56",
"1.66"
],
"stock": [
"1.59",
"1.61"
],
"twin": [
"1.34",
"1.52"
]
}
},
"mean_bpw": "10.884",
"model": "Qwen/Qwen3-4B",
"pack": {
"dtype": "bf16",
"fallback_tensors": 0,
"formatVersion": 1,
"meanBpw": "10.884",
"packedBytes": 4946199508,
"profile": "gulp",
"profileWidths": [
2,
2,
4,
8
],
"radixBytes": 4946199508,
"radixTensorCount": 252,
"rawFallbackBytes": 0,
"rawFallbackTensorCount": 0,
"rawResidentBytes": 778304512,
"rawTensorCount": 146,
"residentBytes": 5733001536,
"tensorCount": 252,
"weightedBpw": "10.89"
},
"prefill_prompt_len": 512,
"resolved_revision": "1cfa9a7208912126459214e8b04321603b3df60c",
"revision": "1cfa9a7208912126459214e8b04321603b3df60c",
"spec": {
"skipped": "no MTP head"
},
"stock_bytes_per_token": {
"counts": {
"bf16": 253
},
"embedding_row_bytes": 5120,
"packed_bytes": 0,
"raw_linear_bytes": 8044544000,
"total_bytes": 8044549120
},
"stock_decode_samples": [
"24.53",
"24.49",
"24.49"
],
"stock_decode_step": "eager",
"stock_decode_tok_s": "24.49",
"stock_loader": "stream",
"stock_outcome": "measured",
"stock_prefill_samples": [
"3088.1",
"3104.29",
"3113.98"
],
"stock_prefill_tok_s": "3104.29",
"stock_routing": {
"deltanet_conv": "none",
"deltanet_kernel": "none",
"narrow_count": 0,
"narrow_gemv": true,
"raw_gemv": "twin",
"stock_gemv": "triton"
},
"stock_ttft_s": "0.1714",
"stock_ttft_samples": [
"0.1727",
"0.171",
"0.1714"
],
"swapped_linears": 252,
"twin_bytes_per_token": {
"counts": {
"bf16": 1,
"drinkme_twin": 252
},
"embedding_row_bytes": 5120,
"packed_bytes": 7266631680,
"raw_linear_bytes": 777912320,
"total_bytes": 8044549120
},
"twin_decode_samples": [
"24.49",
"24.49",
"24.5"
],
"twin_decode_step": "eager",
"twin_decode_tok_s": "24.49",
"twin_outcome": "measured",
"twin_routing": {
"deltanet_conv": "none",
"deltanet_kernel": "none",
"narrow_count": 0,
"narrow_gemv": true,
"raw_gemv": "twin",
"stock_gemv": "triton"
},
"verification": "pack",
"verified_tensors": 252,
"vision_tower_included": false,
"vram_bf16_bytes": 8101714432,
"vram_compressed_bytes": 5800153600,
"vram_twin_bytes": 8135268864,
"warmup_rep": true,
"weighted_bpw": "10.89"
},
"stock": {
"outcome": "measured"
},
"version": "1.0.0"
}did:plc:k7th57d6hylncxd3qu47yfi3 | at://did:plc:k7th57d6hylncxd3qu47yfi3/wtf.petrichor.drinkme.measurement/3mwuqoxf5ed2v