{ "status": "untrained_runtime_probe_not_quality_or_calibration_evidence", "model": "models/gemma-4-e2b-it-4bit", "state_layers": 4, "compute_dtype": "float16", "head_config": { "input_dims": 1536, "dims": 256, "heads": 4, "query_layers": 1, "decision_layers": 2 }, "timing_scope": "warm_model_fresh_state_including_media_decode_preprocess_encoding_and_all_fields", "field_scaling": "four representative question templates repeated to the requested count", "results": [ { "state": "text", "fields": 1, "attempted": 5, "completed": 5, "samples": [ { "execution": "shared_state_parallel_decision_head", "state_encodes": 1, "state_tokens": 50, "state_layers": 4, "compute_dtype": "float16", "field_count": 1, "candidate_count": 3, "query_tokens": 59, "padded_query_tokens": 60, "state_memory_batch_size": 1, "preprocess_seconds": 0.0003582499921321869, "state_encode_seconds": 0.009155500010820106, "query_embed_seconds": 0.0003761249827221036, "trained": false, "head_seconds": 0.005064208002295345, "total_seconds": 0.014962124987505376, "status": "runtime_probe_only" }, { "execution": "shared_state_parallel_decision_head", "state_encodes": 1, "state_tokens": 50, "state_layers": 4, "compute_dtype": "float16", "field_count": 1, "candidate_count": 3, "query_tokens": 59, "padded_query_tokens": 60, "state_memory_batch_size": 1, "preprocess_seconds": 0.00029191598878242075, "state_encode_seconds": 0.010532542015425861, "query_embed_seconds": 0.002476582973031327, "trained": false, "head_seconds": 0.0009858330013230443, "total_seconds": 0.014297916000941768, "status": "runtime_probe_only" }, { "execution": "shared_state_parallel_decision_head", "state_encodes": 1, "state_tokens": 50, "state_layers": 4, "compute_dtype": "float16", "field_count": 1, "candidate_count": 3, "query_tokens": 59, "padded_query_tokens": 60, "state_memory_batch_size": 1, "preprocess_seconds": 0.00030654200236313045, "state_encode_seconds": 0.00897304099635221, "query_embed_seconds": 0.0003955420106649399, "trained": false, "head_seconds": 0.000890791998244822, "total_seconds": 0.010574249987257645, "status": "runtime_probe_only" }, { "execution": "shared_state_parallel_decision_head", "state_encodes": 1, "state_tokens": 50, "state_layers": 4, "compute_dtype": "float16", "field_count": 1, "candidate_count": 3, "query_tokens": 59, "padded_query_tokens": 60, "state_memory_batch_size": 1, "preprocess_seconds": 0.0002838750078808516, "state_encode_seconds": 0.013059666991466656, "query_embed_seconds": 0.000626958004431799, "trained": false, "head_seconds": 0.0010681250132620335, "total_seconds": 0.015045083011500537, "status": "runtime_probe_only" }, { "execution": "shared_state_parallel_decision_head", "state_encodes": 1, "state_tokens": 50, "state_layers": 4, "compute_dtype": "float16", "field_count": 1, "candidate_count": 3, "query_tokens": 59, "padded_query_tokens": 60, "state_memory_batch_size": 1, "preprocess_seconds": 0.0004568339791148901, "state_encode_seconds": 0.009174833016004413, "query_embed_seconds": 0.00041737500578165054, "trained": false, "head_seconds": 0.005109791993163526, "total_seconds": 0.015165832999628037, "status": "runtime_probe_only" } ], "median_seconds": { "preprocess_seconds": 0.00030654200236313045, "state_encode_seconds": 0.009174833016004413, "query_embed_seconds": 0.00041737500578165054, "head_seconds": 0.0010681250132620335, "total_seconds": 0.014962124987505376 }, "max_seconds": 0.015165832999628037, "under_300ms": 5 }, { "state": "text", "fields": 4, "attempted": 5, "completed": 5, "samples": [ { "execution": "shared_state_parallel_decision_head", "state_encodes": 1, "state_tokens": 50, "state_layers": 4, "compute_dtype": "float16", "field_count": 4, "candidate_count": 10, "query_tokens": 184, "padded_query_tokens": 200, "state_memory_batch_size": 1, "preprocess_seconds": 0.0003633340238593519, "state_encode_seconds": 0.009137249988270923, "query_embed_seconds": 0.0005036249931436032, "trained": false, "head_seconds": 0.0009447919728700072, "total_seconds": 0.010959916980937123, "status": "runtime_probe_only" }, { "execution": "shared_state_parallel_decision_head", "state_encodes": 1, "state_tokens": 50, "state_layers": 4, "compute_dtype": "float16", "field_count": 4, "candidate_count": 10, "query_tokens": 184, "padded_query_tokens": 200, "state_memory_batch_size": 1, "preprocess_seconds": 0.0003382920112926513, "state_encode_seconds": 0.009165499999653548, "query_embed_seconds": 0.0004992079921066761, "trained": false, "head_seconds": 0.0009395410015713423, "total_seconds": 0.010956999991321936, "status": "runtime_probe_only" }, { "execution": "shared_state_parallel_decision_head", "state_encodes": 1, "state_tokens": 50, "state_layers": 4, "compute_dtype": "float16", "field_count": 4, "candidate_count": 10, "query_tokens": 184, "padded_query_tokens": 200, "state_memory_batch_size": 1, "preprocess_seconds": 0.00029712499235756695, "state_encode_seconds": 0.009060875017894432, "query_embed_seconds": 0.002391207992332056, "trained": false, "head_seconds": 0.002695167000638321, "total_seconds": 0.014458124991506338, "status": "runtime_probe_only" }, { "execution": "shared_state_parallel_decision_head", "state_encodes": 1, "state_tokens": 50, "state_layers": 4, "compute_dtype": "float16", "field_count": 4, "candidate_count": 10, "query_tokens": 184, "padded_query_tokens": 200, "state_memory_batch_size": 1, "preprocess_seconds": 0.00033212502603419125, "state_encode_seconds": 0.009143332979874685, "query_embed_seconds": 0.000518000015290454, "trained": false, "head_seconds": 0.0008980410057120025, "total_seconds": 0.01089912498719059, "status": "runtime_probe_only" }, { "execution": "shared_state_parallel_decision_head", "state_encodes": 1, "state_tokens": 50, "state_layers": 4, "compute_dtype": "float16", "field_count": 4, "candidate_count": 10, "query_tokens": 184, "padded_query_tokens": 200, "state_memory_batch_size": 1, "preprocess_seconds": 0.00032458300120197237, "state_encode_seconds": 0.009185041999444366, "query_embed_seconds": 0.0005267079977784306, "trained": false, "head_seconds": 0.0009112079860642552, "total_seconds": 0.010954792000120506, "status": "runtime_probe_only" } ], "median_seconds": { "preprocess_seconds": 0.00033212502603419125, "state_encode_seconds": 0.009143332979874685, "query_embed_seconds": 0.000518000015290454, "head_seconds": 0.0009395410015713423, "total_seconds": 0.010956999991321936 }, "max_seconds": 0.014458124991506338, "under_300ms": 5 }, { "state": "text", "fields": 16, "attempted": 5, "completed": 5, "samples": [ { "execution": "shared_state_parallel_decision_head", "state_encodes": 1, "state_tokens": 50, "state_layers": 4, "compute_dtype": "float16", "field_count": 16, "candidate_count": 40, "query_tokens": 736, "padded_query_tokens": 800, "state_memory_batch_size": 1, "preprocess_seconds": 0.00028337500407360494, "state_encode_seconds": 0.008968207985162735, "query_embed_seconds": 0.0045255419972818345, "trained": false, "head_seconds": 0.001288625004235655, "total_seconds": 0.015080208977451548, "status": "runtime_probe_only" }, { "execution": "shared_state_parallel_decision_head", "state_encodes": 1, "state_tokens": 50, "state_layers": 4, "compute_dtype": "float16", "field_count": 16, "candidate_count": 40, "query_tokens": 736, "padded_query_tokens": 800, "state_memory_batch_size": 1, "preprocess_seconds": 0.00033658300526440144, "state_encode_seconds": 0.009252166986698285, "query_embed_seconds": 0.0010412500123493373, "trained": false, "head_seconds": 0.00546983300591819, "total_seconds": 0.016106625000247732, "status": "runtime_probe_only" }, { "execution": "shared_state_parallel_decision_head", "state_encodes": 1, "state_tokens": 50, "state_layers": 4, "compute_dtype": "float16", "field_count": 16, "candidate_count": 40, "query_tokens": 736, "padded_query_tokens": 800, "state_memory_batch_size": 1, "preprocess_seconds": 0.000327582994941622, "state_encode_seconds": 0.009130791993811727, "query_embed_seconds": 0.0009590410045348108, "trained": false, "head_seconds": 0.0036415409995242953, "total_seconds": 0.014067874988541007, "status": "runtime_probe_only" }, { "execution": "shared_state_parallel_decision_head", "state_encodes": 1, "state_tokens": 50, "state_layers": 4, "compute_dtype": "float16", "field_count": 16, "candidate_count": 40, "query_tokens": 736, "padded_query_tokens": 800, "state_memory_batch_size": 1, "preprocess_seconds": 0.0003196659963577986, "state_encode_seconds": 0.009104249998927116, "query_embed_seconds": 0.0008916250080801547, "trained": false, "head_seconds": 0.005565125000430271, "total_seconds": 0.01589004200650379, "status": "runtime_probe_only" }, { "execution": "shared_state_parallel_decision_head", "state_encodes": 1, "state_tokens": 50, "state_layers": 4, "compute_dtype": "float16", "field_count": 16, "candidate_count": 40, "query_tokens": 736, "padded_query_tokens": 800, "state_memory_batch_size": 1, "preprocess_seconds": 0.0003958329907618463, "state_encode_seconds": 0.01121483399765566, "query_embed_seconds": 0.0021067500056233257, "trained": false, "head_seconds": 0.0011826670088339597, "total_seconds": 0.014906291995430365, "status": "runtime_probe_only" } ], "median_seconds": { "preprocess_seconds": 0.000327582994941622, "state_encode_seconds": 0.009130791993811727, "query_embed_seconds": 0.0010412500123493373, "head_seconds": 0.0036415409995242953, "total_seconds": 0.015080208977451548 }, "max_seconds": 0.016106625000247732, "under_300ms": 5 }, { "state": "text", "fields": 28, "attempted": 5, "completed": 5, "samples": [ { "execution": "shared_state_parallel_decision_head", "state_encodes": 1, "state_tokens": 50, "state_layers": 4, "compute_dtype": "float16", "field_count": 28, "candidate_count": 70, "query_tokens": 1288, "padded_query_tokens": 1400, "state_memory_batch_size": 1, "preprocess_seconds": 0.0007404169882647693, "state_encode_seconds": 0.009168874996248633, "query_embed_seconds": 0.005016290990170091, "trained": false, "head_seconds": 0.001556792005430907, "total_seconds": 0.016499207995366305, "status": "runtime_probe_only" }, { "execution": "shared_state_parallel_decision_head", "state_encodes": 1, "state_tokens": 50, "state_layers": 4, "compute_dtype": "float16", "field_count": 28, "candidate_count": 70, "query_tokens": 1288, "padded_query_tokens": 1400, "state_memory_batch_size": 1, "preprocess_seconds": 0.00032904199906624854, "state_encode_seconds": 0.009080457995878533, "query_embed_seconds": 0.0013523329980671406, "trained": false, "head_seconds": 0.00548516699927859, "total_seconds": 0.01626124998438172, "status": "runtime_probe_only" }, { "execution": "shared_state_parallel_decision_head", "state_encodes": 1, "state_tokens": 50, "state_layers": 4, "compute_dtype": "float16", "field_count": 28, "candidate_count": 70, "query_tokens": 1288, "padded_query_tokens": 1400, "state_memory_batch_size": 1, "preprocess_seconds": 0.0003387090109754354, "state_encode_seconds": 0.008824874996207654, "query_embed_seconds": 0.0013330410001799464, "trained": false, "head_seconds": 0.0015479580033570528, "total_seconds": 0.0120523750083521, "status": "runtime_probe_only" }, { "execution": "shared_state_parallel_decision_head", "state_encodes": 1, "state_tokens": 50, "state_layers": 4, "compute_dtype": "float16", "field_count": 28, "candidate_count": 70, "query_tokens": 1288, "padded_query_tokens": 1400, "state_memory_batch_size": 1, "preprocess_seconds": 0.00028283297433517873, "state_encode_seconds": 0.008951167023042217, "query_embed_seconds": 0.005482624983415008, "trained": false, "head_seconds": 0.0015221250068861991, "total_seconds": 0.016255374997854233, "status": "runtime_probe_only" }, { "execution": "shared_state_parallel_decision_head", "state_encodes": 1, "state_tokens": 50, "state_layers": 4, "compute_dtype": "float16", "field_count": 28, "candidate_count": 70, "query_tokens": 1288, "padded_query_tokens": 1400, "state_memory_batch_size": 1, "preprocess_seconds": 0.00028179099899716675, "state_encode_seconds": 0.008923124987632036, "query_embed_seconds": 0.0013541670050472021, "trained": false, "head_seconds": 0.005875832983292639, "total_seconds": 0.01644183398457244, "status": "runtime_probe_only" } ], "median_seconds": { "preprocess_seconds": 0.00032904199906624854, "state_encode_seconds": 0.008951167023042217, "query_embed_seconds": 0.0013541670050472021, "head_seconds": 0.001556792005430907, "total_seconds": 0.01626124998438172 }, "max_seconds": 0.016499207995366305, "under_300ms": 5 }, { "state": "image", "fields": 1, "attempted": 5, "completed": 5, "samples": [ { "execution": "shared_state_parallel_decision_head", "state_encodes": 1, "state_tokens": 295, "state_layers": 4, "compute_dtype": "float16", "field_count": 1, "candidate_count": 3, "query_tokens": 59, "padded_query_tokens": 60, "state_memory_batch_size": 1, "preprocess_seconds": 0.005595291993813589, "state_encode_seconds": 0.36221091699553654, "query_embed_seconds": 0.0004836659936700016, "trained": false, "head_seconds": 0.0018185420194640756, "total_seconds": 0.37011733299004845, "status": "runtime_probe_only" }, { "execution": "shared_state_parallel_decision_head", "state_encodes": 1, "state_tokens": 295, "state_layers": 4, "compute_dtype": "float16", "field_count": 1, "candidate_count": 3, "query_tokens": 59, "padded_query_tokens": 60, "state_memory_batch_size": 1, "preprocess_seconds": 0.004878624982666224, "state_encode_seconds": 0.36213829199550673, "query_embed_seconds": 0.0005032500193919986, "trained": false, "head_seconds": 0.0017768749967217445, "total_seconds": 0.3693049169960432, "status": "runtime_probe_only" }, { "execution": "shared_state_parallel_decision_head", "state_encodes": 1, "state_tokens": 295, "state_layers": 4, "compute_dtype": "float16", "field_count": 1, "candidate_count": 3, "query_tokens": 59, "padded_query_tokens": 60, "state_memory_batch_size": 1, "preprocess_seconds": 0.004812458995729685, "state_encode_seconds": 0.32575804099906236, "query_embed_seconds": 0.0004389169916976243, "trained": false, "head_seconds": 0.0009497079881839454, "total_seconds": 0.3319677500112448, "status": "runtime_probe_only" }, { "execution": "shared_state_parallel_decision_head", "state_encodes": 1, "state_tokens": 295, "state_layers": 4, "compute_dtype": "float16", "field_count": 1, "candidate_count": 3, "query_tokens": 59, "padded_query_tokens": 60, "state_memory_batch_size": 1, "preprocess_seconds": 0.004024583991849795, "state_encode_seconds": 0.320138750015758, "query_embed_seconds": 0.0004577910003717989, "trained": false, "head_seconds": 0.000972208013990894, "total_seconds": 0.3256002090056427, "status": "runtime_probe_only" }, { "execution": "shared_state_parallel_decision_head", "state_encodes": 1, "state_tokens": 295, "state_layers": 4, "compute_dtype": "float16", "field_count": 1, "candidate_count": 3, "query_tokens": 59, "padded_query_tokens": 60, "state_memory_batch_size": 1, "preprocess_seconds": 0.004406041989568621, "state_encode_seconds": 0.3195346250140574, "query_embed_seconds": 0.000557791005121544, "trained": false, "head_seconds": 0.0009460000146646053, "total_seconds": 0.3254530410049483, "status": "runtime_probe_only" } ], "median_seconds": { "preprocess_seconds": 0.004812458995729685, "state_encode_seconds": 0.32575804099906236, "query_embed_seconds": 0.0004836659936700016, "head_seconds": 0.000972208013990894, "total_seconds": 0.3319677500112448 }, "max_seconds": 0.37011733299004845, "under_300ms": 0 }, { "state": "image", "fields": 4, "attempted": 5, "completed": 5, "samples": [ { "execution": "shared_state_parallel_decision_head", "state_encodes": 1, "state_tokens": 295, "state_layers": 4, "compute_dtype": "float16", "field_count": 4, "candidate_count": 10, "query_tokens": 184, "padded_query_tokens": 200, "state_memory_batch_size": 1, "preprocess_seconds": 0.005467084003612399, "state_encode_seconds": 0.3563353749923408, "query_embed_seconds": 0.0005522909923456609, "trained": false, "head_seconds": 0.0010795419802889228, "total_seconds": 0.3634449999954086, "status": "runtime_probe_only" }, { "execution": "shared_state_parallel_decision_head", "state_encodes": 1, "state_tokens": 295, "state_layers": 4, "compute_dtype": "float16", "field_count": 4, "candidate_count": 10, "query_tokens": 184, "padded_query_tokens": 200, "state_memory_batch_size": 1, "preprocess_seconds": 0.004778458009241149, "state_encode_seconds": 0.35963158300728537, "query_embed_seconds": 0.000567584007512778, "trained": false, "head_seconds": 0.0010940000065602362, "total_seconds": 0.3661220410140231, "status": "runtime_probe_only" }, { "execution": "shared_state_parallel_decision_head", "state_encodes": 1, "state_tokens": 295, "state_layers": 4, "compute_dtype": "float16", "field_count": 4, "candidate_count": 10, "query_tokens": 184, "padded_query_tokens": 200, "state_memory_batch_size": 1, "preprocess_seconds": 0.004420834011398256, "state_encode_seconds": 0.3197649579960853, "query_embed_seconds": 0.00055233298917301, "trained": false, "head_seconds": 0.004800875001819804, "total_seconds": 0.3295473750040401, "status": "runtime_probe_only" }, { "execution": "shared_state_parallel_decision_head", "state_encodes": 1, "state_tokens": 295, "state_layers": 4, "compute_dtype": "float16", "field_count": 4, "candidate_count": 10, "query_tokens": 184, "padded_query_tokens": 200, "state_memory_batch_size": 1, "preprocess_seconds": 0.004490583000006154, "state_encode_seconds": 0.3360624579945579, "query_embed_seconds": 0.0005665000062435865, "trained": false, "head_seconds": 0.0010190409957431257, "total_seconds": 0.34214566700393334, "status": "runtime_probe_only" }, { "execution": "shared_state_parallel_decision_head", "state_encodes": 1, "state_tokens": 295, "state_layers": 4, "compute_dtype": "float16", "field_count": 4, "candidate_count": 10, "query_tokens": 184, "padded_query_tokens": 200, "state_memory_batch_size": 1, "preprocess_seconds": 0.004159957985393703, "state_encode_seconds": 0.31989066701498814, "query_embed_seconds": 0.0005612079985439777, "trained": false, "head_seconds": 0.001043124997522682, "total_seconds": 0.3256640409817919, "status": "runtime_probe_only" } ], "median_seconds": { "preprocess_seconds": 0.004490583000006154, "state_encode_seconds": 0.3360624579945579, "query_embed_seconds": 0.0005612079985439777, "head_seconds": 0.0010795419802889228, "total_seconds": 0.34214566700393334 }, "max_seconds": 0.3661220410140231, "under_300ms": 0 }, { "state": "image", "fields": 16, "attempted": 5, "completed": 5, "samples": [ { "execution": "shared_state_parallel_decision_head", "state_encodes": 1, "state_tokens": 295, "state_layers": 4, "compute_dtype": "float16", "field_count": 16, "candidate_count": 40, "query_tokens": 736, "padded_query_tokens": 800, "state_memory_batch_size": 1, "preprocess_seconds": 0.0053989159932825714, "state_encode_seconds": 0.36778295898693614, "query_embed_seconds": 0.0010856660082936287, "trained": false, "head_seconds": 0.0014709999959450215, "total_seconds": 0.3757588339794893, "status": "runtime_probe_only" }, { "execution": "shared_state_parallel_decision_head", "state_encodes": 1, "state_tokens": 295, "state_layers": 4, "compute_dtype": "float16", "field_count": 16, "candidate_count": 40, "query_tokens": 736, "padded_query_tokens": 800, "state_memory_batch_size": 1, "preprocess_seconds": 0.005323958001099527, "state_encode_seconds": 0.36008012501406483, "query_embed_seconds": 0.0011775839957408607, "trained": false, "head_seconds": 0.0013068749976810068, "total_seconds": 0.36790024998481385, "status": "runtime_probe_only" }, { "execution": "shared_state_parallel_decision_head", "state_encodes": 1, "state_tokens": 295, "state_layers": 4, "compute_dtype": "float16", "field_count": 16, "candidate_count": 40, "query_tokens": 736, "padded_query_tokens": 800, "state_memory_batch_size": 1, "preprocess_seconds": 0.003959333000238985, "state_encode_seconds": 0.3220959590107668, "query_embed_seconds": 0.0010212500055786222, "trained": false, "head_seconds": 0.0014404999965336174, "total_seconds": 0.32852766700671054, "status": "runtime_probe_only" }, { "execution": "shared_state_parallel_decision_head", "state_encodes": 1, "state_tokens": 295, "state_layers": 4, "compute_dtype": "float16", "field_count": 16, "candidate_count": 40, "query_tokens": 736, "padded_query_tokens": 800, "state_memory_batch_size": 1, "preprocess_seconds": 0.004326291993493214, "state_encode_seconds": 0.3152669999981299, "query_embed_seconds": 0.004162583005381748, "trained": false, "head_seconds": 0.001320417009992525, "total_seconds": 0.32509137500892393, "status": "runtime_probe_only" }, { "execution": "shared_state_parallel_decision_head", "state_encodes": 1, "state_tokens": 295, "state_layers": 4, "compute_dtype": "float16", "field_count": 16, "candidate_count": 40, "query_tokens": 736, "padded_query_tokens": 800, "state_memory_batch_size": 1, "preprocess_seconds": 0.004568875010591, "state_encode_seconds": 0.318459583009826, "query_embed_seconds": 0.0009824579756241292, "trained": false, "head_seconds": 0.005152708996320143, "total_seconds": 0.3291739590058569, "status": "runtime_probe_only" } ], "median_seconds": { "preprocess_seconds": 0.004568875010591, "state_encode_seconds": 0.3220959590107668, "query_embed_seconds": 0.0010856660082936287, "head_seconds": 0.0014404999965336174, "total_seconds": 0.3291739590058569 }, "max_seconds": 0.3757588339794893, "under_300ms": 0 }, { "state": "image", "fields": 28, "attempted": 5, "completed": 5, "samples": [ { "execution": "shared_state_parallel_decision_head", "state_encodes": 1, "state_tokens": 295, "state_layers": 4, "compute_dtype": "float16", "field_count": 28, "candidate_count": 70, "query_tokens": 1288, "padded_query_tokens": 1400, "state_memory_batch_size": 1, "preprocess_seconds": 0.004977917007636279, "state_encode_seconds": 0.3676435420056805, "query_embed_seconds": 0.0016840829921420664, "trained": false, "head_seconds": 0.001872833992820233, "total_seconds": 0.376187125017168, "status": "runtime_probe_only" }, { "execution": "shared_state_parallel_decision_head", "state_encodes": 1, "state_tokens": 295, "state_layers": 4, "compute_dtype": "float16", "field_count": 28, "candidate_count": 70, "query_tokens": 1288, "padded_query_tokens": 1400, "state_memory_batch_size": 1, "preprocess_seconds": 0.005286208994220942, "state_encode_seconds": 0.3605314160231501, "query_embed_seconds": 0.00140470897895284, "trained": false, "head_seconds": 0.005512458010343835, "total_seconds": 0.3727506250143051, "status": "runtime_probe_only" }, { "execution": "shared_state_parallel_decision_head", "state_encodes": 1, "state_tokens": 295, "state_layers": 4, "compute_dtype": "float16", "field_count": 28, "candidate_count": 70, "query_tokens": 1288, "padded_query_tokens": 1400, "state_memory_batch_size": 1, "preprocess_seconds": 0.0044749169901479036, "state_encode_seconds": 0.3184295000100974, "query_embed_seconds": 0.001550916989799589, "trained": false, "head_seconds": 0.005317625007592142, "total_seconds": 0.3297832499956712, "status": "runtime_probe_only" }, { "execution": "shared_state_parallel_decision_head", "state_encodes": 1, "state_tokens": 295, "state_layers": 4, "compute_dtype": "float16", "field_count": 28, "candidate_count": 70, "query_tokens": 1288, "padded_query_tokens": 1400, "state_memory_batch_size": 1, "preprocess_seconds": 0.0042472500063013285, "state_encode_seconds": 0.3147895839938428, "query_embed_seconds": 0.005313125002430752, "trained": false, "head_seconds": 0.0016402080073021352, "total_seconds": 0.3260105410008691, "status": "runtime_probe_only" }, { "execution": "shared_state_parallel_decision_head", "state_encodes": 1, "state_tokens": 295, "state_layers": 4, "compute_dtype": "float16", "field_count": 28, "candidate_count": 70, "query_tokens": 1288, "padded_query_tokens": 1400, "state_memory_batch_size": 1, "preprocess_seconds": 0.003875791997415945, "state_encode_seconds": 0.2992909580061678, "query_embed_seconds": 0.005023916979553178, "trained": false, "head_seconds": 0.0018469999777153134, "total_seconds": 0.31005666599958204, "status": "runtime_probe_only" } ], "median_seconds": { "preprocess_seconds": 0.0044749169901479036, "state_encode_seconds": 0.3184295000100974, "query_embed_seconds": 0.0016840829921420664, "head_seconds": 0.001872833992820233, "total_seconds": 0.3297832499956712 }, "max_seconds": 0.376187125017168, "under_300ms": 0 }, { "state": "speech", "fields": 1, "attempted": 5, "completed": 5, "samples": [ { "execution": "shared_state_parallel_decision_head", "state_encodes": 1, "state_tokens": 86, "state_layers": 4, "compute_dtype": "float16", "field_count": 1, "candidate_count": 3, "query_tokens": 59, "padded_query_tokens": 60, "state_memory_batch_size": 1, "preprocess_seconds": 0.031396124977618456, "state_encode_seconds": 0.0436764580081217, "query_embed_seconds": 0.00042275001760572195, "trained": false, "head_seconds": 0.0008944589935708791, "total_seconds": 0.07640795901534148, "status": "runtime_probe_only" }, { "execution": "shared_state_parallel_decision_head", "state_encodes": 1, "state_tokens": 86, "state_layers": 4, "compute_dtype": "float16", "field_count": 1, "candidate_count": 3, "query_tokens": 59, "padded_query_tokens": 60, "state_memory_batch_size": 1, "preprocess_seconds": 0.03511866700137034, "state_encode_seconds": 0.06589658299344592, "query_embed_seconds": 0.0004990000161342323, "trained": false, "head_seconds": 0.0009330830071121454, "total_seconds": 0.10248795899678953, "status": "runtime_probe_only" }, { "execution": "shared_state_parallel_decision_head", "state_encodes": 1, "state_tokens": 86, "state_layers": 4, "compute_dtype": "float16", "field_count": 1, "candidate_count": 3, "query_tokens": 59, "padded_query_tokens": 60, "state_memory_batch_size": 1, "preprocess_seconds": 0.03605891700135544, "state_encode_seconds": 0.070999416988343, "query_embed_seconds": 0.00045508300536312163, "trained": false, "head_seconds": 0.000944707979215309, "total_seconds": 0.10846958300680853, "status": "runtime_probe_only" }, { "execution": "shared_state_parallel_decision_head", "state_encodes": 1, "state_tokens": 86, "state_layers": 4, "compute_dtype": "float16", "field_count": 1, "candidate_count": 3, "query_tokens": 59, "padded_query_tokens": 60, "state_memory_batch_size": 1, "preprocess_seconds": 0.034525166993262246, "state_encode_seconds": 0.07261270799790509, "query_embed_seconds": 0.0005415000196080655, "trained": false, "head_seconds": 0.0009656249894760549, "total_seconds": 0.10865670899511315, "status": "runtime_probe_only" }, { "execution": "shared_state_parallel_decision_head", "state_encodes": 1, "state_tokens": 86, "state_layers": 4, "compute_dtype": "float16", "field_count": 1, "candidate_count": 3, "query_tokens": 59, "padded_query_tokens": 60, "state_memory_batch_size": 1, "preprocess_seconds": 0.03247049998026341, "state_encode_seconds": 0.07464787500794046, "query_embed_seconds": 0.00046429099165834486, "trained": false, "head_seconds": 0.0009176659805234522, "total_seconds": 0.10851237500901334, "status": "runtime_probe_only" } ], "median_seconds": { "preprocess_seconds": 0.034525166993262246, "state_encode_seconds": 0.070999416988343, "query_embed_seconds": 0.00046429099165834486, "head_seconds": 0.0009330830071121454, "total_seconds": 0.10846958300680853 }, "max_seconds": 0.10865670899511315, "under_300ms": 5 }, { "state": "speech", "fields": 4, "attempted": 5, "completed": 5, "samples": [ { "execution": "shared_state_parallel_decision_head", "state_encodes": 1, "state_tokens": 86, "state_layers": 4, "compute_dtype": "float16", "field_count": 4, "candidate_count": 10, "query_tokens": 184, "padded_query_tokens": 200, "state_memory_batch_size": 1, "preprocess_seconds": 0.03192487498745322, "state_encode_seconds": 0.0524836250115186, "query_embed_seconds": 0.0005765410023741424, "trained": false, "head_seconds": 0.001097832981031388, "total_seconds": 0.08609954197891057, "status": "runtime_probe_only" }, { "execution": "shared_state_parallel_decision_head", "state_encodes": 1, "state_tokens": 86, "state_layers": 4, "compute_dtype": "float16", "field_count": 4, "candidate_count": 10, "query_tokens": 184, "padded_query_tokens": 200, "state_memory_batch_size": 1, "preprocess_seconds": 0.03559183399192989, "state_encode_seconds": 0.06971049998537637, "query_embed_seconds": 0.0014668330259155482, "trained": false, "head_seconds": 0.0010487090039532632, "total_seconds": 0.10784304200205952, "status": "runtime_probe_only" }, { "execution": "shared_state_parallel_decision_head", "state_encodes": 1, "state_tokens": 86, "state_layers": 4, "compute_dtype": "float16", "field_count": 4, "candidate_count": 10, "query_tokens": 184, "padded_query_tokens": 200, "state_memory_batch_size": 1, "preprocess_seconds": 0.03607825000653975, "state_encode_seconds": 0.07108808399061672, "query_embed_seconds": 0.0006489579973276705, "trained": false, "head_seconds": 0.0015701670199632645, "total_seconds": 0.10939620799035765, "status": "runtime_probe_only" }, { "execution": "shared_state_parallel_decision_head", "state_encodes": 1, "state_tokens": 86, "state_layers": 4, "compute_dtype": "float16", "field_count": 4, "candidate_count": 10, "query_tokens": 184, "padded_query_tokens": 200, "state_memory_batch_size": 1, "preprocess_seconds": 0.03409149998333305, "state_encode_seconds": 0.07258620799984783, "query_embed_seconds": 0.0006699580117128789, "trained": false, "head_seconds": 0.0009677080088295043, "total_seconds": 0.1083422499941662, "status": "runtime_probe_only" }, { "execution": "shared_state_parallel_decision_head", "state_encodes": 1, "state_tokens": 86, "state_layers": 4, "compute_dtype": "float16", "field_count": 4, "candidate_count": 10, "query_tokens": 184, "padded_query_tokens": 200, "state_memory_batch_size": 1, "preprocess_seconds": 0.034657917014556006, "state_encode_seconds": 0.07322229197598062, "query_embed_seconds": 0.0005928330065216869, "trained": false, "head_seconds": 0.004836958018131554, "total_seconds": 0.11332204201607965, "status": "runtime_probe_only" } ], "median_seconds": { "preprocess_seconds": 0.034657917014556006, "state_encode_seconds": 0.07108808399061672, "query_embed_seconds": 0.0006489579973276705, "head_seconds": 0.001097832981031388, "total_seconds": 0.1083422499941662 }, "max_seconds": 0.11332204201607965, "under_300ms": 5 }, { "state": "speech", "fields": 16, "attempted": 5, "completed": 5, "samples": [ { "execution": "shared_state_parallel_decision_head", "state_encodes": 1, "state_tokens": 86, "state_layers": 4, "compute_dtype": "float16", "field_count": 16, "candidate_count": 40, "query_tokens": 736, "padded_query_tokens": 800, "state_memory_batch_size": 1, "preprocess_seconds": 0.03545691698673181, "state_encode_seconds": 0.0629102920065634, "query_embed_seconds": 0.0010870409896597266, "trained": false, "head_seconds": 0.001274041016586125, "total_seconds": 0.10074049999821, "status": "runtime_probe_only" }, { "execution": "shared_state_parallel_decision_head", "state_encodes": 1, "state_tokens": 86, "state_layers": 4, "compute_dtype": "float16", "field_count": 16, "candidate_count": 40, "query_tokens": 736, "padded_query_tokens": 800, "state_memory_batch_size": 1, "preprocess_seconds": 0.04608766600722447, "state_encode_seconds": 0.06811424999614246, "query_embed_seconds": 0.004294416983611882, "trained": false, "head_seconds": 0.0013261249987408519, "total_seconds": 0.11984108400065452, "status": "runtime_probe_only" }, { "execution": "shared_state_parallel_decision_head", "state_encodes": 1, "state_tokens": 86, "state_layers": 4, "compute_dtype": "float16", "field_count": 16, "candidate_count": 40, "query_tokens": 736, "padded_query_tokens": 800, "state_memory_batch_size": 1, "preprocess_seconds": 0.03649733398924582, "state_encode_seconds": 0.06450708300690167, "query_embed_seconds": 0.003968458011513576, "trained": false, "head_seconds": 0.0018946249911095947, "total_seconds": 0.10690174999763258, "status": "runtime_probe_only" }, { "execution": "shared_state_parallel_decision_head", "state_encodes": 1, "state_tokens": 86, "state_layers": 4, "compute_dtype": "float16", "field_count": 16, "candidate_count": 40, "query_tokens": 736, "padded_query_tokens": 800, "state_memory_batch_size": 1, "preprocess_seconds": 0.037452542019309476, "state_encode_seconds": 0.07514387500123121, "query_embed_seconds": 0.0011247499787714332, "trained": false, "head_seconds": 0.0012524579942692071, "total_seconds": 0.1149917500151787, "status": "runtime_probe_only" }, { "execution": "shared_state_parallel_decision_head", "state_encodes": 1, "state_tokens": 86, "state_layers": 4, "compute_dtype": "float16", "field_count": 16, "candidate_count": 40, "query_tokens": 736, "padded_query_tokens": 800, "state_memory_batch_size": 1, "preprocess_seconds": 0.032752040977356955, "state_encode_seconds": 0.07653779201791622, "query_embed_seconds": 0.0014647079806309193, "trained": false, "head_seconds": 0.0017112079949583858, "total_seconds": 0.11248954199254513, "status": "runtime_probe_only" } ], "median_seconds": { "preprocess_seconds": 0.03649733398924582, "state_encode_seconds": 0.06811424999614246, "query_embed_seconds": 0.0014647079806309193, "head_seconds": 0.0013261249987408519, "total_seconds": 0.11248954199254513 }, "max_seconds": 0.11984108400065452, "under_300ms": 5 }, { "state": "speech", "fields": 28, "attempted": 5, "completed": 5, "samples": [ { "execution": "shared_state_parallel_decision_head", "state_encodes": 1, "state_tokens": 86, "state_layers": 4, "compute_dtype": "float16", "field_count": 28, "candidate_count": 70, "query_tokens": 1288, "padded_query_tokens": 1400, "state_memory_batch_size": 1, "preprocess_seconds": 0.035324249998666346, "state_encode_seconds": 0.06583820801461115, "query_embed_seconds": 0.005072542000561953, "trained": false, "head_seconds": 0.002535957988584414, "total_seconds": 0.10882504101027735, "status": "runtime_probe_only" }, { "execution": "shared_state_parallel_decision_head", "state_encodes": 1, "state_tokens": 86, "state_layers": 4, "compute_dtype": "float16", "field_count": 28, "candidate_count": 70, "query_tokens": 1288, "padded_query_tokens": 1400, "state_memory_batch_size": 1, "preprocess_seconds": 0.06560937498579733, "state_encode_seconds": 0.0822420000040438, "query_embed_seconds": 0.001652042003115639, "trained": false, "head_seconds": 0.0032965410209726542, "total_seconds": 0.1528195420105476, "status": "runtime_probe_only" }, { "execution": "shared_state_parallel_decision_head", "state_encodes": 1, "state_tokens": 86, "state_layers": 4, "compute_dtype": "float16", "field_count": 28, "candidate_count": 70, "query_tokens": 1288, "padded_query_tokens": 1400, "state_memory_batch_size": 1, "preprocess_seconds": 0.03487399997538887, "state_encode_seconds": 0.062173792015528306, "query_embed_seconds": 0.0015589999966323376, "trained": false, "head_seconds": 0.001533333008410409, "total_seconds": 0.10015141699113883, "status": "runtime_probe_only" }, { "execution": "shared_state_parallel_decision_head", "state_encodes": 1, "state_tokens": 86, "state_layers": 4, "compute_dtype": "float16", "field_count": 28, "candidate_count": 70, "query_tokens": 1288, "padded_query_tokens": 1400, "state_memory_batch_size": 1, "preprocess_seconds": 0.0338916250038892, "state_encode_seconds": 0.07222370800445788, "query_embed_seconds": 0.001549959008116275, "trained": false, "head_seconds": 0.002160500007448718, "total_seconds": 0.1098396250163205, "status": "runtime_probe_only" }, { "execution": "shared_state_parallel_decision_head", "state_encodes": 1, "state_tokens": 86, "state_layers": 4, "compute_dtype": "float16", "field_count": 28, "candidate_count": 70, "query_tokens": 1288, "padded_query_tokens": 1400, "state_memory_batch_size": 1, "preprocess_seconds": 0.034672875015530735, "state_encode_seconds": 0.07144483298179694, "query_embed_seconds": 0.0014543330180458724, "trained": false, "head_seconds": 0.005861708981683478, "total_seconds": 0.11344654101412743, "status": "runtime_probe_only" } ], "median_seconds": { "preprocess_seconds": 0.03487399997538887, "state_encode_seconds": 0.07144483298179694, "query_embed_seconds": 0.0015589999966323376, "head_seconds": 0.002535957988584414, "total_seconds": 0.1098396250163205 }, "max_seconds": 0.1528195420105476, "under_300ms": 5 }, { "state": "video", "fields": 1, "attempted": 5, "completed": 5, "samples": [ { "execution": "shared_state_parallel_decision_head", "state_encodes": 1, "state_tokens": 328, "state_layers": 4, "compute_dtype": "float16", "field_count": 1, "candidate_count": 3, "query_tokens": 59, "padded_query_tokens": 60, "state_memory_batch_size": 1, "preprocess_seconds": 0.04621179201058112, "state_encode_seconds": 0.33686112498980947, "query_embed_seconds": 0.0006558330205734819, "trained": false, "head_seconds": 0.001800333004212007, "total_seconds": 0.38557074998971075, "status": "runtime_probe_only" }, { "execution": "shared_state_parallel_decision_head", "state_encodes": 1, "state_tokens": 328, "state_layers": 4, "compute_dtype": "float16", "field_count": 1, "candidate_count": 3, "query_tokens": 59, "padded_query_tokens": 60, "state_memory_batch_size": 1, "preprocess_seconds": 0.044420499994885176, "state_encode_seconds": 0.32251583301695064, "query_embed_seconds": 0.0004813330015167594, "trained": false, "head_seconds": 0.0010356249986216426, "total_seconds": 0.3684699170116801, "status": "runtime_probe_only" }, { "execution": "shared_state_parallel_decision_head", "state_encodes": 1, "state_tokens": 328, "state_layers": 4, "compute_dtype": "float16", "field_count": 1, "candidate_count": 3, "query_tokens": 59, "padded_query_tokens": 60, "state_memory_batch_size": 1, "preprocess_seconds": 0.04364258301211521, "state_encode_seconds": 0.33034320900333114, "query_embed_seconds": 0.0004352079995442182, "trained": false, "head_seconds": 0.0017733339918777347, "total_seconds": 0.37622466700850055, "status": "runtime_probe_only" }, { "execution": "shared_state_parallel_decision_head", "state_encodes": 1, "state_tokens": 328, "state_layers": 4, "compute_dtype": "float16", "field_count": 1, "candidate_count": 3, "query_tokens": 59, "padded_query_tokens": 60, "state_memory_batch_size": 1, "preprocess_seconds": 0.04329629198764451, "state_encode_seconds": 0.3338427499984391, "query_embed_seconds": 0.0004555000050459057, "trained": false, "head_seconds": 0.0017669170047156513, "total_seconds": 0.37937816701014526, "status": "runtime_probe_only" }, { "execution": "shared_state_parallel_decision_head", "state_encodes": 1, "state_tokens": 328, "state_layers": 4, "compute_dtype": "float16", "field_count": 1, "candidate_count": 3, "query_tokens": 59, "padded_query_tokens": 60, "state_memory_batch_size": 1, "preprocess_seconds": 0.042434208997292444, "state_encode_seconds": 0.32306783300009556, "query_embed_seconds": 0.0011342499929014593, "trained": false, "head_seconds": 0.003663417010102421, "total_seconds": 0.370322708011372, "status": "runtime_probe_only" } ], "median_seconds": { "preprocess_seconds": 0.04364258301211521, "state_encode_seconds": 0.33034320900333114, "query_embed_seconds": 0.0004813330015167594, "head_seconds": 0.0017733339918777347, "total_seconds": 0.37622466700850055 }, "max_seconds": 0.38557074998971075, "under_300ms": 0 }, { "state": "video", "fields": 4, "attempted": 5, "completed": 5, "samples": [ { "execution": "shared_state_parallel_decision_head", "state_encodes": 1, "state_tokens": 328, "state_layers": 4, "compute_dtype": "float16", "field_count": 4, "candidate_count": 10, "query_tokens": 184, "padded_query_tokens": 200, "state_memory_batch_size": 1, "preprocess_seconds": 0.049829958006739616, "state_encode_seconds": 0.3729558330087457, "query_embed_seconds": 0.0007689589983783662, "trained": false, "head_seconds": 0.0011911670153494924, "total_seconds": 0.42476745799649507, "status": "runtime_probe_only" }, { "execution": "shared_state_parallel_decision_head", "state_encodes": 1, "state_tokens": 328, "state_layers": 4, "compute_dtype": "float16", "field_count": 4, "candidate_count": 10, "query_tokens": 184, "padded_query_tokens": 200, "state_memory_batch_size": 1, "preprocess_seconds": 0.04350520798470825, "state_encode_seconds": 0.3315720419923309, "query_embed_seconds": 0.0005626250058412552, "trained": false, "head_seconds": 0.0010704589949455112, "total_seconds": 0.37672991698491387, "status": "runtime_probe_only" }, { "execution": "shared_state_parallel_decision_head", "state_encodes": 1, "state_tokens": 328, "state_layers": 4, "compute_dtype": "float16", "field_count": 4, "candidate_count": 10, "query_tokens": 184, "padded_query_tokens": 200, "state_memory_batch_size": 1, "preprocess_seconds": 0.045985583012225106, "state_encode_seconds": 0.33338720799656585, "query_embed_seconds": 0.0005533339863177389, "trained": false, "head_seconds": 0.0011158340203110129, "total_seconds": 0.3810606669867411, "status": "runtime_probe_only" }, { "execution": "shared_state_parallel_decision_head", "state_encodes": 1, "state_tokens": 328, "state_layers": 4, "compute_dtype": "float16", "field_count": 4, "candidate_count": 10, "query_tokens": 184, "padded_query_tokens": 200, "state_memory_batch_size": 1, "preprocess_seconds": 0.044491207983810455, "state_encode_seconds": 0.33000245899893343, "query_embed_seconds": 0.0005855830095242709, "trained": false, "head_seconds": 0.004968624998582527, "total_seconds": 0.38007554100477137, "status": "runtime_probe_only" }, { "execution": "shared_state_parallel_decision_head", "state_encodes": 1, "state_tokens": 328, "state_layers": 4, "compute_dtype": "float16", "field_count": 4, "candidate_count": 10, "query_tokens": 184, "padded_query_tokens": 200, "state_memory_batch_size": 1, "preprocess_seconds": 0.046067707997281104, "state_encode_seconds": 0.3311900419939775, "query_embed_seconds": 0.0006082080071792006, "trained": false, "head_seconds": 0.0016893749998416752, "total_seconds": 0.3795728339755442, "status": "runtime_probe_only" } ], "median_seconds": { "preprocess_seconds": 0.045985583012225106, "state_encode_seconds": 0.3315720419923309, "query_embed_seconds": 0.0005855830095242709, "head_seconds": 0.0011911670153494924, "total_seconds": 0.38007554100477137 }, "max_seconds": 0.42476745799649507, "under_300ms": 0 }, { "state": "video", "fields": 16, "attempted": 5, "completed": 5, "samples": [ { "execution": "shared_state_parallel_decision_head", "state_encodes": 1, "state_tokens": 328, "state_layers": 4, "compute_dtype": "float16", "field_count": 16, "candidate_count": 40, "query_tokens": 736, "padded_query_tokens": 800, "state_memory_batch_size": 1, "preprocess_seconds": 0.04789545899257064, "state_encode_seconds": 0.37454600000637583, "query_embed_seconds": 0.0010934160090982914, "trained": false, "head_seconds": 0.0014610829821322113, "total_seconds": 0.42501229199115187, "status": "runtime_probe_only" }, { "execution": "shared_state_parallel_decision_head", "state_encodes": 1, "state_tokens": 328, "state_layers": 4, "compute_dtype": "float16", "field_count": 16, "candidate_count": 40, "query_tokens": 736, "padded_query_tokens": 800, "state_memory_batch_size": 1, "preprocess_seconds": 0.043852124974364415, "state_encode_seconds": 0.33194679202279076, "query_embed_seconds": 0.0010320000001229346, "trained": false, "head_seconds": 0.005168417003005743, "total_seconds": 0.3820179169997573, "status": "runtime_probe_only" }, { "execution": "shared_state_parallel_decision_head", "state_encodes": 1, "state_tokens": 328, "state_layers": 4, "compute_dtype": "float16", "field_count": 16, "candidate_count": 40, "query_tokens": 736, "padded_query_tokens": 800, "state_memory_batch_size": 1, "preprocess_seconds": 0.04767404100857675, "state_encode_seconds": 0.33274062498821877, "query_embed_seconds": 0.0012474590039346367, "trained": false, "head_seconds": 0.0015768329903949052, "total_seconds": 0.38325920898932964, "status": "runtime_probe_only" }, { "execution": "shared_state_parallel_decision_head", "state_encodes": 1, "state_tokens": 328, "state_layers": 4, "compute_dtype": "float16", "field_count": 16, "candidate_count": 40, "query_tokens": 736, "padded_query_tokens": 800, "state_memory_batch_size": 1, "preprocess_seconds": 0.03970925000612624, "state_encode_seconds": 0.3346290829940699, "query_embed_seconds": 0.0010665000008884817, "trained": false, "head_seconds": 0.0013300839927978814, "total_seconds": 0.37675291701452807, "status": "runtime_probe_only" }, { "execution": "shared_state_parallel_decision_head", "state_encodes": 1, "state_tokens": 328, "state_layers": 4, "compute_dtype": "float16", "field_count": 16, "candidate_count": 40, "query_tokens": 736, "padded_query_tokens": 800, "state_memory_batch_size": 1, "preprocess_seconds": 0.043237000005319715, "state_encode_seconds": 0.328029249998508, "query_embed_seconds": 0.001045000011799857, "trained": false, "head_seconds": 0.0014704159984830767, "total_seconds": 0.3738062500196975, "status": "runtime_probe_only" } ], "median_seconds": { "preprocess_seconds": 0.043852124974364415, "state_encode_seconds": 0.33274062498821877, "query_embed_seconds": 0.0010665000008884817, "head_seconds": 0.0014704159984830767, "total_seconds": 0.3820179169997573 }, "max_seconds": 0.42501229199115187, "under_300ms": 0 }, { "state": "video", "fields": 28, "attempted": 5, "completed": 5, "samples": [ { "execution": "shared_state_parallel_decision_head", "state_encodes": 1, "state_tokens": 328, "state_layers": 4, "compute_dtype": "float16", "field_count": 28, "candidate_count": 70, "query_tokens": 1288, "padded_query_tokens": 1400, "state_memory_batch_size": 1, "preprocess_seconds": 0.04449545798706822, "state_encode_seconds": 0.3301790420082398, "query_embed_seconds": 0.005318707990227267, "trained": false, "head_seconds": 0.0019158330105710775, "total_seconds": 0.38195004098815843, "status": "runtime_probe_only" }, { "execution": "shared_state_parallel_decision_head", "state_encodes": 1, "state_tokens": 328, "state_layers": 4, "compute_dtype": "float16", "field_count": 28, "candidate_count": 70, "query_tokens": 1288, "padded_query_tokens": 1400, "state_memory_batch_size": 1, "preprocess_seconds": 0.04533074999926612, "state_encode_seconds": 0.3308066249883268, "query_embed_seconds": 0.005158874992048368, "trained": false, "head_seconds": 0.0017686249921098351, "total_seconds": 0.3830923750065267, "status": "runtime_probe_only" }, { "execution": "shared_state_parallel_decision_head", "state_encodes": 1, "state_tokens": 328, "state_layers": 4, "compute_dtype": "float16", "field_count": 28, "candidate_count": 70, "query_tokens": 1288, "padded_query_tokens": 1400, "state_memory_batch_size": 1, "preprocess_seconds": 0.044738083000993356, "state_encode_seconds": 0.33114108399604447, "query_embed_seconds": 0.0046508329978678375, "trained": false, "head_seconds": 0.001830082997912541, "total_seconds": 0.3823893750086427, "status": "runtime_probe_only" }, { "execution": "shared_state_parallel_decision_head", "state_encodes": 1, "state_tokens": 328, "state_layers": 4, "compute_dtype": "float16", "field_count": 28, "candidate_count": 70, "query_tokens": 1288, "padded_query_tokens": 1400, "state_memory_batch_size": 1, "preprocess_seconds": 0.04238412500126287, "state_encode_seconds": 0.3350877920165658, "query_embed_seconds": 0.004307082999730483, "trained": false, "head_seconds": 0.0017003749962896109, "total_seconds": 0.38350266599445604, "status": "runtime_probe_only" }, { "execution": "shared_state_parallel_decision_head", "state_encodes": 1, "state_tokens": 328, "state_layers": 4, "compute_dtype": "float16", "field_count": 28, "candidate_count": 70, "query_tokens": 1288, "padded_query_tokens": 1400, "state_memory_batch_size": 1, "preprocess_seconds": 0.041229791997466236, "state_encode_seconds": 0.3392020000028424, "query_embed_seconds": 0.0014917499793227762, "trained": false, "head_seconds": 0.0018632090068422258, "total_seconds": 0.3838059580011759, "status": "runtime_probe_only" } ], "median_seconds": { "preprocess_seconds": 0.04449545798706822, "state_encode_seconds": 0.33114108399604447, "query_embed_seconds": 0.0046508329978678375, "head_seconds": 0.001830082997912541, "total_seconds": 0.3830923750065267 }, "max_seconds": 0.3838059580011759, "under_300ms": 0 }, { "state": "video_speech", "fields": 1, "attempted": 5, "completed": 5, "samples": [ { "execution": "shared_state_parallel_decision_head", "state_encodes": 1, "state_tokens": 431, "state_layers": 4, "compute_dtype": "float16", "field_count": 1, "candidate_count": 3, "query_tokens": 59, "padded_query_tokens": 60, "state_memory_batch_size": 1, "preprocess_seconds": 0.08852754099643789, "state_encode_seconds": 0.4275756249844562, "query_embed_seconds": 0.0005857920041307807, "trained": false, "head_seconds": 0.0015457499830517918, "total_seconds": 0.5182641669816803, "status": "runtime_probe_only" }, { "execution": "shared_state_parallel_decision_head", "state_encodes": 1, "state_tokens": 431, "state_layers": 4, "compute_dtype": "float16", "field_count": 1, "candidate_count": 3, "query_tokens": 59, "padded_query_tokens": 60, "state_memory_batch_size": 1, "preprocess_seconds": 0.08192562498152256, "state_encode_seconds": 0.5915635830024257, "query_embed_seconds": 0.00043966699740849435, "trained": false, "head_seconds": 0.0010459170152898878, "total_seconds": 0.6749866659811232, "status": "runtime_probe_only" }, { "execution": "shared_state_parallel_decision_head", "state_encodes": 1, "state_tokens": 431, "state_layers": 4, "compute_dtype": "float16", "field_count": 1, "candidate_count": 3, "query_tokens": 59, "padded_query_tokens": 60, "state_memory_batch_size": 1, "preprocess_seconds": 0.08051412500208244, "state_encode_seconds": 0.4263494999904651, "query_embed_seconds": 0.00040320900734514, "trained": false, "head_seconds": 0.0010301249858457595, "total_seconds": 0.5083087499951944, "status": "runtime_probe_only" }, { "execution": "shared_state_parallel_decision_head", "state_encodes": 1, "state_tokens": 431, "state_layers": 4, "compute_dtype": "float16", "field_count": 1, "candidate_count": 3, "query_tokens": 59, "padded_query_tokens": 60, "state_memory_batch_size": 1, "preprocess_seconds": 0.07919287501135841, "state_encode_seconds": 0.4182466249912977, "query_embed_seconds": 0.0004272499936632812, "trained": false, "head_seconds": 0.004997334006475285, "total_seconds": 0.5028767500189133, "status": "runtime_probe_only" }, { "execution": "shared_state_parallel_decision_head", "state_encodes": 1, "state_tokens": 431, "state_layers": 4, "compute_dtype": "float16", "field_count": 1, "candidate_count": 3, "query_tokens": 59, "padded_query_tokens": 60, "state_memory_batch_size": 1, "preprocess_seconds": 0.07751470801304094, "state_encode_seconds": 0.4165015000035055, "query_embed_seconds": 0.0004484999808482826, "trained": false, "head_seconds": 0.0024128340010065585, "total_seconds": 0.4968912500189617, "status": "runtime_probe_only" } ], "median_seconds": { "preprocess_seconds": 0.08051412500208244, "state_encode_seconds": 0.4263494999904651, "query_embed_seconds": 0.00043966699740849435, "head_seconds": 0.0015457499830517918, "total_seconds": 0.5083087499951944 }, "max_seconds": 0.6749866659811232, "under_300ms": 0 }, { "state": "video_speech", "fields": 4, "attempted": 5, "completed": 5, "samples": [ { "execution": "shared_state_parallel_decision_head", "state_encodes": 1, "state_tokens": 431, "state_layers": 4, "compute_dtype": "float16", "field_count": 4, "candidate_count": 10, "query_tokens": 184, "padded_query_tokens": 200, "state_memory_batch_size": 1, "preprocess_seconds": 0.08572858298430219, "state_encode_seconds": 0.42855154201970436, "query_embed_seconds": 0.0005762499931734055, "trained": false, "head_seconds": 0.0012091250100638717, "total_seconds": 0.5160795839910861, "status": "runtime_probe_only" }, { "execution": "shared_state_parallel_decision_head", "state_encodes": 1, "state_tokens": 431, "state_layers": 4, "compute_dtype": "float16", "field_count": 4, "candidate_count": 10, "query_tokens": 184, "padded_query_tokens": 200, "state_memory_batch_size": 1, "preprocess_seconds": 0.08521187500446104, "state_encode_seconds": 0.5899692079983652, "query_embed_seconds": 0.0005725419905502349, "trained": false, "head_seconds": 0.0061562079936265945, "total_seconds": 0.6819245000078809, "status": "runtime_probe_only" }, { "execution": "shared_state_parallel_decision_head", "state_encodes": 1, "state_tokens": 431, "state_layers": 4, "compute_dtype": "float16", "field_count": 4, "candidate_count": 10, "query_tokens": 184, "padded_query_tokens": 200, "state_memory_batch_size": 1, "preprocess_seconds": 0.07832637499086559, "state_encode_seconds": 0.43640574999153614, "query_embed_seconds": 0.0005622919998131692, "trained": false, "head_seconds": 0.001158207975095138, "total_seconds": 0.5164668340003118, "status": "runtime_probe_only" }, { "execution": "shared_state_parallel_decision_head", "state_encodes": 1, "state_tokens": 431, "state_layers": 4, "compute_dtype": "float16", "field_count": 4, "candidate_count": 10, "query_tokens": 184, "padded_query_tokens": 200, "state_memory_batch_size": 1, "preprocess_seconds": 0.08255774999270216, "state_encode_seconds": 0.4119445420219563, "query_embed_seconds": 0.000577290978981182, "trained": false, "head_seconds": 0.002631083014421165, "total_seconds": 0.49772358301561326, "status": "runtime_probe_only" }, { "execution": "shared_state_parallel_decision_head", "state_encodes": 1, "state_tokens": 431, "state_layers": 4, "compute_dtype": "float16", "field_count": 4, "candidate_count": 10, "query_tokens": 184, "padded_query_tokens": 200, "state_memory_batch_size": 1, "preprocess_seconds": 0.08021629200084135, "state_encode_seconds": 0.4227657080045901, "query_embed_seconds": 0.000570792006328702, "trained": false, "head_seconds": 0.001174540986539796, "total_seconds": 0.5047400000039488, "status": "runtime_probe_only" } ], "median_seconds": { "preprocess_seconds": 0.08255774999270216, "state_encode_seconds": 0.42855154201970436, "query_embed_seconds": 0.0005725419905502349, "head_seconds": 0.0012091250100638717, "total_seconds": 0.5160795839910861 }, "max_seconds": 0.6819245000078809, "under_300ms": 0 }, { "state": "video_speech", "fields": 16, "attempted": 5, "completed": 5, "samples": [ { "execution": "shared_state_parallel_decision_head", "state_encodes": 1, "state_tokens": 431, "state_layers": 4, "compute_dtype": "float16", "field_count": 16, "candidate_count": 40, "query_tokens": 736, "padded_query_tokens": 800, "state_memory_batch_size": 1, "preprocess_seconds": 0.07821920901187696, "state_encode_seconds": 0.4116681659943424, "query_embed_seconds": 0.0012529590167105198, "trained": false, "head_seconds": 0.0055263749964069575, "total_seconds": 0.49668491701595485, "status": "runtime_probe_only" }, { "execution": "shared_state_parallel_decision_head", "state_encodes": 1, "state_tokens": 431, "state_layers": 4, "compute_dtype": "float16", "field_count": 16, "candidate_count": 40, "query_tokens": 736, "padded_query_tokens": 800, "state_memory_batch_size": 1, "preprocess_seconds": 0.0808782080130186, "state_encode_seconds": 0.5879356669902336, "query_embed_seconds": 0.0036664170038420707, "trained": false, "head_seconds": 0.001552417001221329, "total_seconds": 0.6740478750143666, "status": "runtime_probe_only" }, { "execution": "shared_state_parallel_decision_head", "state_encodes": 1, "state_tokens": 431, "state_layers": 4, "compute_dtype": "float16", "field_count": 16, "candidate_count": 40, "query_tokens": 736, "padded_query_tokens": 800, "state_memory_batch_size": 1, "preprocess_seconds": 0.086143082997296, "state_encode_seconds": 0.44048808299703524, "query_embed_seconds": 0.004388459026813507, "trained": false, "head_seconds": 0.0016619589878246188, "total_seconds": 0.5327002090052702, "status": "runtime_probe_only" }, { "execution": "shared_state_parallel_decision_head", "state_encodes": 1, "state_tokens": 431, "state_layers": 4, "compute_dtype": "float16", "field_count": 16, "candidate_count": 40, "query_tokens": 736, "padded_query_tokens": 800, "state_memory_batch_size": 1, "preprocess_seconds": 0.08164820898673497, "state_encode_seconds": 0.42478445800952613, "query_embed_seconds": 0.0010577919892966747, "trained": false, "head_seconds": 0.005263041995931417, "total_seconds": 0.5127704580081627, "status": "runtime_probe_only" }, { "execution": "shared_state_parallel_decision_head", "state_encodes": 1, "state_tokens": 431, "state_layers": 4, "compute_dtype": "float16", "field_count": 16, "candidate_count": 40, "query_tokens": 736, "padded_query_tokens": 800, "state_memory_batch_size": 1, "preprocess_seconds": 0.08167154199327342, "state_encode_seconds": 0.4108351659961045, "query_embed_seconds": 0.0010459170152898878, "trained": false, "head_seconds": 0.005381499999202788, "total_seconds": 0.4989492919994518, "status": "runtime_probe_only" } ], "median_seconds": { "preprocess_seconds": 0.08164820898673497, "state_encode_seconds": 0.42478445800952613, "query_embed_seconds": 0.0012529590167105198, "head_seconds": 0.005263041995931417, "total_seconds": 0.5127704580081627 }, "max_seconds": 0.6740478750143666, "under_300ms": 0 }, { "state": "video_speech", "fields": 28, "attempted": 5, "completed": 5, "samples": [ { "execution": "shared_state_parallel_decision_head", "state_encodes": 1, "state_tokens": 431, "state_layers": 4, "compute_dtype": "float16", "field_count": 28, "candidate_count": 70, "query_tokens": 1288, "padded_query_tokens": 1400, "state_memory_batch_size": 1, "preprocess_seconds": 0.0808088339981623, "state_encode_seconds": 0.41237237499444745, "query_embed_seconds": 0.0015076659910846502, "trained": false, "head_seconds": 0.005776999983936548, "total_seconds": 0.5004812089900952, "status": "runtime_probe_only" }, { "execution": "shared_state_parallel_decision_head", "state_encodes": 1, "state_tokens": 431, "state_layers": 4, "compute_dtype": "float16", "field_count": 28, "candidate_count": 70, "query_tokens": 1288, "padded_query_tokens": 1400, "state_memory_batch_size": 1, "preprocess_seconds": 0.07998029197915457, "state_encode_seconds": 0.40825279100681655, "query_embed_seconds": 0.005527834000531584, "trained": false, "head_seconds": 0.001817334006773308, "total_seconds": 0.49560216697864234, "status": "runtime_probe_only" }, { "execution": "shared_state_parallel_decision_head", "state_encodes": 1, "state_tokens": 431, "state_layers": 4, "compute_dtype": "float16", "field_count": 28, "candidate_count": 70, "query_tokens": 1288, "padded_query_tokens": 1400, "state_memory_batch_size": 1, "preprocess_seconds": 0.09298524999758229, "state_encode_seconds": 0.40932662499835715, "query_embed_seconds": 0.004486375022679567, "trained": false, "head_seconds": 0.0019725000020116568, "total_seconds": 0.5087852920114528, "status": "runtime_probe_only" }, { "execution": "shared_state_parallel_decision_head", "state_encodes": 1, "state_tokens": 431, "state_layers": 4, "compute_dtype": "float16", "field_count": 28, "candidate_count": 70, "query_tokens": 1288, "padded_query_tokens": 1400, "state_memory_batch_size": 1, "preprocess_seconds": 0.07603641698369756, "state_encode_seconds": 0.4471763750188984, "query_embed_seconds": 0.001449249975848943, "trained": false, "head_seconds": 0.001757292018737644, "total_seconds": 0.526431166974362, "status": "runtime_probe_only" }, { "execution": "shared_state_parallel_decision_head", "state_encodes": 1, "state_tokens": 431, "state_layers": 4, "compute_dtype": "float16", "field_count": 28, "candidate_count": 70, "query_tokens": 1288, "padded_query_tokens": 1400, "state_memory_batch_size": 1, "preprocess_seconds": 0.07963683397974819, "state_encode_seconds": 0.4145796250086278, "query_embed_seconds": 0.004962500010151416, "trained": false, "head_seconds": 0.0018729580042418092, "total_seconds": 0.5010684590088204, "status": "runtime_probe_only" } ], "median_seconds": { "preprocess_seconds": 0.07998029197915457, "state_encode_seconds": 0.41237237499444745, "query_embed_seconds": 0.004486375022679567, "head_seconds": 0.0018729580042418092, "total_seconds": 0.5010684590088204 }, "max_seconds": 0.526431166974362, "under_300ms": 0 } ] }