oellm-9b-256k-sft / eval /niah-hex-long.json
birgermoell's picture
Document tokenizer provenance and long-context retrieval evals
fc63b07 verified
Raw History Blame Contribute Delete
10.5 kB
{
"model": "/scratch/project_465002530/users/bmoell/qwen35-posttrain/output/oellm9b-256k-sft-bf16",
"rope_theta": 64000000.0,
"max_position_embeddings": 262144,
"requested_lengths": [
8192,
32768,
131072,
262144
],
"depths_percent": [
0,
25,
50,
75,
100
],
"max_new_tokens": 64,
"results": [
{
"requested_length": 8192,
"target_input_length": 8192,
"actual_input_tokens": 8158,
"depth_percent": 0,
"record_id": "record-8b002158",
"secret": "NIAH-B7622D5EE25D",
"filler_blocks": 122,
"needle_token_index": 52,
"needle_prompt_fraction": 0.006374,
"answer": "NIAH-B7622D5EE25D",
"strict_exact_match": true,
"contains_match": true,
"elapsed_seconds": 12.246
},
{
"requested_length": 8192,
"target_input_length": 8192,
"actual_input_tokens": 8154,
"depth_percent": 25,
"record_id": "record-7dedf730",
"secret": "NIAH-C745BD96A5D6",
"filler_blocks": 122,
"needle_token_index": 2030,
"needle_prompt_fraction": 0.248958,
"answer": "NIAH-C745BD96A5D6",
"strict_exact_match": true,
"contains_match": true,
"elapsed_seconds": 4.736
},
{
"requested_length": 8192,
"target_input_length": 8192,
"actual_input_tokens": 8153,
"depth_percent": 50,
"record_id": "record-42efa78b",
"secret": "NIAH-4859184CAD75",
"filler_blocks": 122,
"needle_token_index": 4076,
"needle_prompt_fraction": 0.499939,
"answer": "42efa78b",
"strict_exact_match": false,
"contains_match": false,
"elapsed_seconds": 2.6
},
{
"requested_length": 8192,
"target_input_length": 8192,
"actual_input_tokens": 8155,
"depth_percent": 75,
"record_id": "record-6f01d6ae",
"secret": "NIAH-3BF43AA1B869",
"filler_blocks": 122,
"needle_token_index": 6123,
"needle_prompt_fraction": 0.750828,
"answer": "6f01d6ae",
"strict_exact_match": false,
"contains_match": false,
"elapsed_seconds": 2.082
},
{
"requested_length": 8192,
"target_input_length": 8192,
"actual_input_tokens": 8154,
"depth_percent": 100,
"record_id": "record-c8da86af",
"secret": "NIAH-0A78C2D44DC2",
"filler_blocks": 122,
"needle_token_index": 8102,
"needle_prompt_fraction": 0.993623,
"answer": "N/A",
"strict_exact_match": false,
"contains_match": false,
"elapsed_seconds": 1.758
},
{
"requested_length": 32768,
"target_input_length": 32768,
"actual_input_tokens": 32707,
"depth_percent": 0,
"record_id": "record-ff160c03",
"secret": "NIAH-BCE52943FB70",
"filler_blocks": 494,
"needle_token_index": 51,
"needle_prompt_fraction": 0.001559,
"answer": "NIAH-BCE52943FB70",
"strict_exact_match": true,
"contains_match": true,
"elapsed_seconds": 10.211
},
{
"requested_length": 32768,
"target_input_length": 32768,
"actual_input_tokens": 32709,
"depth_percent": 25,
"record_id": "record-1855984e",
"secret": "NIAH-7AC46E9BB40A",
"filler_blocks": 494,
"needle_token_index": 8236,
"needle_prompt_fraction": 0.251796,
"answer": "The passkey for record-1855984e is NIAH-7AC46E9BB40A.",
"strict_exact_match": false,
"contains_match": true,
"elapsed_seconds": 12.821
},
{
"requested_length": 32768,
"target_input_length": 32768,
"actual_input_tokens": 32707,
"depth_percent": 50,
"record_id": "record-524f75bb",
"secret": "NIAH-DB7FD665F149",
"filler_blocks": 494,
"needle_token_index": 16353,
"needle_prompt_fraction": 0.499985,
"answer": "The passkey for record-524f75bb is NIAH-DB7FD665F149.",
"strict_exact_match": false,
"contains_match": true,
"elapsed_seconds": 12.891
},
{
"requested_length": 32768,
"target_input_length": 32768,
"actual_input_tokens": 32709,
"depth_percent": 75,
"record_id": "record-c7336bc6",
"secret": "NIAH-24A960D29254",
"filler_blocks": 494,
"needle_token_index": 24471,
"needle_prompt_fraction": 0.748143,
"answer": "The passkey for record-c7336bc6 is NIAH-24A960D29254.",
"strict_exact_match": false,
"contains_match": true,
"elapsed_seconds": 12.983
},
{
"requested_length": 32768,
"target_input_length": 32768,
"actual_input_tokens": 32705,
"depth_percent": 100,
"record_id": "record-93bacde9",
"secret": "NIAH-C905A6096888",
"filler_blocks": 494,
"needle_token_index": 32653,
"needle_prompt_fraction": 0.99841,
"answer": "NIAH-C905A6096888",
"strict_exact_match": true,
"contains_match": true,
"elapsed_seconds": 10.523
},
{
"requested_length": 131072,
"target_input_length": 131072,
"actual_input_tokens": 131044,
"depth_percent": 0,
"record_id": "record-275a8fbf",
"secret": "NIAH-F060DF9BE3BD",
"filler_blocks": 1984,
"needle_token_index": 50,
"needle_prompt_fraction": 0.000382,
"answer": "The passkey for record-275a8fbf is NIAH-F060DF9BE3BD.",
"strict_exact_match": false,
"contains_match": true,
"elapsed_seconds": 83.261
},
{
"requested_length": 131072,
"target_input_length": 131072,
"actual_input_tokens": 131048,
"depth_percent": 25,
"record_id": "record-77a65ae3",
"secret": "NIAH-51F1F58CC855",
"filler_blocks": 1984,
"needle_token_index": 32787,
"needle_prompt_fraction": 0.250191,
"answer": "The passkey for record-77a65ae3 is NIAH-51F1F58CC855.",
"strict_exact_match": false,
"contains_match": true,
"elapsed_seconds": 84.484
},
{
"requested_length": 131072,
"target_input_length": 131072,
"actual_input_tokens": 131045,
"depth_percent": 50,
"record_id": "record-ef276ebe",
"secret": "NIAH-7F94A934229F",
"filler_blocks": 1984,
"needle_token_index": 65521,
"needle_prompt_fraction": 0.499989,
"answer": "The passkey for record-ef276ebe is NIAH-7F94A934229F",
"strict_exact_match": false,
"contains_match": true,
"elapsed_seconds": 83.533
},
{
"requested_length": 131072,
"target_input_length": 131072,
"actual_input_tokens": 131047,
"depth_percent": 75,
"record_id": "record-4a3ef6f4",
"secret": "NIAH-CB433BE20187",
"filler_blocks": 1984,
"needle_token_index": 98259,
"needle_prompt_fraction": 0.7498,
"answer": "The passkey for record-4a3ef6f4 is NIAH-CB433BE20187.",
"strict_exact_match": false,
"contains_match": true,
"elapsed_seconds": 84.008
},
{
"requested_length": 131072,
"target_input_length": 131072,
"actual_input_tokens": 131045,
"depth_percent": 100,
"record_id": "record-311aacd4",
"secret": "NIAH-956C7D5E1ACA",
"filler_blocks": 1984,
"needle_token_index": 130994,
"needle_prompt_fraction": 0.999611,
"answer": "The passkey for record-311aacd4 is NIAH-956C7D5E1ACA.",
"strict_exact_match": false,
"contains_match": true,
"elapsed_seconds": 83.52
},
{
"requested_length": 262144,
"target_input_length": 262080,
"actual_input_tokens": 262060,
"depth_percent": 0,
"record_id": "record-28709c4e",
"secret": "NIAH-8D2349DA1B8B",
"filler_blocks": 3969,
"needle_token_index": 52,
"needle_prompt_fraction": 0.000198,
"answer": "The passkey for record-28709c4e is NIAH-8D2349DA1B8B.",
"strict_exact_match": false,
"contains_match": true,
"elapsed_seconds": 269.095
},
{
"requested_length": 262144,
"target_input_length": 262080,
"actual_input_tokens": 262059,
"depth_percent": 25,
"record_id": "record-524a19bf",
"secret": "NIAH-337054561569",
"filler_blocks": 3969,
"needle_token_index": 65523,
"needle_prompt_fraction": 0.250031,
"answer": "The passkey for record-524a19bf is NIAH-337054561569.",
"strict_exact_match": false,
"contains_match": true,
"elapsed_seconds": 269.044
},
{
"requested_length": 262144,
"target_input_length": 262080,
"actual_input_tokens": 262056,
"depth_percent": 50,
"record_id": "record-b52bade9",
"secret": "NIAH-7AFD87186345",
"filler_blocks": 3969,
"needle_token_index": 130994,
"needle_prompt_fraction": 0.49987,
"answer": "The passkey for record-b52bade9 is NIAH-7AFD87186345.",
"strict_exact_match": false,
"contains_match": true,
"elapsed_seconds": 267.341
},
{
"requested_length": 262144,
"target_input_length": 262080,
"actual_input_tokens": 262061,
"depth_percent": 75,
"record_id": "record-3e633807",
"secret": "NIAH-E7814D0E171D",
"filler_blocks": 3969,
"needle_token_index": 196534,
"needle_prompt_fraction": 0.749955,
"answer": "The passkey for record-3e633807 is NIAH-E7814D0E171D.",
"strict_exact_match": false,
"contains_match": true,
"elapsed_seconds": 269.949
},
{
"requested_length": 262144,
"target_input_length": 262080,
"actual_input_tokens": 262057,
"depth_percent": 100,
"record_id": "record-f3bbed67",
"secret": "NIAH-B147F7228673",
"filler_blocks": 3969,
"needle_token_index": 262004,
"needle_prompt_fraction": 0.999798,
"answer": "The passkey for record-f3bbed67 is NIAH-B147F7228673.",
"strict_exact_match": false,
"contains_match": true,
"elapsed_seconds": 268.271
}
],
"strict_exact_match_grid": {
"8192": {
"0": true,
"25": true,
"50": false,
"75": false,
"100": false
},
"32768": {
"0": true,
"25": false,
"50": false,
"75": false,
"100": true
},
"131072": {
"0": false,
"25": false,
"50": false,
"75": false,
"100": false
},
"262144": {
"0": false,
"25": false,
"50": false,
"75": false,
"100": false
}
}
}