Update TensorBoard training metrics

benchmark_cache/performance.json

@@ -49,11 +49,11 @@
   "selected": {
     "microbatch": 90,
     "mode": "compiled",
-    "microbatch_seconds": 0.9388940140000273,
+    "microbatch_seconds": 0.9371568949999869,
     "peak_gb": 89.37790203094482,
     "fits_budget": true,
-    "estimated_update_seconds": 0.9388940140000273,
-    "input_tokens_per_second": 196316.08813302612
+    "estimated_update_seconds": 0.9371568949999869,
+    "input_tokens_per_second": 196679.98067709096
   },
   "measurements": [
     {

benchmark_cache/shard_scheduler.json

@@ -22,8 +22,8 @@
       "tokens_per_second": 598315.063954628,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 75631.67090441682,
-      "required_pages_per_second_upper_estimate": 0.00523237448282181
+      "required_token_positions_per_second": 78671.99227083639,
+      "required_pages_per_second_upper_estimate": 0.005442710969468209
     },
     {
       "source": "rewrite6",
@@ -36,8 +36,8 @@
       "tokens_per_second": 8533.957213384612,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 9453.958863052103,
-      "required_pages_per_second_upper_estimate": 0.0039581751548493525
+      "required_token_positions_per_second": 9833.999033854549,
+      "required_pages_per_second_upper_estimate": 0.004117290038223121
     },
     {
       "source": "openmath",
@@ -50,8 +50,8 @@
       "tokens_per_second": 380144.35761751566,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 18907.917726104206,
-      "required_pages_per_second_upper_estimate": 0.003938020627025296
+      "required_token_positions_per_second": 19667.998067709097,
+      "required_pages_per_second_upper_estimate": 0.004096325317515041
     },
     {
       "source": "ultra_qa",
@@ -64,8 +64,8 @@
       "tokens_per_second": 43742.16556775435,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28361.87658915631,
-      "required_pages_per_second_upper_estimate": 0.0017002511954418724
+      "required_token_positions_per_second": 29501.997101563647,
+      "required_pages_per_second_upper_estimate": 0.001768599679297469
     },
     {
       "source": "ultra_style",
@@ -78,8 +78,8 @@
       "tokens_per_second": 22018.86276523721,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28361.87658915631,
-      "required_pages_per_second_upper_estimate": 0.0022050111540958487
+      "required_token_positions_per_second": 29501.997101563647,
+      "required_pages_per_second_upper_estimate": 0.002293650509075372
     },
     {
       "source": "cortex",
@@ -91,9 +91,9 @@
       "read_seconds": 0.04061180899998362,
       "tokens_per_second": 910315.184299418,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
-      "estimated_pages_for_buffer": 22,
-      "required_token_positions_per_second": 28361.87658915631,
-      "required_pages_per_second_upper_estimate": 0.3637957002752185
+      "estimated_pages_for_buffer": 24,
+      "required_token_positions_per_second": 29501.997101563647,
+      "required_pages_per_second_upper_estimate": 0.37841994204234997
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,415 +155,30 @@
   "shard_benchmark_scope": "Local packing/read measurements; upload capacity measured only on real data commits. Targets larger than available pages are sample-limited.",
   "estimated_fresh_generation_batch_seconds": 2.558590814491007,
   "selected": {
-    "prefetch_batches": 62,
+    "prefetch_batches": 65,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.09063524006239732,
-    "producer_p95_seconds": 0.32995927400043,
-    "gpu_update_seconds": 0.974829712451776,
-    "producer_batches_per_second": 11.033236071439273,
-    "gpu_batches_per_second": 1.0258201891332577,
-    "producer_headroom": 10.75552634693371,
-    "sustainable": true,
-    "buffer_seconds": 60.43944217201011,
-    "buffer_drain_seconds": null,
+    "producer_batch_seconds": 1.643433355500008,
+    "producer_p95_seconds": 3.2689711630000033,
+    "gpu_update_seconds": 0.9371568949999869,
+    "producer_batches_per_second": 0.6084822342526656,
+    "gpu_batches_per_second": 1.067057186833176,
+    "producer_headroom": 0.5702433213148826,
+    "sustainable": false,
+    "buffer_seconds": 60.91519817499915,
+    "buffer_drain_seconds": 141.7434590228479,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 84,
-    "max_pending_pages": 4096,
-    "generation_bytes_per_second": 364533.2995165233,
-    "upload_bytes_per_second": 5619729.502013892,
-    "upload_sustainable": true,
+    "upload_pages_per_commit": 6,
+    "max_pending_pages": 1445,
+    "generation_bytes_per_second": 0,
+    "upload_bytes_per_second": null,
+    "upload_sustainable": null,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 78.22036017534042,
+    "estimated_full_shards_per_hour": 0.0,
     "selection_policy": "Smallest shard within 90% of measured local peak, enlarged to amortize 30 seconds of measured page production; bounded to 256 MiB."
   },
-  "upload_measurements": [
-    {
-      "seconds": 1.149907068000175,
-      "bytes": 11779236,
-      "pages": 33,
-      "shards": 3
-    },
-    {
-      "seconds": 1.051513154998247,
-      "bytes": 2589054,
-      "pages": 34,
-      "shards": 2
-    },
-    {
-      "seconds": 1.0851055740058655,
-      "bytes": 5876732,
-      "pages": 32,
-      "shards": 2
-    },
-    {
-      "seconds": 1.0857899119946524,
-      "bytes": 1135012,
-      "pages": 33,
-      "shards": 1
-    },
-    {
-      "seconds": 1.3140387780003948,
-      "bytes": 11211496,
-      "pages": 39,
-      "shards": 4
-    },
-    {
-      "seconds": 1.1181462930035195,
-      "bytes": 5980396,
-      "pages": 35,
-      "shards": 2
-    },
-    {
-      "seconds": 1.134683780001069,
-      "bytes": 8428323,
-      "pages": 38,
-      "shards": 3
-    },
-    {
-      "seconds": 1.2588409719974152,
-      "bytes": 10473930,
-      "pages": 30,
-      "shards": 3
-    },
-    {
-      "seconds": 0.8643372480000835,
-      "bytes": 910531,
-      "pages": 27,
-      "shards": 1
-    },
-    {
-      "seconds": 1.064887002001342,
-      "bytes": 7934391,
-      "pages": 43,
-      "shards": 3
-    },
-    {
-      "seconds": 1.0906057760003023,
-      "bytes": 6005491,
-      "pages": 32,
-      "shards": 2
-    },
-    {
-      "seconds": 0.9286484479962382,
-      "bytes": 714826,
-      "pages": 21,
-      "shards": 1
-    },
-    {
-      "seconds": 1.5466609349969076,
-      "bytes": 4766956,
-      "pages": 1,
-      "shards": 1
-    },
-    {
-      "seconds": 1.2835849650000455,
-      "bytes": 14052046,
-      "pages": 50,
-      "shards": 4
-    },
-    {
-      "seconds": 1.258647842994833,
-      "bytes": 5993179,
-      "pages": 33,
-      "shards": 2
-    },
-    {
-      "seconds": 1.1118833660002565,
-      "bytes": 2568385,
-      "pages": 33,
-      "shards": 2
-    },
-    {
-      "seconds": 0.9927107309995336,
-      "bytes": 6005616,
-      "pages": 31,
-      "shards": 3
-    },
-    {
-      "seconds": 1.3781777679978404,
-      "bytes": 5044232,
-      "pages": 36,
-      "shards": 2
-    },
-    {
-      "seconds": 1.3652096680016257,
-      "bytes": 7517328,
-      "pages": 32,
-      "shards": 3
-    },
-    {
-      "seconds": 1.3108274569967762,
-      "bytes": 11956950,
-      "pages": 36,
-      "shards": 3
-    },
-    {
-      "seconds": 0.923527514001762,
-      "bytes": 789312,
-      "pages": 22,
-      "shards": 1
-    },
-    {
-      "seconds": 1.4157909589994233,
-      "bytes": 12468598,
-      "pages": 49,
-      "shards": 4
-    },
-    {
-      "seconds": 0.873544275003951,
-      "bytes": 984851,
-      "pages": 27,
-      "shards": 1
-    },
-    {
-      "seconds": 1.161854639001831,
-      "bytes": 7666603,
-      "pages": 40,
-      "shards": 3
-    },
-    {
-      "seconds": 0.9383765300008235,
-      "bytes": 5995522,
-      "pages": 27,
-      "shards": 2
-    },
-    {
-      "seconds": 1.2260548230042332,
-      "bytes": 10646786,
-      "pages": 36,
-      "shards": 3
-    },
-    {
-      "seconds": 1.1512752330017975,
-      "bytes": 7594320,
-      "pages": 40,
-      "shards": 3
-    },
-    {
-      "seconds": 1.0619875439952011,
-      "bytes": 5471838,
-      "pages": 28,
-      "shards": 2
-    },
-    {
-      "seconds": 1.1333524080037023,
-      "bytes": 7101020,
-      "pages": 41,
-      "shards": 3
-    },
-    {
-      "seconds": 1.235225882002851,
-      "bytes": 5723234,
-      "pages": 29,
-      "shards": 2
-    },
-    {
-      "seconds": 0.9995385679940227,
-      "bytes": 6053776,
-      "pages": 37,
-      "shards": 2
-    },
-    {
-      "seconds": 1.3147696169980918,
-      "bytes": 8376347,
-      "pages": 36,
-      "shards": 3
-    },
-    {
-      "seconds": 1.1205798119990504,
-      "bytes": 10286006,
-      "pages": 27,
-      "shards": 3
-    },
-    {
-      "seconds": 0.9068919589990401,
-      "bytes": 1229989,
-      "pages": 35,
-      "shards": 1
-    },
-    {
-      "seconds": 1.4636801879969425,
-      "bytes": 7357161,
-      "pages": 31,
-      "shards": 3
-    },
-    {
-      "seconds": 1.33853296199959,
-      "bytes": 6323168,
-      "pages": 37,
-      "shards": 2
-    },
-    {
-      "seconds": 1.1956154220024473,
-      "bytes": 6398380,
-      "pages": 33,
-      "shards": 3
-    },
-    {
-      "seconds": 1.3141280880008708,
-      "bytes": 12092914,
-      "pages": 40,
-      "shards": 3
-    },
-    {
-      "seconds": 0.8376461750012822,
-      "bytes": 875423,
-      "pages": 26,
-      "shards": 1
-    },
-    {
-      "seconds": 1.1449231579972547,
-      "bytes": 7698354,
-      "pages": 40,
-      "shards": 3
-    },
-    {
-      "seconds": 1.4822090849993401,
-      "bytes": 10438246,
-      "pages": 34,
-      "shards": 3
-    },
-    {
-      "seconds": 1.2825779239938129,
-      "bytes": 2656869,
-      "pages": 38,
-      "shards": 2
-    },
-    {
-      "seconds": 1.2434003369999118,
-      "bytes": 5829196,
-      "pages": 28,
-      "shards": 2
-    },
-    {
-      "seconds": 1.570511810001335,
-      "bytes": 12268120,
-      "pages": 37,
-      "shards": 3
-    },
-    {
-      "seconds": 1.4351412589967367,
-      "bytes": 7373378,
-      "pages": 40,
-      "shards": 3
-    },
-    {
-      "seconds": 1.1535751260016696,
-      "bytes": 5918763,
-      "pages": 29,
-      "shards": 2
-    },
-    {
-      "seconds": 1.0554045609969762,
-      "bytes": 2714106,
-      "pages": 38,
-      "shards": 2
-    },
-    {
-      "seconds": 1.0593990999987,
-      "bytes": 5830929,
-      "pages": 27,
-      "shards": 3
-    },
-    {
-      "seconds": 1.5048105800015037,
-      "bytes": 9859192,
-      "pages": 38,
-      "shards": 3
-    },
-    {
-      "seconds": 1.1245696650003083,
-      "bytes": 8444588,
-      "pages": 40,
-      "shards": 3
-    },
-    {
-      "seconds": 1.4192982609965838,
-      "bytes": 6049845,
-      "pages": 29,
-      "shards": 2
-    },
-    {
-      "seconds": 1.1915414559989586,
-      "bytes": 1026854,
-      "pages": 30,
-      "shards": 1
-    },
-    {
-      "seconds": 1.3134237790000043,
-      "bytes": 12455748,
-      "pages": 43,
-      "shards": 4
-    },
-    {
-      "seconds": 1.187473874000716,
-      "bytes": 6085388,
-      "pages": 30,
-      "shards": 2
-    },
-    {
-      "seconds": 0.9651436459971592,
-      "bytes": 2516489,
-      "pages": 34,
-      "shards": 2
-    },
-    {
-      "seconds": 1.3385733559989603,
-      "bytes": 11847980,
-      "pages": 36,
-      "shards": 3
-    },
-    {
-      "seconds": 1.301661803998286,
-      "bytes": 11947886,
-      "pages": 38,
-      "shards": 4
-    },
-    {
-      "seconds": 1.3009620130033,
-      "bytes": 995955,
-      "pages": 28,
-      "shards": 1
-    },
-    {
-      "seconds": 1.2097940460007521,
-      "bytes": 6272554,
-      "pages": 37,
-      "shards": 2
-    },
-    {
-      "seconds": 1.2908939679982723,
-      "bytes": 6343120,
-      "pages": 38,
-      "shards": 3
-    },
-    {
-      "seconds": 0.9974526400037576,
-      "bytes": 5745868,
-      "pages": 27,
-      "shards": 2
-    },
-    {
-      "seconds": 1.414747607006575,
-      "bytes": 11951303,
-      "pages": 37,
-      "shards": 3
-    },
-    {
-      "seconds": 1.082737166005245,
-      "bytes": 2735426,
-      "pages": 37,
-      "shards": 2
-    },
-    {
-      "seconds": 1.4271837970009074,
-      "bytes": 10546211,
-      "pages": 37,
-      "shards": 3
-    }
-  ],
+  "upload_measurements": [],
   "cache_key": {
     "version": 1,
     "kind": "data",
@@ -606,7 +221,7 @@
   },
   "cache_reused": true,
   "resume_probe_seconds": [
-    2.7244317560000013,
-    0.015869192000025123
+    3.2689711630000033,
+    0.017895548000012695
   ]
 }

runs/void-byte-v1/events.out.tfevents.1791209746.ebd20fd8-0561-42d8-bf2b-56d08e7106ef-6qqk2.329.0.1791209746770022608-f3fe5fb7

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:90ccd1a0934fddd691a0254fb4840efb4f195c58139fe7ac6e04d1306b3ffd92
+size 165