Update TensorBoard training metrics

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": 76747.7266955817,
-      "required_pages_per_second_upper_estimate": 0.0053095858120608104
+      "required_token_positions_per_second": 76816.24975895784,
+      "required_pages_per_second_upper_estimate": 0.005314326396580584
     },
     {
       "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": 9593.465836947713,
-      "required_pages_per_second_upper_estimate": 0.0040165838115825536
+      "required_token_positions_per_second": 9602.03121986973,
+      "required_pages_per_second_upper_estimate": 0.0040201699585464675
     },
     {
       "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": 19186.931673895426,
-      "required_pages_per_second_upper_estimate": 0.003996131874257593
+      "required_token_positions_per_second": 19204.06243973946,
+      "required_pages_per_second_upper_estimate": 0.003999699761014231
     },
     {
       "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": 28780.397510843137,
-      "required_pages_per_second_upper_estimate": 0.001725340885652552
+      "required_token_positions_per_second": 28806.09365960919,
+      "required_pages_per_second_upper_estimate": 0.0017268813305353296
     },
     {
       "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": 28780.397510843137,
-      "required_pages_per_second_upper_estimate": 0.0022375493148780153
+      "required_token_positions_per_second": 28806.09365960919,
+      "required_pages_per_second_upper_estimate": 0.0022395470774191624
     },
     {
       "source": "cortex",
@@ -92,8 +92,8 @@
       "tokens_per_second": 910315.184299418,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 23,
-      "required_token_positions_per_second": 28780.397510843137,
-      "required_pages_per_second_upper_estimate": 0.3691640372858626
+      "required_token_positions_per_second": 28806.09365960919,
+      "required_pages_per_second_upper_estimate": 0.3694936398918586
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 704566,
-    "producer_batch_seconds": 0.08302896245308489,
-    "producer_p95_seconds": 0.2205657099984819,
-    "gpu_update_seconds": 0.9606538613507162,
-    "producer_batches_per_second": 12.043990078342183,
-    "gpu_batches_per_second": 1.0409576645993612,
-    "producer_headroom": 11.570105574829132,
+    "producer_batch_seconds": 0.06424719368746423,
+    "producer_p95_seconds": 0.21371257299688295,
+    "gpu_update_seconds": 0.959796920981583,
+    "producer_batches_per_second": 15.564882177805032,
+    "gpu_batches_per_second": 1.041887068128226,
+    "producer_headroom": 14.939125989698386,
     "sustainable": true,
-    "buffer_seconds": 60.52119326509512,
+    "buffer_seconds": 60.46720602183973,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 82,
+    "upload_pages_per_commit": 84,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 329161.4894213708,
-    "upload_bytes_per_second": 5482438.459894346,
+    "generation_bytes_per_second": 328038.00648047135,
+    "upload_bytes_per_second": 5544108.7807667395,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 70.63039314251749,
+    "estimated_full_shards_per_hour": 70.38931985674482,
     "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": 1.1323460120001982,
-      "bytes": 7094094,
-      "pages": 25,
-      "shards": 3
-    },
     {
       "seconds": 1.3368760009998368,
       "bytes": 10726853,
@@ -562,6 +556,12 @@
       "bytes": 2234803,
       "pages": 29,
       "shards": 2
+    },
+    {
+      "seconds": 0.9967152519966476,
+      "bytes": 11168227,
+      "pages": 38,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790956870018829698-ff939fa5/train.jsonl

@@ -164,3 +164,4 @@
 {"step": 541400, "loss": 0.5476915240287781, "lr": 0.0003, "grad_norm": 0.1653532639298301, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 79243004314, "tokens_per_second": 154845.0489894277, "data_wait_s": 1.3023000065004453e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 541500, "loss": 0.5443552732467651, "lr": 0.0003, "grad_norm": 0.15322251732361983, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 79257695084, "tokens_per_second": 138901.01671732092, "data_wait_s": 1.4998997357906774e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 541600, "loss": 0.5914584994316101, "lr": 0.0003, "grad_norm": 0.16577160816929673, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 79272297585, "tokens_per_second": 152755.2319532559, "data_wait_s": 1.001299824565649e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 541700, "loss": 0.5614579916000366, "lr": 0.0003, "grad_norm": 0.17160825678431055, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 79287086550, "tokens_per_second": 154765.4882797034, "data_wait_s": 1.0728002962423488e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1790956870.abdf5fd7-18c6-4e42-b51e-c994b21d5ba1-dh6p5.3284.0.1790956870018829698-ff939fa5

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:9fcc54276ef96af08e2307ea1fc991600a6dbc0c5f5677a05474d10f446d6b08
-size 354102
+oid sha256:f1f3d52e7a814164f8178ed37cfdef45f8a10998e1b475e09bad86852b9fde84
+size 356220