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": 76987.14752496383,
-      "required_pages_per_second_upper_estimate": 0.005326149500570327
+      "required_token_positions_per_second": 76398.9175778448,
+      "required_pages_per_second_upper_estimate": 0.005285454387947111
     },
     {
       "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": 9623.393440620479,
-      "required_pages_per_second_upper_estimate": 0.00402911387428091
+      "required_token_positions_per_second": 9549.8646972306,
+      "required_pages_per_second_upper_estimate": 0.003998328924878332
     },
     {
       "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": 19246.786881240958,
-      "required_pages_per_second_upper_estimate": 0.004008598135459701
+      "required_token_positions_per_second": 19099.7293944612,
+      "required_pages_per_second_upper_estimate": 0.003977969939130199
     },
     {
       "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": 28870.180321861433,
-      "required_pages_per_second_upper_estimate": 0.0017307232280826845
+      "required_token_positions_per_second": 28649.594091691797,
+      "required_pages_per_second_upper_estimate": 0.001717499420399686
     },
     {
       "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": 28870.180321861433,
-      "required_pages_per_second_upper_estimate": 0.0022445295335218377
+      "required_token_positions_per_second": 28649.594091691797,
+      "required_pages_per_second_upper_estimate": 0.0022273799243824354
     },
     {
       "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": 28870.180321861433,
-      "required_pages_per_second_upper_estimate": 0.37031567478433364
+      "required_token_positions_per_second": 28649.594091691797,
+      "required_pages_per_second_upper_estimate": 0.36748623147075843
     }
   ],
   "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": 720750,
-    "producer_batch_seconds": 0.11324974896098183,
-    "producer_p95_seconds": 0.3429523670056369,
-    "gpu_update_seconds": 0.9576663426332686,
-    "producer_batches_per_second": 8.830041648432546,
-    "gpu_batches_per_second": 1.0442050174284374,
-    "producer_headroom": 8.456233690753834,
+    "producer_batch_seconds": 0.13347179028960454,
+    "producer_p95_seconds": 0.356411078995734,
+    "gpu_update_seconds": 0.9650398505302994,
+    "producer_batches_per_second": 7.492219875302632,
+    "gpu_batches_per_second": 1.036226638154362,
+    "producer_headroom": 7.23029074860219,
     "sustainable": true,
-    "buffer_seconds": 60.33297958589592,
+    "buffer_seconds": 60.797510583408865,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 79,
+    "upload_pages_per_commit": 78,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 366781.59164030344,
-    "upload_bytes_per_second": 4105385.428279686,
+    "generation_bytes_per_second": 366868.7057661156,
+    "upload_bytes_per_second": 4126154.576234558,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 78.70279132754162,
+    "estimated_full_shards_per_hour": 78.72148399102785,
     "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": 1.5531271910003852,
-      "bytes": 10487346,
-      "pages": 37,
-      "shards": 3
-    },
     {
       "seconds": 1.5084796250012005,
       "bytes": 1063958,
@@ -562,6 +556,12 @@
       "bytes": 5494719,
       "pages": 25,
       "shards": 2
+    },
+    {
+      "seconds": 1.394116926996503,
+      "bytes": 11997464,
+      "pages": 39,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791300814123050567-72be911d/train.jsonl

@@ -384,3 +384,4 @@
 {"step": 838400, "loss": 0.5197517275810242, "lr": 3e-05, "grad_norm": 0.2894518852575219, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 122717666672, "tokens_per_second": 150872.9618805877, "data_wait_s": 1.113000325858593e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 838500, "loss": 0.5905106067657471, "lr": 3e-05, "grad_norm": 0.35634493423609104, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 122732311458, "tokens_per_second": 132077.6518711381, "data_wait_s": 7.971997547429055e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 838600, "loss": 0.5332279801368713, "lr": 3e-05, "grad_norm": 0.29592874082051485, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 122746902988, "tokens_per_second": 151017.26300918433, "data_wait_s": 7.921000360511243e-06, "gpu_peak_gb": 89.37899398803711}
+{"step": 838700, "loss": 0.5397185683250427, "lr": 3e-05, "grad_norm": 0.28648062397296886, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 122761678745, "tokens_per_second": 152947.86210918566, "data_wait_s": 9.016002877615392e-06, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791300814.0c9d201b-2dbc-40fb-b739-4ee2350a0f54-7b2tx.840.0.1791300814123050567-72be911d

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:9ced99bc2c869f703131889f1bc1cab7b9c4b909d5927931883bae18b6b31e3f
-size 823194
+oid sha256:b8eb3e433d2eb1659c8e35f8d1b08666cb159738a7bc3a4fade1659be905d995
+size 825312