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": 76679.688537802,
-      "required_pages_per_second_upper_estimate": 0.005304878774435333
+      "required_token_positions_per_second": 76619.31638446944,
+      "required_pages_per_second_upper_estimate": 0.005300702088785092
     },
     {
       "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": 9584.96106722525,
-      "required_pages_per_second_upper_estimate": 0.004013023042099545
+      "required_token_positions_per_second": 9577.41454805868,
+      "required_pages_per_second_upper_estimate": 0.004009863472113743
     },
     {
       "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": 19169.9221344505,
-      "required_pages_per_second_upper_estimate": 0.003992589235762935
+      "required_token_positions_per_second": 19154.82909611736,
+      "required_pages_per_second_upper_estimate": 0.003989445753908329
     },
     {
       "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": 28754.883201675748,
-      "required_pages_per_second_upper_estimate": 0.0017238113417760615
+      "required_token_positions_per_second": 28732.24364417604,
+      "required_pages_per_second_upper_estimate": 0.0017224541348725567
     },
     {
       "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": 28754.883201675748,
-      "required_pages_per_second_upper_estimate": 0.0022355656895658364
+      "required_token_positions_per_second": 28732.24364417604,
+      "required_pages_per_second_upper_estimate": 0.0022338055635511235
     },
     {
       "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": 28754.883201675748,
-      "required_pages_per_second_upper_estimate": 0.3688367671229942
+      "required_token_positions_per_second": 28732.24364417604,
+      "required_pages_per_second_upper_estimate": 0.36854637118785083
     }
   ],
   "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": 0.0587892256952216,
-    "producer_p95_seconds": 0.20206039500044426,
-    "gpu_update_seconds": 0.9615062529062458,
-    "producer_batches_per_second": 17.009919558802423,
-    "gpu_batches_per_second": 1.040034838023573,
-    "producer_headroom": 16.35514401722078,
+    "producer_batch_seconds": 0.07736619592174065,
+    "producer_p95_seconds": 0.2163933880001423,
+    "gpu_update_seconds": 0.9622638712937472,
+    "producer_batches_per_second": 12.925541809132563,
+    "gpu_batches_per_second": 1.039215988287617,
+    "producer_headroom": 12.437781899825085,
     "sustainable": true,
-    "buffer_seconds": 60.574893933093485,
+    "buffer_seconds": 60.62262389150607,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 84,
+    "upload_pages_per_commit": 85,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 319548.94790189626,
-    "upload_bytes_per_second": 5360548.999982087,
+    "generation_bytes_per_second": 320114.27023038676,
+    "upload_bytes_per_second": 5321536.556669298,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 68.56776550095239,
+    "estimated_full_shards_per_hour": 68.68907051261618,
     "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": 1.1272267000003922,
-      "bytes": 8136270,
-      "pages": 29,
-      "shards": 3
-    },
     {
       "seconds": 1.007810305000021,
       "bytes": 6126457,
@@ -562,6 +556,12 @@
       "bytes": 5730323,
       "pages": 27,
       "shards": 2
+    },
+    {
+      "seconds": 1.3926623539991851,
+      "bytes": 6408754,
+      "pages": 36,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790989041640695387-36510d53/train.jsonl

@@ -174,3 +174,4 @@
 {"step": 574500, "loss": 0.5851013660430908, "lr": 0.0003, "grad_norm": 0.20274901251846236, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 84089482779, "tokens_per_second": 135131.29057012842, "data_wait_s": 1.2450000212993473e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 574600, "loss": 0.5650253295898438, "lr": 0.0003, "grad_norm": 0.1595260223471998, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 84104123104, "tokens_per_second": 153091.42426447838, "data_wait_s": 2.3594999220222235e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 574700, "loss": 0.6336092352867126, "lr": 0.0003, "grad_norm": 0.15673328562349978, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 84118631127, "tokens_per_second": 151586.2848648544, "data_wait_s": 9.937000868376344e-06, "gpu_peak_gb": 89.37899398803711}
+{"step": 574800, "loss": 0.7473382353782654, "lr": 0.0003, "grad_norm": 0.3193019051904646, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 84133290730, "tokens_per_second": 152766.53100402327, "data_wait_s": 1.2799999240087345e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1790989041.fb9d4967-10f3-4fa9-9622-1d5a28f8589b-vh8db.427.0.1790989041640695387-36510d53

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:d40bc9c59264e1afdc97f02645acebf176a037f3e8ca487a1bc41c271207107d
-size 375282
+oid sha256:b4a01d5ea7dce57a78551f4bd68a15ecd1403bf428a40a664e7e1fcdcf6dfc80
+size 377400