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": 77397.18008935252,
-      "required_pages_per_second_upper_estimate": 0.005354516504781364
+      "required_token_positions_per_second": 77328.52199913055,
+      "required_pages_per_second_upper_estimate": 0.005349766578791091
     },
     {
       "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": 9674.647511169065,
-      "required_pages_per_second_upper_estimate": 0.004050572883312901
+      "required_token_positions_per_second": 9666.065249891319,
+      "required_pages_per_second_upper_estimate": 0.004046979669733904
     },
     {
       "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": 19349.29502233813,
-      "required_pages_per_second_upper_estimate": 0.004029947877928771
+      "required_token_positions_per_second": 19332.130499782637,
+      "required_pages_per_second_upper_estimate": 0.004026372960539362
     },
     {
       "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": 29023.942533507197,
-      "required_pages_per_second_upper_estimate": 0.001739941037889545
+      "required_token_positions_per_second": 28998.195749673956,
+      "required_pages_per_second_upper_estimate": 0.0017383975575118048
     },
     {
       "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": 29023.942533507197,
-      "required_pages_per_second_upper_estimate": 0.002256483869148804
+      "required_token_positions_per_second": 28998.195749673956,
+      "required_pages_per_second_upper_estimate": 0.002254482169953903
     },
     {
       "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": 29023.942533507197,
-      "required_pages_per_second_upper_estimate": 0.37228797133832553
+      "required_token_positions_per_second": 28998.195749673956,
+      "required_pages_per_second_upper_estimate": 0.3719577192400554
     }
   ],
   "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": 0.0807314597967661,
-    "producer_p95_seconds": 0.21290857999701984,
-    "gpu_update_seconds": 0.9525928453062944,
-    "producer_batches_per_second": 12.38674492592363,
-    "gpu_batches_per_second": 1.0497664400140045,
-    "producer_headroom": 11.799524593068895,
+    "producer_batch_seconds": 0.06198457837501792,
+    "producer_p95_seconds": 0.21088658799999394,
+    "gpu_update_seconds": 0.9534386290330102,
+    "producer_batches_per_second": 16.13304512535068,
+    "gpu_batches_per_second": 1.0488352050663323,
+    "producer_headroom": 15.381868426442042,
     "sustainable": true,
-    "buffer_seconds": 60.01334925429655,
+    "buffer_seconds": 60.06663362907965,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 78,
+    "upload_pages_per_commit": 88,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 334596.88302656414,
-    "upload_bytes_per_second": 5689834.199822879,
+    "generation_bytes_per_second": 416408.40571232233,
+    "upload_bytes_per_second": 5620643.45869113,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 71.79670208070462,
+    "estimated_full_shards_per_hour": 89.35155037429097,
     "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": 1.1290311129996553,
-      "bytes": 7603249,
-      "pages": 38,
-      "shards": 3
-    },
     {
       "seconds": 1.4725222399993072,
       "bytes": 12055069,
@@ -562,6 +556,12 @@
       "bytes": 6195296,
       "pages": 36,
       "shards": 2
+    },
+    {
+      "seconds": 0.9275896610015479,
+      "bytes": 1143867,
+      "pages": 35,
+      "shards": 1
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790914434605211575-090606d4/train.jsonl

@@ -158,3 +158,4 @@
 {"step": 500800, "loss": 0.5830085873603821, "lr": 0.0003, "grad_norm": 0.1689505062686292, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 73300241182, "tokens_per_second": 153446.22490661117, "data_wait_s": 1.5358000382548198e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 500900, "loss": 0.5516468286514282, "lr": 0.0003, "grad_norm": 0.1395681611502911, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 73314884916, "tokens_per_second": 152671.84020874542, "data_wait_s": 1.1242998880334198e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 501000, "loss": 0.5778363943099976, "lr": 0.0003, "grad_norm": 0.15539351286475997, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 73329643855, "tokens_per_second": 154126.9569822981, "data_wait_s": 8.320999768329784e-06, "gpu_peak_gb": 89.37899398803711}
+{"step": 501100, "loss": 0.5862400531768799, "lr": 0.0003, "grad_norm": 0.17308679642894115, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 73344268038, "tokens_per_second": 151663.20512792928, "data_wait_s": 1.4340999769046903e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1790914434.4dec8a9d-3312-413a-9119-9058bec1adb7-bx5st.8214.0.1790914434605211575-090606d4

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:561c00617011c8c7d4bbfc7c2074caa78a1a34cd3fe0d4c03447aa847495129c
-size 341394
+oid sha256:d2ad39fbc64a49b7fa7fb6adc7aeba6962b92b89245e8e3d54cf5f447612327f
+size 343512