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": 76402.4705577821,
-      "required_pages_per_second_upper_estimate": 0.005285700191343223
+      "required_token_positions_per_second": 77238.7251275425,
+      "required_pages_per_second_upper_estimate": 0.005343554222857181
     },
     {
       "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": 9550.308819722763,
-      "required_pages_per_second_upper_estimate": 0.003998514869691468
+      "required_token_positions_per_second": 9654.840640942812,
+      "required_pages_per_second_upper_estimate": 0.004042280160363653
     },
     {
       "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": 19100.617639445525,
-      "required_pages_per_second_upper_estimate": 0.003978154937135839
+      "required_token_positions_per_second": 19309.681281885623,
+      "required_pages_per_second_upper_estimate": 0.0040216973804771015
     },
     {
       "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": 28650.926459168288,
-      "required_pages_per_second_upper_estimate": 0.0017175792937954941
+      "required_token_positions_per_second": 28964.521922828437,
+      "required_pages_per_second_upper_estimate": 0.0017363788630093723
     },
     {
       "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": 28650.926459168288,
-      "required_pages_per_second_upper_estimate": 0.0022274835100932673
+      "required_token_positions_per_second": 28964.521922828437,
+      "required_pages_per_second_upper_estimate": 0.0022518641780321746
     },
     {
       "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": 28650.926459168288,
-      "required_pages_per_second_upper_estimate": 0.3675033216501621
+      "required_token_positions_per_second": 28964.521922828437,
+      "required_pages_per_second_upper_estimate": 0.3715257875454193
     }
   ],
   "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": 0.12254344599216438,
-    "producer_p95_seconds": 0.33139444600237766,
-    "gpu_update_seconds": 0.9649949728293218,
-    "producer_batches_per_second": 8.160371139423821,
-    "gpu_batches_per_second": 1.0362748285289456,
-    "producer_headroom": 7.874717125965472,
+    "producer_batch_seconds": 0.14394954231210022,
+    "producer_p95_seconds": 0.3518039959963062,
+    "gpu_update_seconds": 0.954547086040774,
+    "producer_batches_per_second": 6.946878634958614,
+    "gpu_batches_per_second": 1.0476172570467461,
+    "producer_headroom": 6.6311227580786545,
     "sustainable": true,
-    "buffer_seconds": 60.79468328824727,
+    "buffer_seconds": 60.13646642056876,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 84,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 491622.93879709806,
-    "upload_bytes_per_second": 5962331.855255548,
+    "generation_bytes_per_second": 492063.19810861495,
+    "upload_bytes_per_second": 5854573.908054152,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 105.49083826956469,
+    "estimated_full_shards_per_hour": 105.58530766910397,
     "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": 1.0825822449987754,
-      "bytes": 10431134,
-      "pages": 32,
-      "shards": 3
-    },
     {
       "seconds": 1.0676360670004215,
       "bytes": 12466956,
@@ -562,6 +556,12 @@
       "bytes": 6038977,
       "pages": 38,
       "shards": 2
+    },
+    {
+      "seconds": 1.0995818070005043,
+      "bytes": 2583346,
+      "pages": 34,
+      "shards": 2
     }
   ],
   "cache_key": {

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

@@ -325,3 +325,4 @@
 {"step": 517500, "loss": 0.6253323554992676, "lr": 0.0003, "grad_norm": 0.16925463581210798, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 75744017905, "tokens_per_second": 153063.1224845746, "data_wait_s": 1.756600249791518e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 517600, "loss": 0.5789720416069031, "lr": 0.0003, "grad_norm": 0.18343476660004074, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 75758639250, "tokens_per_second": 152493.2833605163, "data_wait_s": 9.866998880170286e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 517700, "loss": 0.567682683467865, "lr": 0.0003, "grad_norm": 0.15705088020636113, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 75773299753, "tokens_per_second": 152709.9012266064, "data_wait_s": 1.7655998817645013e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 517800, "loss": 0.5665090084075928, "lr": 0.0003, "grad_norm": 0.13725582150416302, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 75788025463, "tokens_per_second": 152683.27364554495, "data_wait_s": 1.7523001588415354e-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:dd91e5f69a4490cd8f7fa725c79a00905b83ab5a185d34452a47ed53682002f9
-size 697449
+oid sha256:cc7849a2d5ed9e86b22e899bae411d1538a8895496af254ea41eb3c9e1e8cad8
+size 699567