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": 76840.80428014703,
-      "required_pages_per_second_upper_estimate": 0.00531602513533079
+      "required_token_positions_per_second": 76784.11146894193,
+      "required_pages_per_second_upper_estimate": 0.005312102995106176
     },
     {
       "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": 9605.100535018379,
-      "required_pages_per_second_upper_estimate": 0.004021455016704618
+      "required_token_positions_per_second": 9598.013933617742,
+      "required_pages_per_second_upper_estimate": 0.004018488004683236
     },
     {
       "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": 19210.201070036757,
-      "required_pages_per_second_upper_estimate": 0.00400097827581859
+      "required_token_positions_per_second": 19196.027867235483,
+      "required_pages_per_second_upper_estimate": 0.003998026371447578
     },
     {
       "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": 28815.301605055138,
-      "required_pages_per_second_upper_estimate": 0.0017274333328050956
+      "required_token_positions_per_second": 28794.041800853225,
+      "required_pages_per_second_upper_estimate": 0.0017261588400050323
     },
     {
       "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": 28815.301605055138,
-      "required_pages_per_second_upper_estimate": 0.002240262954676113
+      "required_token_positions_per_second": 28794.041800853225,
+      "required_pages_per_second_upper_estimate": 0.0022386100984114103
     },
     {
       "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": 28815.301605055138,
-      "required_pages_per_second_upper_estimate": 0.3696117495293177
+      "required_token_positions_per_second": 28794.041800853225,
+      "required_pages_per_second_upper_estimate": 0.36933905158801483
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 704566,
-    "producer_batch_seconds": 0.07585416217968088,
+    "producer_batch_seconds": 0.07090073270312303,
     "producer_p95_seconds": 0.17503454999950918,
-    "gpu_update_seconds": 0.9594902173486064,
-    "producer_batches_per_second": 13.183192210748203,
-    "gpu_batches_per_second": 1.042220110136543,
-    "producer_headroom": 12.649143959639249,
+    "gpu_update_seconds": 0.9601986477348496,
+    "producer_batches_per_second": 14.104226597871987,
+    "gpu_batches_per_second": 1.041451164672064,
+    "producer_headroom": 13.542859306622582,
     "sustainable": true,
-    "buffer_seconds": 60.4478836929622,
+    "buffer_seconds": 60.49251480729553,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 357462.0739709278,
-    "upload_bytes_per_second": 4295212.239902175,
+    "generation_bytes_per_second": 356889.1483537111,
+    "upload_bytes_per_second": 4303635.813146845,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 76.70303978296161,
+    "estimated_full_shards_per_hour": 76.58010328253269,
     "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": [
@@ -502,6 +502,12 @@
       "bytes": 2702720,
       "pages": 37,
       "shards": 2
+    },
+    {
+      "seconds": 1.212294140999802,
+      "bytes": 5935208,
+      "pages": 36,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790868774355361490-de221e1d/train.jsonl

@@ -68,3 +68,4 @@
 {"step": 447000, "loss": 0.4998493194580078, "lr": 0.0003, "grad_norm": 0.1414208802714209, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 65427210081, "tokens_per_second": 152827.77021420054, "data_wait_s": 1.2985000466869678e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 447100, "loss": 0.6052762269973755, "lr": 0.0003, "grad_norm": 0.14999873744594025, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 65441963550, "tokens_per_second": 153995.48498493223, "data_wait_s": 1.2706000234175008e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 447200, "loss": 0.6125456094741821, "lr": 0.0003, "grad_norm": 0.18039655206486915, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 65456727001, "tokens_per_second": 153944.95189737467, "data_wait_s": 1.4642999303760007e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 447300, "loss": 0.563656747341156, "lr": 0.0003, "grad_norm": 0.1657343197335768, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 65471367768, "tokens_per_second": 152871.88170210554, "data_wait_s": 2.185199991799891e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1790868774.f92f6aaf-743a-4c24-a54a-850bd75954cb-8js8b.612.0.1790868774355361490-de221e1d

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:3518d841653e1b60776a0c50791d4e8683876e48e01b433f1daee3eb6db72181
-size 149208
+oid sha256:9a0bbaf3920b093b5b13ab27a491335dc6ee3ef91e36ac491e413958caf34425
+size 151326