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": 76301.63082085963,
-      "required_pages_per_second_upper_estimate": 0.005278723864362501
+      "required_token_positions_per_second": 76536.61347537351,
+      "required_pages_per_second_upper_estimate": 0.005294980509636648
     },
     {
       "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": 9537.703852607454,
-      "required_pages_per_second_upper_estimate": 0.003993237433181933
+      "required_token_positions_per_second": 9567.076684421689,
+      "required_pages_per_second_upper_estimate": 0.004005535224488077
     },
     {
       "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": 19075.407705214908,
-      "required_pages_per_second_upper_estimate": 0.003972904372666275
+      "required_token_positions_per_second": 19134.153368843377,
+      "required_pages_per_second_upper_estimate": 0.003985139545172754
     },
     {
       "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": 28613.11155782236,
-      "required_pages_per_second_upper_estimate": 0.001715312348199817
+      "required_token_positions_per_second": 28701.23005326507,
+      "required_pages_per_second_upper_estimate": 0.0017205949174524554
     },
     {
       "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": 28613.11155782236,
-      "required_pages_per_second_upper_estimate": 0.0022245435678437944
+      "required_token_positions_per_second": 28701.23005326507,
+      "required_pages_per_second_upper_estimate": 0.0022313943932721655
     },
     {
       "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": 28613.11155782236,
-      "required_pages_per_second_upper_estimate": 0.3670182726981742
+      "required_token_positions_per_second": 28701.23005326507,
+      "required_pages_per_second_upper_estimate": 0.36814856214344444
     }
   ],
   "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": 0.35315888176555177,
-    "producer_p95_seconds": 0.2587006169997039,
-    "gpu_update_seconds": 0.966270303882469,
-    "producer_batches_per_second": 2.8315867209701397,
-    "gpu_batches_per_second": 1.0349071020624407,
-    "producer_headroom": 2.736078161341381,
+    "producer_batch_seconds": 0.07051779232031663,
+    "producer_p95_seconds": 0.2533027809986379,
+    "gpu_update_seconds": 0.9633036615047357,
+    "producer_batches_per_second": 14.180818302672439,
+    "gpu_batches_per_second": 1.038094258292284,
+    "producer_headroom": 13.66043419409773,
     "sustainable": true,
-    "buffer_seconds": 60.87502914459555,
+    "buffer_seconds": 60.688130674798344,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 79,
+    "upload_pages_per_commit": 80,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 441671.5583556982,
-    "upload_bytes_per_second": 5508944.008561795,
+    "generation_bytes_per_second": 444504.35773233767,
+    "upload_bytes_per_second": 5569695.701669307,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 94.77243483546457,
+    "estimated_full_shards_per_hour": 95.38028763749692,
     "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": 1.1697606620000442,
-      "bytes": 7536660,
-      "pages": 38,
-      "shards": 3
-    },
     {
       "seconds": 0.8494176380008867,
       "bytes": 947075,
@@ -562,6 +556,12 @@
       "bytes": 6252708,
       "pages": 25,
       "shards": 3
+    },
+    {
+      "seconds": 1.107486043998506,
+      "bytes": 11895475,
+      "pages": 41,
+      "shards": 3
     }
   ],
   "cache_key": {

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

@@ -216,3 +216,4 @@
 {"step": 506600, "loss": 0.5785301327705383, "lr": 0.0003, "grad_norm": 0.16547915430050822, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 74149480947, "tokens_per_second": 150849.97201876153, "data_wait_s": 8.549002814106643e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 506700, "loss": 0.5619561076164246, "lr": 0.0003, "grad_norm": 0.1554463108727053, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 74164155673, "tokens_per_second": 152544.85657892728, "data_wait_s": 1.4468998415395617e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 506800, "loss": 0.5715804100036621, "lr": 0.0003, "grad_norm": 0.1542340269721092, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 74178678963, "tokens_per_second": 150929.81368149302, "data_wait_s": 9.966999641619623e-06, "gpu_peak_gb": 89.37899398803711}
+{"step": 506900, "loss": 0.5570117235183716, "lr": 0.0003, "grad_norm": 0.15393902032819415, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 74193322151, "tokens_per_second": 152792.72910903775, "data_wait_s": 2.1479998395079747e-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 sha256:8fdd5d8e89dfc9de03f05e7ffc69e8b8d1bb79508582cd195dd53fcc3c10869f
-size 465021
+oid sha256:0c3b61aaa3a00d1afdf78b8576a84cc6a5685f33f3fa731a2202af0e89f242aa
+size 467139