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": 77135.57706257077,
-      "required_pages_per_second_upper_estimate": 0.005336418200385952
+      "required_token_positions_per_second": 77239.26463336572,
+      "required_pages_per_second_upper_estimate": 0.005343591547121857
     },
     {
       "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": 9641.947132821346,
-      "required_pages_per_second_upper_estimate": 0.004036881917760262
+      "required_token_positions_per_second": 9654.908079170715,
+      "required_pages_per_second_upper_estimate": 0.004042308395341405
     },
     {
       "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": 19283.89426564269,
-      "required_pages_per_second_upper_estimate": 0.004016326625043048
+      "required_token_positions_per_second": 19309.81615834143,
+      "required_pages_per_second_upper_estimate": 0.004021725471685915
     },
     {
       "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": 28925.841398464036,
-      "required_pages_per_second_upper_estimate": 0.0017340600246354674
+      "required_token_positions_per_second": 28964.724237512146,
+      "required_pages_per_second_upper_estimate": 0.0017363909914657441
     },
     {
       "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": 28925.841398464036,
-      "required_pages_per_second_upper_estimate": 0.0022488569374004884
+      "required_token_positions_per_second": 28964.724237512146,
+      "required_pages_per_second_upper_estimate": 0.002251879907108945
     },
     {
       "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": 28925.841398464036,
-      "required_pages_per_second_upper_estimate": 0.37102963531078403
+      "required_token_positions_per_second": 28964.724237512146,
+      "required_pages_per_second_upper_estimate": 0.3715283826209534
     }
   ],
   "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": 0.5049835351328156,
-    "producer_p95_seconds": 0.15728663800018694,
-    "gpu_update_seconds": 0.9558235357491834,
-    "producer_batches_per_second": 1.980262583684021,
-    "gpu_batches_per_second": 1.0462182218773162,
-    "producer_headroom": 1.8927815844486742,
+    "producer_batch_seconds": 0.06433194296095834,
+    "producer_p95_seconds": 0.15103475400019306,
+    "gpu_update_seconds": 0.9545404186584018,
+    "producer_batches_per_second": 15.54437739595209,
+    "gpu_batches_per_second": 1.0476245745627946,
+    "producer_headroom": 14.837736507316304,
     "sustainable": true,
-    "buffer_seconds": 60.21688275219856,
+    "buffer_seconds": 60.13604637547931,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 78,
+    "upload_pages_per_commit": 84,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 266019.63741450995,
-    "upload_bytes_per_second": 5382510.268907458,
+    "generation_bytes_per_second": 270210.9501805142,
+    "upload_bytes_per_second": 5484788.469311288,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 57.081621568932285,
+    "estimated_full_shards_per_hour": 57.98097971974916,
     "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": [
@@ -346,6 +346,12 @@
       "bytes": 5490081,
       "pages": 21,
       "shards": 2
+    },
+    {
+      "seconds": 1.3486151890001565,
+      "bytes": 10508121,
+      "pages": 37,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790824801708454555-41033672/train.jsonl

@@ -31,3 +31,4 @@
 {"step": 400200, "loss": 0.5335736274719238, "lr": 0.0003, "grad_norm": 0.1470821112963024, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 58578299903, "tokens_per_second": 149474.68890538777, "data_wait_s": 8.923999757826095e-06, "gpu_peak_gb": 89.37861251831055}
 {"step": 400300, "loss": 0.5574657917022705, "lr": 0.0003, "grad_norm": 0.15063563789242532, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 58592754920, "tokens_per_second": 125612.52985703979, "data_wait_s": 7.310000000870787e-06, "gpu_peak_gb": 89.37861251831055}
 {"step": 400400, "loss": 0.6056730151176453, "lr": 0.0003, "grad_norm": 0.16038392755760572, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 58607306918, "tokens_per_second": 151749.90916819338, "data_wait_s": 6.699000095977681e-06, "gpu_peak_gb": 89.37861251831055}
+{"step": 400500, "loss": 0.5657328367233276, "lr": 0.0003, "grad_norm": 0.1569044007497574, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 58621964614, "tokens_per_second": 152573.03605078027, "data_wait_s": 9.83200015980401e-06, "gpu_peak_gb": 89.37861251831055}

runs/void-byte-v1/events.out.tfevents.1790824801.afc92292-ba09-4e51-8d82-bc1983323308-sdcdh.469.0.1790824801708454555-41033672

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:66eb1ff6e4254ac69ad8d7a138d1ff80db79d06e66f77bc201d0b4be40271918
-size 70842
+oid sha256:588f05ebfbba3cf97180000b76978e44ba227d9021db09d5e7fa277c84a943e6
+size 72960