Update TensorBoard training metrics

benchmark_cache/shard_scheduler.json

@@ -22,8 +22,8 @@
       "tokens_per_second": 1062967.6168812835,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 76091.83360235163,
-      "required_pages_per_second_upper_estimate": 0.006595447283662722
+      "required_token_positions_per_second": 76035.71264076048,
+      "required_pages_per_second_upper_estimate": 0.006590582860949278
     },
     {
       "source": "rewrite6",
@@ -36,8 +36,8 @@
       "tokens_per_second": 1206672.263377472,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 9511.479200293954,
-      "required_pages_per_second_upper_estimate": 0.0037942712622841684
+      "required_token_positions_per_second": 9504.46408009506,
+      "required_pages_per_second_upper_estimate": 0.0037914728259514363
     },
     {
       "source": "openmath",
@@ -50,8 +50,8 @@
       "tokens_per_second": 410042.7581134458,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 19022.958400587908,
-      "required_pages_per_second_upper_estimate": 0.00440062663334275
+      "required_token_positions_per_second": 19008.92816019012,
+      "required_pages_per_second_upper_estimate": 0.00439738098415052
     },
     {
       "source": "ultra_qa",
@@ -64,8 +64,8 @@
       "tokens_per_second": 2163486.5624068812,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28534.437600881858,
-      "required_pages_per_second_upper_estimate": 0.0017375975412697055
+      "required_token_positions_per_second": 28513.392240285182,
+      "required_pages_per_second_upper_estimate": 0.0017363159892258414
     },
     {
       "source": "ultra_style",
@@ -78,8 +78,8 @@
       "tokens_per_second": 1546694.2737633225,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28534.437600881858,
-      "required_pages_per_second_upper_estimate": 0.002196659994894646
+      "required_token_positions_per_second": 28513.392240285182,
+      "required_pages_per_second_upper_estimate": 0.002195039864778627
     },
     {
       "source": "cortex",
@@ -92,8 +92,8 @@
       "tokens_per_second": 1015177.8395898147,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 22,
-      "required_token_positions_per_second": 28534.437600881858,
-      "required_pages_per_second_upper_estimate": 0.35430221638354864
+      "required_token_positions_per_second": 28513.392240285182,
+      "required_pages_per_second_upper_estimate": 0.35404090343922895
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.3702038395390943,
-    "producer_p95_seconds": 0.19250512600046932,
-    "gpu_update_seconds": 0.9689344639175764,
-    "producer_batches_per_second": 2.7012145558646967,
-    "gpu_batches_per_second": 1.0320615451707849,
-    "producer_headroom": 2.6172998776131142,
+    "producer_batch_seconds": 0.3244580188672046,
+    "producer_p95_seconds": 0.1937449770002786,
+    "gpu_update_seconds": 0.9696496217288902,
+    "producer_batches_per_second": 3.0820628304744835,
+    "gpu_batches_per_second": 1.0313003559130924,
+    "producer_headroom": 2.9885210577142556,
     "sustainable": true,
-    "buffer_seconds": 60.07393676288974,
+    "buffer_seconds": 60.11827654719119,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 348897.1178558215,
-    "upload_bytes_per_second": 5023652.371054542,
+    "generation_bytes_per_second": 328950.2923228368,
+    "upload_bytes_per_second": 5276775.385825173,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 74.86519958263382,
+    "estimated_full_shards_per_hour": 70.58507516158895,
     "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": 860910,
       "pages": 25,
       "shards": 1
+    },
+    {
+      "seconds": 1.6520775649996722,
+      "bytes": 17392082,
+      "pages": 43,
+      "shards": 5
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790563279934888633-82973089/train.jsonl

@@ -62,3 +62,4 @@
 {"step": 161200, "loss": 0.5865148399025202, "lr": 0.0003, "grad_norm": 0.13777146085472525, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 23592569679, "tokens_per_second": 147987.39522450994, "data_wait_s": 1.0426000699226279e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 161300, "loss": 0.6452075373381376, "lr": 0.0003, "grad_norm": 0.14168178188903088, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 23607329339, "tokens_per_second": 151364.45356469764, "data_wait_s": 1.2057000276399776e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 161400, "loss": 0.5813808608800173, "lr": 0.0003, "grad_norm": 0.16874280129162456, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 23621895067, "tokens_per_second": 148976.63425909603, "data_wait_s": 1.3281000065035187e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 161500, "loss": 0.5784920454025269, "lr": 0.0003, "grad_norm": 0.1443170699606449, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 23636375426, "tokens_per_second": 148385.01331216842, "data_wait_s": 1.0714999916672241e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790563279.bdfa290c-aca9-4e96-a5d2-8425921a3278-d85hq.1304.0.1790563279934888633-82973089

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:7a6f76b0d5d27254571332fd3e0ddaf54f7af425f1f30638bcc59dc03d803daf
-size 136500
+oid sha256:f0130442ad873f7988fbe47a1127dd5d65660ee9b241a70ad194603acf60f50f
+size 138618