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": 75410.11962323569,
-      "required_pages_per_second_upper_estimate": 0.006536358043741236
+      "required_token_positions_per_second": 75751.90664772637,
+      "required_pages_per_second_upper_estimate": 0.006565983276772794
     },
     {
       "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": 9426.264952904461,
-      "required_pages_per_second_upper_estimate": 0.003760278024933964
+      "required_token_positions_per_second": 9468.988330965796,
+      "required_pages_per_second_upper_estimate": 0.0037773210192140563
     },
     {
       "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": 18852.529905808922,
-      "required_pages_per_second_upper_estimate": 0.004361201000514696
+      "required_token_positions_per_second": 18937.976661931592,
+      "required_pages_per_second_upper_estimate": 0.00438096760373213
     },
     {
       "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": 28278.794858713383,
-      "required_pages_per_second_upper_estimate": 0.001722030239525456
+      "required_token_positions_per_second": 28406.96499289739,
+      "required_pages_per_second_upper_estimate": 0.0017298351282405379
     },
     {
       "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": 28278.794858713383,
-      "required_pages_per_second_upper_estimate": 0.002176979908938114
+      "required_token_positions_per_second": 28406.96499289739,
+      "required_pages_per_second_upper_estimate": 0.002186846800665238
     },
     {
       "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": 28278.794858713383,
-      "required_pages_per_second_upper_estimate": 0.3511279891070363
+      "required_token_positions_per_second": 28406.96499289739,
+      "required_pages_per_second_upper_estimate": 0.3527194332157566
     }
   ],
   "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": 0.04961033224998346,
-    "producer_p95_seconds": 0.17037309699935577,
-    "gpu_update_seconds": 0.9776937149597441,
-    "producer_batches_per_second": 20.15709136881125,
-    "gpu_batches_per_second": 1.022815207563418,
-    "producer_headroom": 19.707461543156064,
+    "producer_batch_seconds": 0.0667450819140214,
+    "producer_p95_seconds": 0.16825275900009729,
+    "gpu_update_seconds": 0.973282432914352,
+    "producer_batches_per_second": 14.982377297673615,
+    "gpu_batches_per_second": 1.0274509907731983,
+    "producer_headroom": 14.582084627120532,
     "sustainable": true,
-    "buffer_seconds": 60.61701032750413,
+    "buffer_seconds": 60.343510840689824,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 78,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 399136.7533068909,
-    "upload_bytes_per_second": 4853717.286250679,
+    "generation_bytes_per_second": 420607.24743669183,
+    "upload_bytes_per_second": 4699819.4940735195,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 85.64545583157583,
+    "estimated_full_shards_per_hour": 90.25252406430785,
     "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": [
@@ -250,6 +250,12 @@
       "bytes": 12398330,
       "pages": 38,
       "shards": 4
+    },
+    {
+      "seconds": 0.7755766290001702,
+      "bytes": 1144894,
+      "pages": 33,
+      "shards": 1
     }
   ],
   "cache_key": {

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

@@ -43,3 +43,4 @@
 {"step": 159300, "loss": 0.5889497511088848, "lr": 0.0003, "grad_norm": 0.14231023699674406, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 23314483323, "tokens_per_second": 149607.7924737093, "data_wait_s": 1.3675999980478082e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 159400, "loss": 0.5490333624184132, "lr": 0.0003, "grad_norm": 0.14879176756412074, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 23329089121, "tokens_per_second": 148774.9649715777, "data_wait_s": 1.605200031917775e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 159500, "loss": 0.5589956678450108, "lr": 0.0003, "grad_norm": 0.14677640845151912, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 23343720107, "tokens_per_second": 149320.57667826503, "data_wait_s": 1.1400999937904999e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 159600, "loss": 0.6243120767176151, "lr": 0.0003, "grad_norm": 0.15560457008791292, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 23358465314, "tokens_per_second": 150561.28663703115, "data_wait_s": 1.5671000255679246e-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 sha256:62db26e72ffc3de219045017df18b84d24bc60cb17c80d1e27b9109fb3ddbdc5
-size 95475
+oid sha256:6a02545280276a7f2bfa051ccef58c26f34cc2af3a571ae9bec6d6cb8389c08d
+size 97593