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": 75361.11575349608,
-      "required_pages_per_second_upper_estimate": 0.006532110512511379
+      "required_token_positions_per_second": 75410.23828024937,
+      "required_pages_per_second_upper_estimate": 0.006536368328630448
     },
     {
       "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": 9420.13946918701,
-      "required_pages_per_second_upper_estimate": 0.0037578344778949296
+      "required_token_positions_per_second": 9426.27978503117,
+      "required_pages_per_second_upper_estimate": 0.0037602839416910686
     },
     {
       "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": 18840.27893837402,
-      "required_pages_per_second_upper_estimate": 0.004358366954808295
+      "required_token_positions_per_second": 18852.55957006234,
+      "required_pages_per_second_upper_estimate": 0.004361207862817652
     },
     {
       "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": 28260.41840756103,
-      "required_pages_per_second_upper_estimate": 0.0017209112100640654
+      "required_token_positions_per_second": 28278.83935509351,
+      "required_pages_per_second_upper_estimate": 0.0017220329491215512
     },
     {
       "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": 28260.41840756103,
-      "required_pages_per_second_upper_estimate": 0.002175565238859133
+      "required_token_positions_per_second": 28278.83935509351,
+      "required_pages_per_second_upper_estimate": 0.00217698333439263
     },
     {
       "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": 28260.41840756103,
-      "required_pages_per_second_upper_estimate": 0.35089981508574974
+      "required_token_positions_per_second": 28278.83935509351,
+      "required_pages_per_second_upper_estimate": 0.3511285416031577
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.09893946624217165,
+    "producer_batch_seconds": 0.22677589012499766,
     "producer_p95_seconds": 0.3258566270014853,
-    "gpu_update_seconds": 0.9783294642446916,
-    "producer_batches_per_second": 10.107190163653451,
-    "gpu_batches_per_second": 1.0221505500419932,
-    "producer_headroom": 9.888161937826299,
+    "gpu_update_seconds": 0.9776921765716001,
+    "producer_batches_per_second": 4.409639840676208,
+    "gpu_batches_per_second": 1.022816816952167,
+    "producer_headroom": 4.311270373727566,
     "sustainable": true,
-    "buffer_seconds": 60.65642678317088,
+    "buffer_seconds": 60.616914947439206,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 88,
+    "upload_pages_per_commit": 84,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 432833.62458750274,
-    "upload_bytes_per_second": 5688061.204038365,
+    "generation_bytes_per_second": 418533.95379857026,
+    "upload_bytes_per_second": 5684079.852398834,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 92.87601998537838,
+    "estimated_full_shards_per_hour": 89.80764351337271,
     "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": 1.1359487030003947,
-      "bytes": 6670820,
-      "pages": 29,
-      "shards": 3
-    },
     {
       "seconds": 0.9071375989988155,
       "bytes": 7422614,
@@ -562,6 +556,12 @@
       "bytes": 1102979,
       "pages": 34,
       "shards": 1
+    },
+    {
+      "seconds": 1.0738140280009247,
+      "bytes": 6022869,
+      "pages": 36,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790390190216476996-4d8dc439/train.jsonl

@@ -227,3 +227,4 @@
 {"step": 22700, "loss": 0.6697792634367943, "lr": 0.0003, "grad_norm": 0.13214199019849326, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 3321172444, "tokens_per_second": 148412.0297910993, "data_wait_s": 1.2770000466844067e-05, "gpu_peak_gb": 85.43514251708984}
 {"step": 22800, "loss": 0.6651651300489902, "lr": 0.0003, "grad_norm": 0.13975925019841318, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 3335845366, "tokens_per_second": 149251.86210631623, "data_wait_s": 1.0806001228047535e-05, "gpu_peak_gb": 85.43514251708984}
 {"step": 22900, "loss": 0.6729341298341751, "lr": 0.0003, "grad_norm": 0.13222387328376883, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 3350480304, "tokens_per_second": 148574.3674867066, "data_wait_s": 9.93300272966735e-06, "gpu_peak_gb": 85.43514251708984}
+{"step": 23000, "loss": 0.6521362792700529, "lr": 0.0003, "grad_norm": 0.13819179859861555, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 3365135354, "tokens_per_second": 149722.4295056493, "data_wait_s": 1.4397999620996416e-05, "gpu_peak_gb": 85.43514251708984}

runs/void-byte-v1/events.out.tfevents.1790390190.c8125a87-1345-4bfd-b42f-0e32ed8b7bb7-6w295.451.0.1790390190216476996-4d8dc439

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:41d00ad282ff0da0061f00031a309ca41a0bdc9c22e8bfc37d59899d2ca53d1c
-size 482545
+oid sha256:230bbc3481daac3ab6b8fd69eb442ce4d4f0a2ee2587d079774449f3e2e592ef
+size 484663