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": 75644.44763824537,
-      "required_pages_per_second_upper_estimate": 0.006556669002183351
+      "required_token_positions_per_second": 76034.35054523053,
+      "required_pages_per_second_upper_estimate": 0.006590464797960941
     },
     {
       "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": 9455.555954780672,
-      "required_pages_per_second_upper_estimate": 0.0037719626435218892
+      "required_token_positions_per_second": 9504.293818153816,
+      "required_pages_per_second_upper_estimate": 0.003791404905917431
     },
     {
       "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": 18911.111909561343,
-      "required_pages_per_second_upper_estimate": 0.004374752916074766
+      "required_token_positions_per_second": 19008.58763630763,
+      "required_pages_per_second_upper_estimate": 0.004397302209943322
     },
     {
       "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": 28366.667864342013,
-      "required_pages_per_second_upper_estimate": 0.0017273812445342096
+      "required_token_positions_per_second": 28512.881454461447,
+      "required_pages_per_second_upper_estimate": 0.001736284885049062
     },
     {
       "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": 28366.667864342013,
-      "required_pages_per_second_upper_estimate": 0.002183744616159454
+      "required_token_positions_per_second": 28512.881454461447,
+      "required_pages_per_second_upper_estimate": 0.00219500054307197
     },
     {
       "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": 28366.667864342013,
-      "required_pages_per_second_upper_estimate": 0.35221907774491246
+      "required_token_positions_per_second": 28512.881454461447,
+      "required_pages_per_second_upper_estimate": 0.35403456118878834
     }
   ],
   "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": 717798,
-    "producer_batch_seconds": 0.13109138810960985,
-    "producer_p95_seconds": 0.34873073600101634,
-    "gpu_update_seconds": 0.9746650587309408,
-    "producer_batches_per_second": 7.6282661616479865,
-    "gpu_batches_per_second": 1.0259934846767222,
-    "producer_headroom": 7.435004486457883,
+    "producer_batch_seconds": 0.10199884532056558,
+    "producer_p95_seconds": 0.3446155480014568,
+    "gpu_update_seconds": 0.9696669922384811,
+    "producer_batches_per_second": 9.80403255406632,
+    "gpu_batches_per_second": 1.0312818813100928,
+    "producer_headroom": 9.50664675850964,
     "sustainable": true,
-    "buffer_seconds": 60.429233641318326,
+    "buffer_seconds": 60.11935351878583,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 78,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 421700.3490654072,
-    "upload_bytes_per_second": 5174887.657271451,
+    "generation_bytes_per_second": 440038.247404976,
+    "upload_bytes_per_second": 5331446.876716542,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 90.48707822772657,
+    "estimated_full_shards_per_hour": 94.42196432697259,
     "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": 4.179794984000182,
-      "bytes": 6264644,
-      "pages": 36,
-      "shards": 2
-    },
     {
       "seconds": 0.9406074870021257,
       "bytes": 762678,
@@ -562,6 +556,12 @@
       "bytes": 11716150,
       "pages": 31,
       "shards": 4
+    },
+    {
+      "seconds": 0.8494216189974395,
+      "bytes": 1181935,
+      "pages": 35,
+      "shards": 1
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790782192342215557-281f6532/train.jsonl

@@ -305,3 +305,4 @@
 {"step": 385500, "loss": 0.5525905806571245, "lr": 0.0003, "grad_norm": 0.15506915940897897, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 56426203371, "tokens_per_second": 148504.1289400462, "data_wait_s": 1.0281000868417323e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 385600, "loss": 0.6451750993728638, "lr": 0.0003, "grad_norm": 0.15489016510070341, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 56440707124, "tokens_per_second": 147566.77223000812, "data_wait_s": 1.3055003364570439e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 385700, "loss": 0.5747754983603954, "lr": 0.0003, "grad_norm": 0.15300323239784439, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 56455342955, "tokens_per_second": 149817.77340950616, "data_wait_s": 1.2177002645330504e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 385800, "loss": 0.5800092741847038, "lr": 0.0003, "grad_norm": 0.14827453131526264, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 56469982706, "tokens_per_second": 148383.5879183736, "data_wait_s": 8.168997737811878e-06, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790782192.062ccfe6-664f-43fe-9e0e-62f43f0c4870-64tpq.429.0.1790782192342215557-281f6532

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:8e258316cf430e6be11747a29df9feadf9b6c607837d5a5874052834b2fd0519
-size 655089
+oid sha256:b21f8017b32f53d79ab31432348f605369ba1ecf72984babba9e0a14b4acbb70
+size 657207