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": 73766.50150923943,
-      "required_pages_per_second_upper_estimate": 0.006393893391332065
+      "required_token_positions_per_second": 74805.39933802045,
+      "required_pages_per_second_upper_estimate": 0.006483942422068329
     },
     {
       "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": 9220.812688654929,
-      "required_pages_per_second_upper_estimate": 0.0036783200449397356
+      "required_token_positions_per_second": 9350.674917252556,
+      "required_pages_per_second_upper_estimate": 0.0037301240295406716
     },
     {
       "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": 18441.625377309858,
-      "required_pages_per_second_upper_estimate": 0.004266145469519147
+      "required_token_positions_per_second": 18701.349834505112,
+      "required_pages_per_second_upper_estimate": 0.004326228151696942
     },
     {
       "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": 27662.438065964787,
-      "required_pages_per_second_upper_estimate": 0.0016844973446212387
+      "required_token_positions_per_second": 28052.024751757668,
+      "required_pages_per_second_upper_estimate": 0.0017082211297826538
     },
     {
       "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": 27662.438065964787,
-      "required_pages_per_second_upper_estimate": 0.0021295310568475237
+      "required_token_positions_per_second": 28052.024751757668,
+      "required_pages_per_second_upper_estimate": 0.0021595225183648294
     },
     {
       "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": 21,
-      "required_token_positions_per_second": 27662.438065964787,
-      "required_pages_per_second_upper_estimate": 0.3434749005545872
+      "required_token_positions_per_second": 28052.024751757668,
+      "required_pages_per_second_upper_estimate": 0.34831226332937243
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 61,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.23404260364846152,
-    "producer_p95_seconds": 0.28490845399937825,
-    "gpu_update_seconds": 0.9994780624205879,
-    "producer_batches_per_second": 4.272726351574979,
-    "gpu_batches_per_second": 1.0005222101405087,
-    "producer_headroom": 4.270496255125548,
+    "producer_batch_seconds": 0.3162211251797231,
+    "producer_p95_seconds": 0.2998101889988902,
+    "gpu_update_seconds": 0.985597305173761,
+    "producer_batches_per_second": 3.162344069934144,
+    "gpu_batches_per_second": 1.0146131637643834,
+    "producer_headroom": 3.1167977933593156,
     "sustainable": true,
-    "buffer_seconds": 60.96816180765586,
+    "buffer_seconds": 60.12143561559942,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 79,
+    "upload_pages_per_commit": 78,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 441505.85580955155,
-    "upload_bytes_per_second": 5147532.125934589,
+    "generation_bytes_per_second": 430180.1909169438,
+    "upload_bytes_per_second": 5205976.992258369,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 94.73687892641935,
+    "estimated_full_shards_per_hour": 92.30665488845095,
     "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": 1.3963090409997676,
-      "bytes": 10793699,
-      "pages": 45,
-      "shards": 3
-    },
     {
       "seconds": 1.3661847930006843,
       "bytes": 7242166,
@@ -562,6 +556,12 @@
       "bytes": 836703,
       "pages": 25,
       "shards": 1
+    },
+    {
+      "seconds": 1.6271591859986074,
+      "bytes": 16803522,
+      "pages": 35,
+      "shards": 4
     }
   ],
   "cache_key": {

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

@@ -145,3 +145,4 @@
 {"step": 169500, "loss": 0.5566157568246126, "lr": 0.0003, "grad_norm": 0.14683305342487346, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 24806932915, "tokens_per_second": 148869.1731775142, "data_wait_s": 8.650000381749123e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 169600, "loss": 0.5995315574109554, "lr": 0.0003, "grad_norm": 0.1519435142341528, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 24821573889, "tokens_per_second": 149355.70367058695, "data_wait_s": 1.0183999620494433e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 169700, "loss": 0.55619752779603, "lr": 0.0003, "grad_norm": 0.15390745482804757, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 24836272163, "tokens_per_second": 149863.83213670508, "data_wait_s": 1.2968999726581387e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 169800, "loss": 0.5832123067229986, "lr": 0.0003, "grad_norm": 0.14845787700264634, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 24850943091, "tokens_per_second": 150583.05351849974, "data_wait_s": 1.3396000213106163e-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 sha256:479d32f9de93affff2fb52e19e4dcabdca08b6d9a3ae99821e63255e80217b4a
-size 313077
+oid sha256:a5de01a30db69022dfd242f4fd0931402ce15663cd1b20d26e0d776d257129fc
+size 315195