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": 74817.5457905918,
-      "required_pages_per_second_upper_estimate": 0.006484995245792312
+      "required_token_positions_per_second": 74297.30676053936,
+      "required_pages_per_second_upper_estimate": 0.006439902245201134
     },
     {
       "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": 9352.193223823975,
-      "required_pages_per_second_upper_estimate": 0.003730729704732717
+      "required_token_positions_per_second": 9287.16334506742,
+      "required_pages_per_second_upper_estimate": 0.0037047883138133955
     },
     {
       "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": 18704.38644764795,
-      "required_pages_per_second_upper_estimate": 0.0043269306187049715
+      "required_token_positions_per_second": 18574.32669013484,
+      "required_pages_per_second_upper_estimate": 0.0042968435827778675
     },
     {
       "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": 28056.579671471925,
-      "required_pages_per_second_upper_estimate": 0.0017084985004954352
+      "required_token_positions_per_second": 27861.49003520226,
+      "required_pages_per_second_upper_estimate": 0.0016966185652028294
     },
     {
       "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": 28056.579671471925,
-      "required_pages_per_second_upper_estimate": 0.002159873168693258
+      "required_token_positions_per_second": 27861.49003520226,
+      "required_pages_per_second_upper_estimate": 0.0021448546284505415
     },
     {
       "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": 28056.579671471925,
-      "required_pages_per_second_upper_estimate": 0.3483688201878879
+      "required_token_positions_per_second": 27861.49003520226,
+      "required_pages_per_second_upper_estimate": 0.3459464598284299
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,39 +158,27 @@
     "prefetch_batches": 61,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.15711286308589933,
-    "producer_p95_seconds": 0.3503623950018664,
-    "gpu_update_seconds": 0.9854372957696134,
-    "producer_batches_per_second": 6.364851230884027,
-    "gpu_batches_per_second": 1.0147779105711778,
-    "producer_headroom": 6.272161784938251,
+    "producer_batch_seconds": 0.10504269463280025,
+    "producer_p95_seconds": 0.33738547699613264,
+    "gpu_update_seconds": 0.9923374509068783,
+    "producer_batches_per_second": 9.51993856874787,
+    "gpu_batches_per_second": 1.0077217171297113,
+    "producer_headroom": 9.446991572101338,
     "sustainable": true,
-    "buffer_seconds": 60.11167504194641,
+    "buffer_seconds": 60.532584505319576,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 79,
+    "upload_pages_per_commit": 80,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 374399.8448116255,
-    "upload_bytes_per_second": 4221736.650489915,
+    "generation_bytes_per_second": 373092.5294340365,
+    "upload_bytes_per_second": 4272189.235532882,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 80.33749111424993,
+    "estimated_full_shards_per_hour": 80.05697166696379,
     "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": 1.3773436970004695,
-      "bytes": 2708180,
-      "pages": 37,
-      "shards": 2
-    },
-    {
-      "seconds": 1.157878524998523,
-      "bytes": 6108198,
-      "pages": 36,
-      "shards": 2
-    },
     {
       "seconds": 1.1510198380019574,
       "bytes": 2279083,
@@ -562,6 +550,18 @@
       "bytes": 6552255,
       "pages": 33,
       "shards": 3
+    },
+    {
+      "seconds": 1.1842839309974806,
+      "bytes": 12073153,
+      "pages": 39,
+      "shards": 3
+    },
+    {
+      "seconds": 0.9938246039964724,
+      "bytes": 235889,
+      "pages": 8,
+      "shards": 1
     }
   ],
   "cache_key": {

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

@@ -348,3 +348,4 @@
 {"step": 189800, "loss": 0.6157660949975252, "lr": 0.0003, "grad_norm": 0.16425549973474607, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 27778820913, "tokens_per_second": 151815.9570990028, "data_wait_s": 2.1535000996664166e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 189900, "loss": 0.6188155319541693, "lr": 0.0003, "grad_norm": 0.14061967457514632, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 27793615160, "tokens_per_second": 151529.41461533174, "data_wait_s": 1.0442003258503973e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 190000, "loss": 0.5794018357992172, "lr": 0.0003, "grad_norm": 0.143421867137207, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 27808256418, "tokens_per_second": 150552.32028538815, "data_wait_s": 7.763999747112393e-06, "gpu_peak_gb": 85.46427917480469}
+{"step": 190100, "loss": 0.5956941954791546, "lr": 0.0003, "grad_norm": 0.15725801850204055, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 27822802377, "tokens_per_second": 120622.25517734191, "data_wait_s": 8.716000593267381e-06, "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 sha256:8a8ca3b183a5a078eb22c50a90ac4e7bf4688de863787a48bf78dab9109d79aa
-size 746946
+oid sha256:328d49eb24371811eb4e79eeaf351eb65a1d525e879155802b20a24d5ff0774b
+size 749064