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": 75549.23201163646,
-      "required_pages_per_second_upper_estimate": 0.00654841595299069
+      "required_token_positions_per_second": 75721.0058532333,
+      "required_pages_per_second_upper_estimate": 0.006563304874223483
     },
     {
       "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": 9443.654001454557,
-      "required_pages_per_second_upper_estimate": 0.0037672147763900417
+      "required_token_positions_per_second": 9465.125731654163,
+      "required_pages_per_second_upper_estimate": 0.0037757801705976396
     },
     {
       "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": 18887.308002909114,
-      "required_pages_per_second_upper_estimate": 0.004369246301205222
+      "required_token_positions_per_second": 18930.251463308326,
+      "required_pages_per_second_upper_estimate": 0.00437918051498949
     },
     {
       "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": 28330.962004363675,
-      "required_pages_per_second_upper_estimate": 0.0017252069449956982
+      "required_token_positions_per_second": 28395.377194962486,
+      "required_pages_per_second_upper_estimate": 0.0017291294921427795
     },
     {
       "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": 28330.962004363675,
-      "required_pages_per_second_upper_estimate": 0.002180995880218175
+      "required_token_positions_per_second": 28395.377194962486,
+      "required_pages_per_second_upper_estimate": 0.0021859547399031313
     },
     {
       "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": 28330.962004363675,
-      "required_pages_per_second_upper_estimate": 0.3517757304638076
+      "required_token_positions_per_second": 28395.377194962486,
+      "required_pages_per_second_upper_estimate": 0.35257555154727005
     }
   ],
   "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": 717798,
-    "producer_batch_seconds": 0.07541534979689146,
-    "producer_p95_seconds": 0.19062133399984305,
-    "gpu_update_seconds": 0.9758934410960531,
-    "producer_batches_per_second": 13.259900042805596,
-    "gpu_batches_per_second": 1.0247020400883853,
-    "producer_headroom": 12.940249481363255,
+    "producer_batch_seconds": 0.2646654758594096,
+    "producer_p95_seconds": 0.19091942900013237,
+    "gpu_update_seconds": 0.9736796172901314,
+    "producer_batches_per_second": 3.778354531329959,
+    "gpu_batches_per_second": 1.0270318719242797,
+    "producer_headroom": 3.678906794051788,
     "sustainable": true,
-    "buffer_seconds": 60.50539334795529,
+    "buffer_seconds": 60.36813627198815,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 80,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 450171.7435391816,
-    "upload_bytes_per_second": 5574222.938753174,
+    "generation_bytes_per_second": 424769.68989277724,
+    "upload_bytes_per_second": 5605308.235209643,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 96.59637670165621,
+    "estimated_full_shards_per_hour": 91.14568731868256,
     "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": [
@@ -352,6 +352,12 @@
       "bytes": 10132121,
       "pages": 35,
       "shards": 3
+    },
+    {
+      "seconds": 1.045252316999722,
+      "bytes": 6879925,
+      "pages": 34,
+      "shards": 2
     }
   ],
   "cache_key": {

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

@@ -59,3 +59,4 @@
 {"step": 360900, "loss": 0.5878585204482079, "lr": 0.0003, "grad_norm": 0.15594778002520077, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 52825694224, "tokens_per_second": 149461.72327852427, "data_wait_s": 8.394000360567588e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 361000, "loss": 0.5639312844723463, "lr": 0.0003, "grad_norm": 0.16376991511278852, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 52840419589, "tokens_per_second": 150113.4355381505, "data_wait_s": 1.2329999663052149e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 361100, "loss": 0.6088229678571224, "lr": 0.0003, "grad_norm": 0.15095574985346707, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 52855198388, "tokens_per_second": 150504.0480620053, "data_wait_s": 1.7798999579099473e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 361200, "loss": 0.5808791425079107, "lr": 0.0003, "grad_norm": 0.14341105196398007, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 52869726912, "tokens_per_second": 147459.70150748314, "data_wait_s": 1.1307999557175208e-05, "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 sha256:05b425a5906e92e545b282ceba6f5d386775e2593a4869c7f5f3d666e93aeaa6
-size 130146
+oid sha256:4e47ca5eb806465b6cb3e043640cc7a1e9d910927d600761a3e9c8f6345115e0
+size 132264