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": 75934.59014793354,
-      "required_pages_per_second_upper_estimate": 0.006581817819563654
+      "required_token_positions_per_second": 75700.07760693831,
+      "required_pages_per_second_upper_estimate": 0.006561490866876789
     },
     {
       "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": 9491.823768491693,
-      "required_pages_per_second_upper_estimate": 0.003786430416663353
+      "required_token_positions_per_second": 9462.50970086729,
+      "required_pages_per_second_upper_estimate": 0.00377473659680361
     },
     {
       "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": 18983.647536983386,
-      "required_pages_per_second_upper_estimate": 0.0043915327568954145
+      "required_token_positions_per_second": 18925.01940173458,
+      "required_pages_per_second_upper_estimate": 0.004377970169625542
     },
     {
       "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": 28475.47130547508,
-      "required_pages_per_second_upper_estimate": 0.001734006803251675
+      "required_token_positions_per_second": 28387.529102601868,
+      "required_pages_per_second_upper_estimate": 0.0017286515844937764
     },
     {
       "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": 28475.47130547508,
-      "required_pages_per_second_upper_estimate": 0.0021921206062450794
+      "required_token_positions_per_second": 28387.529102601868,
+      "required_pages_per_second_upper_estimate": 0.002185350572028301
     },
     {
       "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": 28475.47130547508,
-      "required_pages_per_second_upper_estimate": 0.3535700523420922
+      "required_token_positions_per_second": 28387.529102601868,
+      "required_pages_per_second_upper_estimate": 0.3524781045060266
     }
   ],
   "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": 688946,
-    "producer_batch_seconds": 0.066400200140599,
-    "producer_p95_seconds": 0.16180038799939211,
-    "gpu_update_seconds": 0.9709409092268131,
-    "producer_batches_per_second": 15.06019556993129,
-    "gpu_batches_per_second": 1.0299287943241855,
-    "producer_headroom": 14.622559979802709,
+    "producer_batch_seconds": 0.11295699792968605,
+    "producer_p95_seconds": 0.18674839699997392,
+    "gpu_update_seconds": 0.9739488033661202,
+    "producer_batches_per_second": 8.852926497059388,
+    "gpu_batches_per_second": 1.0267480144170236,
+    "producer_headroom": 8.622297168099209,
     "sustainable": true,
-    "buffer_seconds": 60.198336372062414,
+    "buffer_seconds": 60.38482580869945,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 419912.88422934583,
-    "upload_bytes_per_second": 4677095.503737477,
+    "generation_bytes_per_second": 433393.456170296,
+    "upload_bytes_per_second": 4593978.50912802,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 90.1035298839596,
+    "estimated_full_shards_per_hour": 92.99614681083355,
     "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": [
@@ -262,6 +262,12 @@
       "bytes": 6265294,
       "pages": 27,
       "shards": 2
+    },
+    {
+      "seconds": 0.9120372190000126,
+      "bytes": 2656908,
+      "pages": 35,
+      "shards": 2
     }
   ],
   "cache_key": {

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

@@ -45,3 +45,4 @@
 {"step": 159500, "loss": 0.5589956678450108, "lr": 0.0003, "grad_norm": 0.14677640845151912, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 23343720107, "tokens_per_second": 149320.57667826503, "data_wait_s": 1.1400999937904999e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 159600, "loss": 0.6243120767176151, "lr": 0.0003, "grad_norm": 0.15560457008791292, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 23358465314, "tokens_per_second": 150561.28663703115, "data_wait_s": 1.5671000255679246e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 159700, "loss": 0.6141259353607893, "lr": 0.0003, "grad_norm": 0.13582600490727087, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 23373370513, "tokens_per_second": 151190.07967858604, "data_wait_s": 1.5458000234502833e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 159800, "loss": 0.5677892863750458, "lr": 0.0003, "grad_norm": 0.1604483381015062, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 23388092979, "tokens_per_second": 150897.233162046, "data_wait_s": 1.2524999874585774e-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 sha256:8dc3b2a22a9cbf8b02cfe7c52bed2b0cc9ed6a2c247f48365372dae6469744f7
-size 99711
+oid sha256:a34778ad3597a6dece88b0560fb738f333f390de9b8c79dce6ade7be0657e5e0
+size 101829