Update TensorBoard training metrics

benchmark_cache/shard_scheduler.json

@@ -22,8 +22,8 @@
       "tokens_per_second": 598315.063954628,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 77011.92164436255,
-      "required_pages_per_second_upper_estimate": 0.005327863431634207
+      "required_token_positions_per_second": 76794.4591712007,
+      "required_pages_per_second_upper_estimate": 0.005312818873158415
     },
     {
       "source": "rewrite6",
@@ -36,8 +36,8 @@
       "tokens_per_second": 8533.957213384612,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 9626.490205545319,
-      "required_pages_per_second_upper_estimate": 0.004030410425087135
+      "required_token_positions_per_second": 9599.307396400087,
+      "required_pages_per_second_upper_estimate": 0.00401902955053963
     },
     {
       "source": "openmath",
@@ -50,8 +50,8 @@
       "tokens_per_second": 380144.35761751566,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 19252.980411090637,
-      "required_pages_per_second_upper_estimate": 0.0040098880843930235
+      "required_token_positions_per_second": 19198.614792800174,
+      "required_pages_per_second_upper_estimate": 0.003998565159820887
     },
     {
       "source": "ultra_qa",
@@ -64,8 +64,8 @@
       "tokens_per_second": 43742.16556775435,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28879.470616635954,
-      "required_pages_per_second_upper_estimate": 0.001731280167068958
+      "required_token_positions_per_second": 28797.92218920026,
+      "required_pages_per_second_upper_estimate": 0.0017263914633614272
     },
     {
       "source": "ultra_style",
@@ -78,8 +78,8 @@
       "tokens_per_second": 22018.86276523721,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28879.470616635954,
-      "required_pages_per_second_upper_estimate": 0.0022452518130768686
+      "required_token_positions_per_second": 28797.92218920026,
+      "required_pages_per_second_upper_estimate": 0.002238911781537368
     },
     {
       "source": "cortex",
@@ -92,8 +92,8 @@
       "tokens_per_second": 910315.184299418,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 23,
-      "required_token_positions_per_second": 28879.470616635954,
-      "required_pages_per_second_upper_estimate": 0.3704348407105598
+      "required_token_positions_per_second": 28797.92218920026,
+      "required_pages_per_second_upper_estimate": 0.36938882504329423
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 704566,
-    "producer_batch_seconds": 0.15161653654638485,
-    "producer_p95_seconds": 0.3704427859993302,
-    "gpu_update_seconds": 0.9573582690284299,
-    "producer_batches_per_second": 6.595586621213081,
-    "gpu_batches_per_second": 1.0445410379280944,
-    "producer_headroom": 6.314339390911625,
+    "producer_batch_seconds": 0.4044776608901657,
+    "producer_p95_seconds": 0.3602808139985427,
+    "gpu_update_seconds": 0.9600692653572243,
+    "producer_batches_per_second": 2.472324424046613,
+    "gpu_batches_per_second": 1.0415915143663288,
+    "producer_headroom": 2.373602693519154,
     "sustainable": true,
-    "buffer_seconds": 60.31357094879108,
+    "buffer_seconds": 60.484363717505126,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 87,
+    "upload_pages_per_commit": 80,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 451012.64164763945,
-    "upload_bytes_per_second": 5031697.54989096,
+    "generation_bytes_per_second": 415109.2820839442,
+    "upload_bytes_per_second": 5022611.3694364345,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 96.77681386062515,
+    "estimated_full_shards_per_hour": 89.07278868569131,
     "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": 1.6113828489978914,
-      "bytes": 12223061,
-      "pages": 49,
-      "shards": 4
-    },
     {
       "seconds": 0.9428402260018629,
       "bytes": 5721024,
@@ -562,6 +556,12 @@
       "bytes": 7318826,
       "pages": 31,
       "shards": 3
+    },
+    {
+      "seconds": 1.3007696439963183,
+      "bytes": 9895316,
+      "pages": 41,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790868774355361490-de221e1d/train.jsonl

@@ -383,3 +383,4 @@
 {"step": 478500, "loss": 0.560171365737915, "lr": 0.0003, "grad_norm": 0.16747739114457763, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 70037184439, "tokens_per_second": 152712.71266018532, "data_wait_s": 9.809999028220773e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 478600, "loss": 0.5615736842155457, "lr": 0.0003, "grad_norm": 0.1617892725936926, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 70051902110, "tokens_per_second": 152980.99716700756, "data_wait_s": 1.5670004358980805e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 478700, "loss": 0.565996527671814, "lr": 0.0003, "grad_norm": 0.15105829437840865, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 70066633918, "tokens_per_second": 153141.2903115773, "data_wait_s": 1.0431998816784471e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 478800, "loss": 0.5429500341415405, "lr": 0.0003, "grad_norm": 0.17888349647095725, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 70081062930, "tokens_per_second": 149682.05512863398, "data_wait_s": 9.66299558058381e-06, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1790868774.f92f6aaf-743a-4c24-a54a-850bd75954cb-8js8b.612.0.1790868774355361490-de221e1d

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:5256cac45cbfc4175110a3ee31eb2de66633b4dd0d7aa33dfb46897d65e2f5cf
-size 821076
+oid sha256:fcf2616163b7d87e654c82c53189c6a678c0aa9f0ae6e68311bf30ad3f4dac5c
+size 823194