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": 76456.7302508931,
-      "required_pages_per_second_upper_estimate": 0.005289454002812476
+      "required_token_positions_per_second": 77396.02442889404,
+      "required_pages_per_second_upper_estimate": 0.005354436553509346
     },
     {
       "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": 9557.091281361638,
-      "required_pages_per_second_upper_estimate": 0.00400135454474576
+      "required_token_positions_per_second": 9674.503053611756,
+      "required_pages_per_second_upper_estimate": 0.004050512401950273
     },
     {
       "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": 19114.182562723276,
-      "required_pages_per_second_upper_estimate": 0.003980980152923511
+      "required_token_positions_per_second": 19349.00610722351,
+      "required_pages_per_second_upper_estimate": 0.0040298877045296
     },
     {
       "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": 28671.273844084913,
-      "required_pages_per_second_upper_estimate": 0.001718799088452588
+      "required_token_positions_per_second": 29023.509160835263,
+      "required_pages_per_second_upper_estimate": 0.0017399150578595844
     },
     {
       "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": 28671.273844084913,
-      "required_pages_per_second_upper_estimate": 0.002229065429771847
+      "required_token_positions_per_second": 29023.509160835263,
+      "required_pages_per_second_upper_estimate": 0.0022564501763297656
     },
     {
       "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": 28671.273844084913,
-      "required_pages_per_second_upper_estimate": 0.3677643160565528
+      "required_token_positions_per_second": 29023.509160835263,
+      "required_pages_per_second_upper_estimate": 0.3722824124990093
     }
   ],
   "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": 720750,
-    "producer_batch_seconds": 0.14120219783603716,
-    "producer_p95_seconds": 0.30769502199837007,
-    "gpu_update_seconds": 0.9643101367016512,
-    "producer_batches_per_second": 7.082042739598089,
-    "gpu_batches_per_second": 1.0370107727171916,
-    "producer_headroom": 6.8292856023487705,
+    "producer_batch_seconds": 0.08218264320328217,
+    "producer_p95_seconds": 0.2909061640020809,
+    "gpu_update_seconds": 0.9526070692136912,
+    "producer_batches_per_second": 12.168019438441018,
+    "gpu_batches_per_second": 1.0497507653658589,
+    "producer_headroom": 11.591341335388522,
     "sustainable": true,
-    "buffer_seconds": 60.75153861220403,
+    "buffer_seconds": 60.014245360462546,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 81,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 382820.74197954044,
-    "upload_bytes_per_second": 5159349.982962749,
+    "generation_bytes_per_second": 382929.7007182376,
+    "upload_bytes_per_second": 5188497.1812688615,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 82.14441961803112,
+    "estimated_full_shards_per_hour": 82.16779962692591,
     "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": 1.58254548900004,
-      "bytes": 5975386,
-      "pages": 28,
-      "shards": 2
-    },
     {
       "seconds": 0.9414393419974658,
       "bytes": 1125185,
@@ -562,6 +556,12 @@
       "bytes": 6806858,
       "pages": 32,
       "shards": 2
+    },
+    {
+      "seconds": 1.3951733760004572,
+      "bytes": 7437431,
+      "pages": 31,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791463982175382982-d4a3b372/train.jsonl

@@ -290,3 +290,4 @@
 {"step": 954000, "loss": 0.5780755281448364, "lr": 3e-05, "grad_norm": 0.36315459425132196, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 139639558104, "tokens_per_second": 149703.5759324512, "data_wait_s": 1.9346000044606626e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 954100, "loss": 0.4935285449028015, "lr": 3e-05, "grad_norm": 0.3355567709790828, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 139654023557, "tokens_per_second": 149981.28990937743, "data_wait_s": 1.1566000466700643e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 954200, "loss": 0.495866984128952, "lr": 3e-05, "grad_norm": 0.34404256068671807, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 139668686581, "tokens_per_second": 152775.0602528793, "data_wait_s": 1.1750998964998871e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 954300, "loss": 0.5260648727416992, "lr": 3e-05, "grad_norm": 0.3393222147297149, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 139683294717, "tokens_per_second": 141898.65189820735, "data_wait_s": 6.786998710595071e-06, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791463982.2e076f0c-3fe2-4316-8f17-4e30b2e706e2-b5972.360.0.1791463982175382982-d4a3b372

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:bf610b0ecaf4bcdec3061ac85a49ad798e1bbc9fbd5e4ea15840042e8a3d7583
-size 622536
+oid sha256:a6c3796e5f82fd5c8a5b3ac34e21da2efd7675cd1006dc6d378da8d2aefebac1
+size 624654