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": 77270.56915788558,
-      "required_pages_per_second_upper_estimate": 0.0053457572667646456
+      "required_token_positions_per_second": 77369.76193348593,
+      "required_pages_per_second_upper_estimate": 0.005352619653139628
     },
     {
       "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": 9658.821144735697,
-      "required_pages_per_second_upper_estimate": 0.004043946714179363
+      "required_token_positions_per_second": 9671.220241685742,
+      "required_pages_per_second_upper_estimate": 0.004049137957149759
     },
     {
       "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": 19317.642289471394,
-      "required_pages_per_second_upper_estimate": 0.004023355448411329
+      "required_token_positions_per_second": 19342.440483371483,
+      "required_pages_per_second_upper_estimate": 0.0040285202582283665
     },
     {
       "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": 28976.46343420709,
-      "required_pages_per_second_upper_estimate": 0.0017370947383828904
+      "required_token_positions_per_second": 29013.660725057223,
+      "required_pages_per_second_upper_estimate": 0.0017393246591723816
     },
     {
       "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": 28976.46343420709,
-      "required_pages_per_second_upper_estimate": 0.0022527925780167074
+      "required_token_positions_per_second": 29013.660725057223,
+      "required_pages_per_second_upper_estimate": 0.002255684503766714
     },
     {
       "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": 28976.46343420709,
-      "required_pages_per_second_upper_estimate": 0.3716789604315887
+      "required_token_positions_per_second": 29013.660725057223,
+      "required_pages_per_second_upper_estimate": 0.3721560873392751
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.08973011307809742,
-    "producer_p95_seconds": 0.15420620899931237,
-    "gpu_update_seconds": 0.9541537069482806,
-    "producer_batches_per_second": 11.144530700965916,
-    "gpu_batches_per_second": 1.0480491693506615,
-    "producer_headroom": 10.63359528052555,
+    "producer_batch_seconds": 0.17527620346091055,
+    "producer_p95_seconds": 0.16011526199963555,
+    "gpu_update_seconds": 0.9529304234305811,
+    "producer_batches_per_second": 5.705281037896375,
+    "gpu_batches_per_second": 1.0493945574745813,
+    "producer_headroom": 5.436735875233058,
     "sustainable": true,
-    "buffer_seconds": 60.111683537741676,
+    "buffer_seconds": 60.03461667612661,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 86,
+    "upload_pages_per_commit": 81,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 496902.5412332813,
-    "upload_bytes_per_second": 833951.5265037145,
+    "generation_bytes_per_second": 476112.8571544018,
+    "upload_bytes_per_second": 1082368.9061254323,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 106.62371804951505,
+    "estimated_full_shards_per_hour": 102.1627358052639,
     "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": [
@@ -220,6 +220,12 @@
       "bytes": 1130814,
       "pages": 33,
       "shards": 1
+    },
+    {
+      "seconds": 1.5193411210002523,
+      "bytes": 12040700,
+      "pages": 35,
+      "shards": 4
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791083219122319327-61ae60ac/train.jsonl

@@ -40,3 +40,4 @@
 {"step": 649000, "loss": 0.5598666071891785, "lr": 0.0001, "grad_norm": 0.18677011749932018, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 94997187006, "tokens_per_second": 152607.00392093902, "data_wait_s": 1.4277999980549794e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 649100, "loss": 0.5617611408233643, "lr": 0.0001, "grad_norm": 0.17680430993966517, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 95011701494, "tokens_per_second": 150758.08111059852, "data_wait_s": 1.2604000403371174e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 649200, "loss": 0.4521869719028473, "lr": 0.0001, "grad_norm": 0.21298838570299822, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 95026242892, "tokens_per_second": 151542.40936257635, "data_wait_s": 1.3130000297678635e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 649300, "loss": 0.5155933499336243, "lr": 0.0001, "grad_norm": 0.16678892690074087, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 95040929247, "tokens_per_second": 151799.1673511, "data_wait_s": 1.3793000107398257e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791083219.1a73f132-ede3-4b51-9e4f-0cafff989449-8p4r7.435.0.1791083219122319327-61ae60ac

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:20a37ed86a089c66304e429e000e71865d1d24f3dead0f1b4111c45548ff8d71
-size 89121
+oid sha256:22bba539aa8576f3523e829d7fde304274ab625b029394c69b0b825acd450460
+size 91239