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": 77377.81542118467,
-      "required_pages_per_second_upper_estimate": 0.005353176812105292
+      "required_token_positions_per_second": 76646.06726201934,
+      "required_pages_per_second_upper_estimate": 0.005302552776564581
     },
     {
       "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": 9672.226927648084,
-      "required_pages_per_second_upper_estimate": 0.004049559435540198
+      "required_token_positions_per_second": 9580.758407752417,
+      "required_pages_per_second_upper_estimate": 0.00401126347633978
     },
     {
       "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": 19344.453855296168,
-      "required_pages_per_second_upper_estimate": 0.0040289395905040905
+      "required_token_positions_per_second": 19161.516815504834,
+      "required_pages_per_second_upper_estimate": 0.00399083862948972
     },
     {
       "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": 29016.68078294425,
-      "required_pages_per_second_upper_estimate": 0.0017395057070313474
+      "required_token_positions_per_second": 28742.27522325725,
+      "required_pages_per_second_upper_estimate": 0.001723055512721141
     },
     {
       "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": 29016.68078294425,
-      "required_pages_per_second_upper_estimate": 0.0022559192999835995
+      "required_token_positions_per_second": 28742.27522325725,
+      "required_pages_per_second_upper_estimate": 0.0022345854747004346
     },
     {
       "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": 29016.68078294425,
-      "required_pages_per_second_upper_estimate": 0.3721948253991643
+      "required_token_positions_per_second": 28742.27522325725,
+      "required_pages_per_second_upper_estimate": 0.36867504551323416
     }
   ],
   "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": 0.11550370173438296,
-    "producer_p95_seconds": 0.3203342449996853,
-    "gpu_update_seconds": 0.9528312423746017,
-    "producer_batches_per_second": 8.657731180768916,
-    "gpu_batches_per_second": 1.0495037898923703,
-    "producer_headroom": 8.249356757117374,
+    "producer_batch_seconds": 0.12439553138312931,
+    "producer_p95_seconds": 0.31447203300194815,
+    "gpu_update_seconds": 0.9619280236252209,
+    "producer_batches_per_second": 8.038873976269064,
+    "gpu_batches_per_second": 1.0395788202856355,
+    "producer_headroom": 7.732818156164723,
     "sustainable": true,
-    "buffer_seconds": 60.02836826959991,
+    "buffer_seconds": 60.60146548838892,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 84,
+    "upload_pages_per_commit": 85,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 318579.3370430871,
-    "upload_bytes_per_second": 5040513.647559811,
+    "generation_bytes_per_second": 347190.89836736047,
+    "upload_bytes_per_second": 5027569.432646201,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 68.35970958203754,
+    "estimated_full_shards_per_hour": 74.49908459916698,
     "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": 1.2248858210004983,
-      "bytes": 5966749,
-      "pages": 32,
-      "shards": 2
-    },
     {
       "seconds": 1.2219619630013767,
       "bytes": 6395351,
@@ -562,6 +556,12 @@
       "bytes": 1179043,
       "pages": 36,
       "shards": 1
+    },
+    {
+      "seconds": 1.417530385995633,
+      "bytes": 5826314,
+      "pages": 31,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791399463488568880-054ba646/train.jsonl

@@ -378,3 +378,4 @@
 {"step": 922800, "loss": 0.5286610722541809, "lr": 3e-05, "grad_norm": 0.32214467890983767, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 135071503388, "tokens_per_second": 152790.25286520587, "data_wait_s": 1.4129996998235583e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 922900, "loss": 0.5633100271224976, "lr": 3e-05, "grad_norm": 0.3362075815111445, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 135086155620, "tokens_per_second": 153346.35760229715, "data_wait_s": 1.3720004062633961e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 923000, "loss": 0.49429020285606384, "lr": 3e-05, "grad_norm": 0.32696403912165645, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 135100832265, "tokens_per_second": 152626.65675231628, "data_wait_s": 1.3780998415313661e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 923100, "loss": 0.5448967218399048, "lr": 3e-05, "grad_norm": 0.3292880024477215, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 135115520222, "tokens_per_second": 153429.7172407346, "data_wait_s": 8.725000952836126e-06, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791399463.60f38c85-66c9-4fe6-8e58-7b3a56dc2cee-wtkw8.346.0.1791399463488568880-054ba646

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:a7b0daaa9243ba29b3e36558046ac41117fc01856b32739ece1f6cf5d09afd43
-size 810486
+oid sha256:c191d4a45c8f74b7f31fe0faf84d8046960b3966b6a367921ef987a44cd31ddd
+size 812604