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": 73783.21251017376,
-      "required_pages_per_second_upper_estimate": 0.0063953418585394085
+      "required_token_positions_per_second": 74150.86811189397,
+      "required_pages_per_second_upper_estimate": 0.006427209314281913
     },
     {
       "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": 9222.90156377172,
-      "required_pages_per_second_upper_estimate": 0.0036791533284552896
+      "required_token_positions_per_second": 9268.858513986746,
+      "required_pages_per_second_upper_estimate": 0.0036974862430137014
     },
     {
       "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": 18445.80312754344,
-      "required_pages_per_second_upper_estimate": 0.004267111918509794
+      "required_token_positions_per_second": 18537.717027973493,
+      "required_pages_per_second_upper_estimate": 0.004288374581744888
     },
     {
       "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": 27668.70469131516,
-      "required_pages_per_second_upper_estimate": 0.0016848789492266343
+      "required_token_positions_per_second": 27806.57554196024,
+      "required_pages_per_second_upper_estimate": 0.0016932745606784698
     },
     {
       "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": 27668.70469131516,
-      "required_pages_per_second_upper_estimate": 0.002130013479014125
+      "required_token_positions_per_second": 27806.57554196024,
+      "required_pages_per_second_upper_estimate": 0.002140627158747724
     },
     {
       "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": 21,
-      "required_token_positions_per_second": 27668.70469131516,
-      "required_pages_per_second_upper_estimate": 0.3435527110683929
+      "required_token_positions_per_second": 27806.57554196024,
+      "required_pages_per_second_upper_estimate": 0.34526460560935024
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 61,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.491016244640619,
-    "producer_p95_seconds": 0.21142883399988932,
-    "gpu_update_seconds": 0.9992516927862669,
-    "producer_batches_per_second": 2.0365924975291043,
-    "gpu_batches_per_second": 1.000748867596758,
-    "producer_headroom": 2.035068500671769,
+    "producer_batch_seconds": 0.30170215192980265,
+    "producer_p95_seconds": 0.22075719499844126,
+    "gpu_update_seconds": 0.9942971927010232,
+    "producer_batches_per_second": 3.314527236891141,
+    "gpu_batches_per_second": 1.0057355158405759,
+    "producer_headroom": 3.2956251267719407,
     "sustainable": true,
-    "buffer_seconds": 60.95435325996228,
+    "buffer_seconds": 60.652128754762415,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 80,
+    "upload_pages_per_commit": 84,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 225606.46516420715,
+    "generation_bytes_per_second": 213019.2855404701,
     "upload_bytes_per_second": 4971313.679411984,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 48.409895574518785,
+    "estimated_full_shards_per_hour": 45.708979841810006,
     "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": [

metrics/void-byte-v1/1790390190216476996-4d8dc439/train.jsonl

@@ -86,3 +86,4 @@
 {"step": 8600, "loss": 0.7029031589627266, "lr": 0.0003, "grad_norm": 0.1475043923616545, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 1257838036, "tokens_per_second": 146864.6213369466, "data_wait_s": 9.540999599266797e-06, "gpu_peak_gb": 85.43514251708984}
 {"step": 8700, "loss": 0.7436679638922215, "lr": 0.0003, "grad_norm": 0.1510031678001198, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 1272326942, "tokens_per_second": 147700.68508203633, "data_wait_s": 1.547600004414562e-05, "gpu_peak_gb": 85.43514251708984}
 {"step": 8800, "loss": 0.7235393300652504, "lr": 0.0003, "grad_norm": 0.17610486994986296, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 1286956254, "tokens_per_second": 131472.96362443778, "data_wait_s": 1.1267999070696533e-05, "gpu_peak_gb": 85.43514251708984}
+{"step": 8900, "loss": 0.7340329103171825, "lr": 0.0003, "grad_norm": 0.1643783651742324, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 1301647101, "tokens_per_second": 148835.91920559725, "data_wait_s": 1.1284000720479526e-05, "gpu_peak_gb": 85.43514251708984}

runs/void-byte-v1/events.out.tfevents.1790390190.c8125a87-1345-4bfd-b42f-0e32ed8b7bb7-6w295.451.0.1790390190216476996-4d8dc439

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:1452926d9d815a682d3b6bf320ed05d8a8d050ff1bd75c0613497d9fe0a61b27
-size 184203
+oid sha256:851bde4d3c94b7f1b884a67c2c0dace31bfe53a3cfb2bb18c9f45ec8662e58fa
+size 186286