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": 75570.48946129721,
-      "required_pages_per_second_upper_estimate": 0.00655025849485077
+      "required_token_positions_per_second": 75410.52745766896,
+      "required_pages_per_second_upper_estimate": 0.006536393393796264
     },
     {
       "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": 9446.311182662152,
-      "required_pages_per_second_upper_estimate": 0.003768274765702151
+      "required_token_positions_per_second": 9426.31593220862,
+      "required_pages_per_second_upper_estimate": 0.003760298361340601
     },
     {
       "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": 18892.622365324303,
-      "required_pages_per_second_upper_estimate": 0.004370475685420392
+      "required_token_positions_per_second": 18852.63186441724,
+      "required_pages_per_second_upper_estimate": 0.00436122458684432
     },
     {
       "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": 28338.933547986457,
-      "required_pages_per_second_upper_estimate": 0.0017256923701788685
+      "required_token_positions_per_second": 28278.947796625856,
+      "required_pages_per_second_upper_estimate": 0.0017220395526419223
     },
     {
       "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": 28338.933547986457,
-      "required_pages_per_second_upper_estimate": 0.0021816095517129016
+      "required_token_positions_per_second": 28278.947796625856,
+      "required_pages_per_second_upper_estimate": 0.0021769916825219752
     },
     {
       "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": 22,
-      "required_token_positions_per_second": 28338.933547986457,
-      "required_pages_per_second_upper_estimate": 0.35187471035656226
+      "required_token_positions_per_second": 28278.947796625856,
+      "required_pages_per_second_upper_estimate": 0.351129888084059
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.13405573252325098,
-    "producer_p95_seconds": 0.3564302119993954,
-    "gpu_update_seconds": 0.9756189291027308,
-    "producer_batches_per_second": 7.459584019106064,
-    "gpu_batches_per_second": 1.0249903627020565,
-    "producer_headroom": 7.277711372272103,
+    "producer_batch_seconds": 0.2838167806953038,
+    "producer_p95_seconds": 0.32916511800067383,
+    "gpu_update_seconds": 0.9776884274067248,
+    "producer_batches_per_second": 3.5233998410882075,
+    "gpu_batches_per_second": 1.0228207391719422,
+    "producer_headroom": 3.4447872497586336,
     "sustainable": true,
-    "buffer_seconds": 60.488373604369315,
+    "buffer_seconds": 60.616682499216935,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 338702.2395095985,
-    "upload_bytes_per_second": 6230463.703111002,
+    "generation_bytes_per_second": 329156.4086216148,
+    "upload_bytes_per_second": 6229472.319956904,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 72.67761601415603,
+    "estimated_full_shards_per_hour": 70.62930292116482,
     "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": 0.979562092001288,
-      "bytes": 7577775,
-      "pages": 38,
-      "shards": 3
-    },
     {
       "seconds": 0.9862535339998431,
       "bytes": 4746157,
@@ -562,6 +556,12 @@
       "bytes": 5798800,
       "pages": 31,
       "shards": 2
+    },
+    {
+      "seconds": 1.0886239710016525,
+      "bytes": 8189805,
+      "pages": 33,
+      "shards": 3
     }
   ],
   "cache_key": {

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

@@ -290,3 +290,4 @@
 {"step": 29000, "loss": 0.6972436346113682, "lr": 0.0003, "grad_norm": 0.14391108421625456, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 4243645074, "tokens_per_second": 149921.3201087006, "data_wait_s": 1.4084002032177523e-05, "gpu_peak_gb": 85.43514251708984}
 {"step": 29100, "loss": 0.6293734144419432, "lr": 0.0003, "grad_norm": 0.13333394917548458, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 4258227975, "tokens_per_second": 149456.71999855785, "data_wait_s": 1.162899934570305e-05, "gpu_peak_gb": 85.43514251708984}
 {"step": 29200, "loss": 0.6442718654870987, "lr": 0.0003, "grad_norm": 0.13196648001111677, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 4272839299, "tokens_per_second": 149047.26761701144, "data_wait_s": 9.216000762535259e-06, "gpu_peak_gb": 85.43514251708984}
+{"step": 29300, "loss": 0.6453717909753323, "lr": 0.0003, "grad_norm": 0.1279015889634107, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 4287437906, "tokens_per_second": 148910.05981293134, "data_wait_s": 1.2165000953245908e-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 sha256:4ec9cba1d35cca770305e8f872d0588884decea3c8317101950d22d67817d773
-size 616762
+oid sha256:e7941f7eb8319911d6b2d727c22173d3fcaa01ae9b548b7775e325f39c3d4159
+size 618880