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": 75288.57051603614,
-      "required_pages_per_second_upper_estimate": 0.006525822475192575
+      "required_token_positions_per_second": 75427.73566052444,
+      "required_pages_per_second_upper_estimate": 0.006537884957205987
     },
     {
       "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": 9411.071314504517,
-      "required_pages_per_second_upper_estimate": 0.0037542170554110887
+      "required_token_positions_per_second": 9428.466957565555,
+      "required_pages_per_second_upper_estimate": 0.003761156437516178
     },
     {
       "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": 18822.142629009035,
-      "required_pages_per_second_upper_estimate": 0.004354171438824848
+      "required_token_positions_per_second": 18856.93391513111,
+      "required_pages_per_second_upper_estimate": 0.004362219790563468
     },
     {
       "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": 28233.21394351355,
-      "required_pages_per_second_upper_estimate": 0.001719254600934366
+      "required_token_positions_per_second": 28285.400872696668,
+      "required_pages_per_second_upper_estimate": 0.0017224325111179612
     },
     {
       "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": 28233.21394351355,
-      "required_pages_per_second_upper_estimate": 0.0021734709639099816
+      "required_token_positions_per_second": 28285.400872696668,
+      "required_pages_per_second_upper_estimate": 0.002177488458181164
     },
     {
       "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": 28233.21394351355,
-      "required_pages_per_second_upper_estimate": 0.35056202668976433
+      "required_token_positions_per_second": 28285.400872696668,
+      "required_pages_per_second_upper_estimate": 0.3512100136918021
     }
   ],
   "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": 0.07977011278140367,
-    "producer_p95_seconds": 0.2210376319999341,
-    "gpu_update_seconds": 0.9792721457541322,
-    "producer_batches_per_second": 12.536023394379907,
-    "gpu_batches_per_second": 1.0211665922856463,
-    "producer_headroom": 12.276178528638411,
+    "producer_batch_seconds": 0.05929656364840241,
+    "producer_p95_seconds": 0.20810847400025523,
+    "gpu_update_seconds": 0.977465376023292,
+    "producer_batches_per_second": 16.864383675409535,
+    "gpu_batches_per_second": 1.023054140360846,
+    "producer_headroom": 16.484351130685248,
     "sustainable": true,
-    "buffer_seconds": 60.7148730367562,
+    "buffer_seconds": 60.6028533134441,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 78,
+    "upload_pages_per_commit": 87,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 221714.02868585457,
-    "upload_bytes_per_second": 6583733.866930622,
+    "generation_bytes_per_second": 246085.66055733646,
+    "upload_bytes_per_second": 6513806.664100802,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 47.574669317548064,
+    "estimated_full_shards_per_hour": 52.80425417461463,
     "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": 0.9783495700003186,
-      "bytes": 7342912,
-      "pages": 40,
-      "shards": 4
-    },
     {
       "seconds": 0.9914971539992621,
       "bytes": 6238955,
@@ -562,6 +556,12 @@
       "bytes": 7745172,
       "pages": 44,
       "shards": 4
+    },
+    {
+      "seconds": 0.7505128630000399,
+      "bytes": 1172556,
+      "pages": 33,
+      "shards": 1
     }
   ],
   "cache_key": {

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

@@ -113,3 +113,4 @@
 {"step": 11300, "loss": 0.6859790533781052, "lr": 0.0003, "grad_norm": 0.16650099415743275, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 1653152694, "tokens_per_second": 150670.7353186621, "data_wait_s": 1.068199890141841e-05, "gpu_peak_gb": 85.43514251708984}
 {"step": 11400, "loss": 0.7066514007747173, "lr": 0.0003, "grad_norm": 0.15130700510484968, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 1667767034, "tokens_per_second": 147853.70048813766, "data_wait_s": 1.3994998880662024e-05, "gpu_peak_gb": 85.43514251708984}
 {"step": 11500, "loss": 0.6995341554284096, "lr": 0.0003, "grad_norm": 0.15938632569844022, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 1682329231, "tokens_per_second": 148173.18191002306, "data_wait_s": 1.5789999451953918e-05, "gpu_peak_gb": 85.43514251708984}
+{"step": 11600, "loss": 0.6731462813913822, "lr": 0.0003, "grad_norm": 0.14444594314923076, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 1696884254, "tokens_per_second": 147854.69487109123, "data_wait_s": 1.1763999282266013e-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 sha256:7fd3529d7aab8d263e0b1842c70fbe75dfab371da14ffa853b80e586289eda13
-size 241217
+oid sha256:f3fbd76cc6167ca51bc64c239857c28d3e4da4f7cf94f525aab6cf80c01354b9
+size 243300