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": 74860.70388277741,
-      "required_pages_per_second_upper_estimate": 0.005179037195374844
+      "required_token_positions_per_second": 77227.71469895172,
+      "required_pages_per_second_upper_estimate": 0.005342792495859557
     },
     {
       "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": 9357.587985347176,
-      "required_pages_per_second_upper_estimate": 0.00391782668080707
+      "required_token_positions_per_second": 9653.464337368965,
+      "required_pages_per_second_upper_estimate": 0.00404170393079777
     },
     {
       "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": 18715.175970694352,
-      "required_pages_per_second_upper_estimate": 0.003897877602315326
+      "required_token_positions_per_second": 19306.92867473793,
+      "required_pages_per_second_upper_estimate": 0.004021124084999369
     },
     {
       "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": 28072.76395604153,
-      "required_pages_per_second_upper_estimate": 0.0016829193345360713
+      "required_token_positions_per_second": 28960.393012106895,
+      "required_pages_per_second_upper_estimate": 0.0017361313411160975
     },
     {
       "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": 28072.76395604153,
-      "required_pages_per_second_upper_estimate": 0.0021825339185432536
+      "required_token_positions_per_second": 28960.393012106895,
+      "required_pages_per_second_upper_estimate": 0.0022515431733847336
     },
     {
       "source": "cortex",
@@ -91,9 +91,9 @@
       "read_seconds": 0.04061180899998362,
       "tokens_per_second": 910315.184299418,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
-      "estimated_pages_for_buffer": 22,
-      "required_token_positions_per_second": 28072.76395604153,
-      "required_pages_per_second_upper_estimate": 0.36008727384258193
+      "estimated_pages_for_buffer": 23,
+      "required_token_positions_per_second": 28960.393012106895,
+      "required_pages_per_second_upper_estimate": 0.3714728263119623
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,36 +155,30 @@
   "shard_benchmark_scope": "Local packing/read measurements; upload capacity measured only on real data commits. Targets larger than available pages are sample-limited.",
   "estimated_fresh_generation_batch_seconds": 2.558590814491007,
   "selected": {
-    "prefetch_batches": 61,
+    "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 18.944940262664062,
-    "producer_p95_seconds": 0.280350291999639,
-    "gpu_update_seconds": 0.9848691793687769,
-    "producer_batches_per_second": 0.05278454226486849,
-    "gpu_batches_per_second": 1.015363279660067,
-    "producer_headroom": 0.05198586882375755,
-    "sustainable": false,
-    "buffer_seconds": 60.077019941495394,
-    "buffer_drain_seconds": 63.371439270588944,
+    "producer_batch_seconds": 0.07544692595320157,
+    "producer_p95_seconds": 0.2617425459975493,
+    "gpu_update_seconds": 0.9546831767249067,
+    "producer_batches_per_second": 13.254350490307356,
+    "gpu_batches_per_second": 1.04746791855132,
+    "producer_headroom": 12.65370543151195,
+    "sustainable": true,
+    "buffer_seconds": 60.14504013366912,
+    "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 78,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 38138.65869711697,
-    "upload_bytes_per_second": 5263667.821253658,
+    "generation_bytes_per_second": 38126.40045991235,
+    "upload_bytes_per_second": 5264770.992110251,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 8.18366833386547,
+    "estimated_full_shards_per_hour": 8.181038001518516,
     "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": 0.8391288720013108,
-      "bytes": 1133832,
-      "pages": 34,
-      "shards": 1
-    },
     {
       "seconds": 1.679156835001777,
       "bytes": 13329088,
@@ -562,6 +556,12 @@
       "bytes": 5769233,
       "pages": 1,
       "shards": 1
+    },
+    {
+      "seconds": 1.0994338210002752,
+      "bytes": 2592435,
+      "pages": 38,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791143623306220418-eb05a537/train.jsonl

@@ -250,3 +250,4 @@
 {"step": 710000, "loss": 0.5671007633209229, "lr": 8e-05, "grad_norm": 0.22037535991920845, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 103923858867, "tokens_per_second": 152737.25572226194, "data_wait_s": 1.7985999875236303e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 710100, "loss": 0.5553442239761353, "lr": 8e-05, "grad_norm": 0.21565878596663604, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 103938573748, "tokens_per_second": 6114.882245130367, "data_wait_s": 1.7717000446282327e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 710200, "loss": 0.5633423924446106, "lr": 8e-05, "grad_norm": 0.2694930185439089, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 103953324137, "tokens_per_second": 153774.64411334082, "data_wait_s": 1.759799852152355e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 710300, "loss": 0.5289316773414612, "lr": 8e-05, "grad_norm": 0.20650412550663017, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 103967896112, "tokens_per_second": 151435.8393303304, "data_wait_s": 1.8667000404093415e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791143623.3eb6ddb3-ffce-42e5-bd48-c0d653189e98-8pb7d.369.0.1791143623306220418-eb05a537

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:8e9200fe37b4c83f48f03165463f65e4f28b5c0dee4ffc9dcf1205a61266dd9d
-size 537816
+oid sha256:3b5972e362498e6337189d6ce59bd37142986fda8cbc1b794179fbef89a16ddb
+size 539934