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": 75763.08410639188,
-      "required_pages_per_second_upper_estimate": 0.00656695211056091
+      "required_token_positions_per_second": 75751.00285788387,
+      "required_pages_per_second_upper_estimate": 0.006565904938559881
     },
     {
       "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": 9470.385513298985,
-      "required_pages_per_second_upper_estimate": 0.0037778783761365027
+      "required_token_positions_per_second": 9468.875357235484,
+      "required_pages_per_second_upper_estimate": 0.0037772759523039268
     },
     {
       "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": 18940.77102659797,
-      "required_pages_per_second_upper_estimate": 0.004381614030818558
+      "required_token_positions_per_second": 18937.750714470967,
+      "required_pages_per_second_upper_estimate": 0.004380915334763654
     },
     {
       "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": 28411.156539896954,
-      "required_pages_per_second_upper_estimate": 0.001730090371461464
+      "required_token_positions_per_second": 28406.62607170645,
+      "required_pages_per_second_upper_estimate": 0.0017298144897181919
     },
     {
       "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": 28411.156539896954,
-      "required_pages_per_second_upper_estimate": 0.002187169477556209
+      "required_token_positions_per_second": 28406.62607170645,
+      "required_pages_per_second_upper_estimate": 0.0021868207095737655
     },
     {
       "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": 28411.156539896954,
-      "required_pages_per_second_upper_estimate": 0.3527714782012858
+      "required_token_positions_per_second": 28406.62607170645,
+      "required_pages_per_second_upper_estimate": 0.3527152249488614
     }
   ],
   "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": 0.5003399182188559,
-    "producer_p95_seconds": 0.3820157649970497,
-    "gpu_update_seconds": 0.9731388428758514,
-    "producer_batches_per_second": 1.998641250851757,
-    "gpu_batches_per_second": 1.0276025947590044,
-    "producer_headroom": 1.944955434177823,
+    "producer_batch_seconds": 0.11386841564836914,
+    "producer_p95_seconds": 0.3714156709975214,
+    "gpu_update_seconds": 0.9732940452065141,
+    "producer_batches_per_second": 8.782066513404786,
+    "gpu_batches_per_second": 1.0274387323389196,
+    "producer_headroom": 8.54753304210441,
     "sustainable": true,
-    "buffer_seconds": 60.33460825830279,
+    "buffer_seconds": 60.34423080280387,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 471347.01268786454,
-    "upload_bytes_per_second": 5570632.752721683,
+    "generation_bytes_per_second": 471304.1224093139,
+    "upload_bytes_per_second": 5560344.660162394,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 101.14009652592614,
+    "estimated_full_shards_per_hour": 101.13089327058375,
     "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": 1.6018576979986392,
-      "bytes": 11859045,
-      "pages": 34,
-      "shards": 4
-    },
     {
       "seconds": 0.8893314249980904,
       "bytes": 1064415,
@@ -562,6 +556,12 @@
       "bytes": 5862208,
       "pages": 37,
       "shards": 2
+    },
+    {
+      "seconds": 1.9020824750041356,
+      "bytes": 12748087,
+      "pages": 37,
+      "shards": 4
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790434090210070261-a05fc3fb/train.jsonl

@@ -342,3 +342,4 @@
 {"step": 74200, "loss": 0.6501174326986074, "lr": 0.0003, "grad_norm": 0.13225831017210676, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 10859205180, "tokens_per_second": 149691.51110297305, "data_wait_s": 1.1162999726366252e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 74300, "loss": 0.5940321460366249, "lr": 0.0003, "grad_norm": 0.1324297217732014, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 10873598026, "tokens_per_second": 147204.36127849793, "data_wait_s": 8.448005246464163e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 74400, "loss": 0.6435681488364935, "lr": 0.0003, "grad_norm": 0.13936324521999424, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 10888103339, "tokens_per_second": 148194.6666601274, "data_wait_s": 1.4944002032279968e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 74500, "loss": 0.5958024971187115, "lr": 0.0003, "grad_norm": 0.13140623676506977, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 10902622843, "tokens_per_second": 148675.76287073368, "data_wait_s": 7.68699828768149e-06, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790434090.372bd512-57cc-4928-b44a-3b375a0fa071-5spg5.451.0.1790434090210070261-a05fc3fb

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:badd7f316f5f1e5de61d8ec6e3178f1d409186f1e9aba0e181481e3c6c45a351
-size 733455
+oid sha256:3feb91aed0005ef8abf90c0281666ce6794c3f6276012ed10d2303b07d24d223
+size 735573