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": 73505.8020697132,
-      "required_pages_per_second_upper_estimate": 0.0063712966246506205
+      "required_token_positions_per_second": 72887.22926082663,
+      "required_pages_per_second_upper_estimate": 0.0063176803013347835
     },
     {
       "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": 9188.22525871415,
-      "required_pages_per_second_upper_estimate": 0.0036653204319108627
+      "required_token_positions_per_second": 9110.903657603329,
+      "required_pages_per_second_upper_estimate": 0.0036344756891667977
     },
     {
       "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": 18376.4505174283,
-      "required_pages_per_second_upper_estimate": 0.0042510684127239065
+      "required_token_positions_per_second": 18221.807315206657,
+      "required_pages_per_second_upper_estimate": 0.004215294429517334
     },
     {
       "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": 27564.67577614245,
-      "required_pages_per_second_upper_estimate": 0.0016785441340901525
+      "required_token_positions_per_second": 27332.710972809986,
+      "required_pages_per_second_upper_estimate": 0.001664418694592472
     },
     {
       "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": 27564.67577614245,
-      "required_pages_per_second_upper_estimate": 0.0021220050451536614
+      "required_token_positions_per_second": 27332.710972809986,
+      "required_pages_per_second_upper_estimate": 0.0021041477524734553
     },
     {
       "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": 27564.67577614245,
-      "required_pages_per_second_upper_estimate": 0.3422610201043303
+      "required_token_positions_per_second": 27332.710972809986,
+      "required_pages_per_second_upper_estimate": 0.3393807935831976
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 60,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 690670,
-    "producer_batch_seconds": 0.11714230824185279,
+    "producer_batch_seconds": 0.12590279293755202,
     "producer_p95_seconds": 0.393772668998281,
-    "gpu_update_seconds": 1.0030228624684085,
-    "producer_batches_per_second": 8.536625366263001,
-    "gpu_batches_per_second": 0.9969862476903374,
-    "producer_headroom": 8.562430410689542,
+    "gpu_update_seconds": 1.0115352270582914,
+    "producer_batches_per_second": 7.942635557703644,
+    "gpu_batches_per_second": 0.98859631701425,
+    "producer_headroom": 8.034255662303014,
     "sustainable": true,
-    "buffer_seconds": 60.18137174810451,
+    "buffer_seconds": 60.692113623497484,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 86,
+    "upload_pages_per_commit": 84,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 530139.7203919418,
-    "upload_bytes_per_second": 3822498.9789663344,
+    "generation_bytes_per_second": 553347.5295024142,
+    "upload_bytes_per_second": 3743462.5350851677,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 113.75564297503178,
+    "estimated_full_shards_per_hour": 118.73549855999298,
     "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": 1.1155471289966954,
-      "bytes": 10846938,
-      "pages": 39,
-      "shards": 3
-    },
     {
       "seconds": 1.1072990959983144,
       "bytes": 7478866,
@@ -562,6 +556,12 @@
       "bytes": 5700254,
       "pages": 28,
       "shards": 2
+    },
+    {
+      "seconds": 0.8491491380045773,
+      "bytes": 1087747,
+      "pages": 34,
+      "shards": 1
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790738827394728502-c744ae20/train.jsonl

@@ -358,3 +358,4 @@
 {"step": 350800, "loss": 0.5778562966734171, "lr": 0.0003, "grad_norm": 0.16013492337526053, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 51347915416, "tokens_per_second": 148592.92630792255, "data_wait_s": 1.1156000255141407e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 350900, "loss": 0.610313281416893, "lr": 0.0003, "grad_norm": 0.15464998559750495, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 51362586949, "tokens_per_second": 149301.19764136526, "data_wait_s": 8.33100057207048e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 351000, "loss": 0.5867506377398968, "lr": 0.0003, "grad_norm": 0.16866896099953796, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 51377137233, "tokens_per_second": 147668.4150535501, "data_wait_s": 8.555995009373873e-06, "gpu_peak_gb": 85.46427917480469}
+{"step": 351100, "loss": 0.5874341055750847, "lr": 0.0003, "grad_norm": 0.1632072478115318, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 51391790571, "tokens_per_second": 148151.8317907659, "data_wait_s": 1.4305005606729537e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790738827.1a46e860-412e-40b5-8c01-42d24543f8c9-dfnph.319.0.1790738827394728502-c744ae20

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:d613b12a7e31e33976d7a5cc9f4dde82a0862abba87caf5c9d4ab7735f31e681
-size 768126
+oid sha256:b9f29811cbd7f9532e1ec95b8f091af2125086a9562a1e85b49cb9af03e8c21e
+size 770244