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": 77638.02404567161,
-      "required_pages_per_second_upper_estimate": 0.00537117864851449
+      "required_token_positions_per_second": 76555.0266324517,
+      "required_pages_per_second_upper_estimate": 0.005296254374567728
     },
     {
       "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": 9704.753005708952,
-      "required_pages_per_second_upper_estimate": 0.004063177425202537
+      "required_token_positions_per_second": 9569.378329056462,
+      "required_pages_per_second_upper_estimate": 0.004006498875032851
     },
     {
       "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": 19409.506011417903,
-      "required_pages_per_second_upper_estimate": 0.004042488239083526
+      "required_token_positions_per_second": 19138.756658112925,
+      "required_pages_per_second_upper_estimate": 0.003986098288930699
     },
     {
       "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": 29114.259017126853,
-      "required_pages_per_second_upper_estimate": 0.0017453553731772578
+      "required_token_positions_per_second": 28708.134987169386,
+      "required_pages_per_second_upper_estimate": 0.0017210088576967955
     },
     {
       "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": 29114.259017126853,
-      "required_pages_per_second_upper_estimate": 0.002263505578490004
+      "required_token_positions_per_second": 28708.134987169386,
+      "required_pages_per_second_upper_estimate": 0.0022319312215116315
     },
     {
       "source": "cortex",
@@ -92,8 +92,8 @@
       "tokens_per_second": 910315.184299418,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 23,
-      "required_token_positions_per_second": 29114.259017126853,
-      "required_pages_per_second_upper_estimate": 0.37344645421591377
+      "required_token_positions_per_second": 28708.134987169386,
+      "required_pages_per_second_upper_estimate": 0.3682371312216286
     }
   ],
   "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": 64,
+    "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.13055367253895156,
-    "producer_p95_seconds": 0.26719631500236574,
-    "gpu_update_seconds": 0.9496377697174327,
-    "producer_batches_per_second": 7.659684944532245,
-    "gpu_batches_per_second": 1.0530330952375164,
-    "producer_headroom": 7.273926127463798,
+    "producer_batch_seconds": 0.8933466412108544,
+    "producer_p95_seconds": 0.2676944689992524,
+    "gpu_update_seconds": 0.9630719659203499,
+    "producer_batches_per_second": 1.1193863097135353,
+    "gpu_batches_per_second": 1.0383440027188,
+    "producer_headroom": 1.0780495739201401,
     "sustainable": true,
-    "buffer_seconds": 60.776817261915696,
+    "buffer_seconds": 60.67353385298205,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 86,
+    "upload_pages_per_commit": 78,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 276929.98004016274,
-    "upload_bytes_per_second": 4630155.953714088,
+    "generation_bytes_per_second": 235841.00489855447,
+    "upload_bytes_per_second": 4597870.363164352,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 59.4227271166197,
+    "estimated_full_shards_per_hour": 50.6059895536182,
     "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": 1.2958958110029926,
-      "bytes": 5955386,
-      "pages": 27,
-      "shards": 2
-    },
     {
       "seconds": 1.584322863000125,
       "bytes": 6087710,
@@ -562,6 +556,12 @@
       "bytes": 5911958,
       "pages": 38,
       "shards": 2
+    },
+    {
+      "seconds": 1.1188029909972101,
+      "bytes": 2149233,
+      "pages": 21,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791399463488568880-054ba646/train.jsonl

@@ -291,3 +291,4 @@
 {"step": 914100, "loss": 0.5752699375152588, "lr": 3e-05, "grad_norm": 0.3338416774421297, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 133797973145, "tokens_per_second": 153718.12544436217, "data_wait_s": 9.086001227842644e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 914200, "loss": 0.5317642092704773, "lr": 3e-05, "grad_norm": 0.3284383063020626, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 133812466888, "tokens_per_second": 151018.95310672675, "data_wait_s": 9.747000149218366e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 914300, "loss": 0.511695384979248, "lr": 3e-05, "grad_norm": 0.3179288216994619, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 133826971546, "tokens_per_second": 123452.86378734301, "data_wait_s": 1.456199970562011e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 914400, "loss": 0.5224106907844543, "lr": 3e-05, "grad_norm": 0.35033095834050637, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 133841463019, "tokens_per_second": 151270.9553593144, "data_wait_s": 1.5842000721022487e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791399463.60f38c85-66c9-4fe6-8e58-7b3a56dc2cee-wtkw8.346.0.1791399463488568880-054ba646

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:0d7933f8ee81d5b803ef4307f11d3a1fe0d575d64e33ee57459687de995e79f6
-size 624654
+oid sha256:d98d7229e3a2f50b6ab69a85dada5d82f1827e155fbfd82a6e54f4e94a201eeb
+size 626772