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": 77149.80488085745,
-      "required_pages_per_second_upper_estimate": 0.005337402513868112
+      "required_token_positions_per_second": 77063.00398372827,
+      "required_pages_per_second_upper_estimate": 0.005331397426398885
     },
     {
       "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": 9643.725610107182,
-      "required_pages_per_second_upper_estimate": 0.004037626529061012
+      "required_token_positions_per_second": 9632.875497966033,
+      "required_pages_per_second_upper_estimate": 0.00403308381368362
     },
     {
       "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": 19287.451220214363,
-      "required_pages_per_second_upper_estimate": 0.004017067444877127
+      "required_token_positions_per_second": 19265.750995932067,
+      "required_pages_per_second_upper_estimate": 0.004012547860432523
     },
     {
       "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": 28931.176830321547,
-      "required_pages_per_second_upper_estimate": 0.0017343798756285925
+      "required_token_positions_per_second": 28898.6264938981,
+      "required_pages_per_second_upper_estimate": 0.0017324285326614929
     },
     {
       "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": 28931.176830321547,
-      "required_pages_per_second_upper_estimate": 0.002249271743759325
+      "required_token_positions_per_second": 28898.6264938981,
+      "required_pages_per_second_upper_estimate": 0.0022467410982762014
     },
     {
       "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": 28931.176830321547,
-      "required_pages_per_second_upper_estimate": 0.3710980725019118
+      "required_token_positions_per_second": 28898.6264938981,
+      "required_pages_per_second_upper_estimate": 0.3706805517361001
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 704566,
-    "producer_batch_seconds": 0.1266549091562439,
-    "producer_p95_seconds": 0.1745979779989284,
-    "gpu_update_seconds": 0.9556472646153577,
-    "producer_batches_per_second": 7.895469718954053,
-    "gpu_batches_per_second": 1.0464111990133662,
-    "producer_headroom": 7.545284039771827,
+    "producer_batch_seconds": 0.07049417017965709,
+    "producer_p95_seconds": 0.1884536540001136,
+    "gpu_update_seconds": 0.9567236700968411,
+    "producer_batches_per_second": 14.185570203202078,
+    "gpu_batches_per_second": 1.0452338864980506,
+    "producer_headroom": 13.571670787223884,
     "sustainable": true,
-    "buffer_seconds": 60.20577767076753,
+    "buffer_seconds": 60.27359121610099,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 78,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 395922.12761421397,
-    "upload_bytes_per_second": 5730038.966956036,
+    "generation_bytes_per_second": 427400.1912407021,
+    "upload_bytes_per_second": 5694872.885711405,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 84.95567199058355,
+    "estimated_full_shards_per_hour": 91.71013167300985,
     "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": 1.0457134279999991,
-      "bytes": 7722680,
-      "pages": 38,
-      "shards": 3
-    },
     {
       "seconds": 1.1198342459999822,
       "bytes": 518226,
@@ -562,6 +556,12 @@
       "bytes": 6510962,
       "pages": 32,
       "shards": 3
+    },
+    {
+      "seconds": 1.2347332800000004,
+      "bytes": 6132473,
+      "pages": 36,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790989041640695387-36510d53/train.jsonl

@@ -89,3 +89,4 @@
 {"step": 566000, "loss": 0.5413221716880798, "lr": 0.0003, "grad_norm": 0.1595919349606399, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 82845245541, "tokens_per_second": 151979.03032959043, "data_wait_s": 1.5294999684556387e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 566100, "loss": 0.5647397041320801, "lr": 0.0003, "grad_norm": 0.15198237748100768, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 82859990226, "tokens_per_second": 154040.11309169207, "data_wait_s": 1.4417999409488402e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 566200, "loss": 0.5574049353599548, "lr": 0.0003, "grad_norm": 0.164634143063845, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 82874788475, "tokens_per_second": 153979.27378567585, "data_wait_s": 1.4764000297873281e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 566300, "loss": 0.5719775557518005, "lr": 0.0003, "grad_norm": 0.17521623141029388, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 82889345800, "tokens_per_second": 151561.8097086228, "data_wait_s": 1.414399957866408e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1790989041.fb9d4967-10f3-4fa9-9622-1d5a28f8589b-vh8db.427.0.1790989041640695387-36510d53

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:92ba317d2b053ceb58295a590634f6dcfb8c69a3a0cf50f1cbd52a193439acc3
-size 194469
+oid sha256:b536c409f6e1ee1c01ca6512ccff545e8955ebdec1eb81395bc82b0790046034
+size 196587