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": 76803.87422740913,
-      "required_pages_per_second_upper_estimate": 0.0053134702285928545
+      "required_token_positions_per_second": 76687.96584220369,
+      "required_pages_per_second_upper_estimate": 0.005305451417559835
     },
     {
       "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": 9600.484278426142,
-      "required_pages_per_second_upper_estimate": 0.004019522286467848
+      "required_token_positions_per_second": 9585.995730275461,
+      "required_pages_per_second_upper_estimate": 0.004013456233912448
     },
     {
       "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": 19200.968556852284,
-      "required_pages_per_second_upper_estimate": 0.00399905538680001
+      "required_token_positions_per_second": 19171.991460550922,
+      "required_pages_per_second_upper_estimate": 0.003993020221817854
     },
     {
       "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": 28801.452835278425,
-      "required_pages_per_second_upper_estimate": 0.0017266031202028288
+      "required_token_positions_per_second": 28757.98719082638,
+      "required_pages_per_second_upper_estimate": 0.001723997421186125
     },
     {
       "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": 28801.452835278425,
-      "required_pages_per_second_upper_estimate": 0.0022391862737402746
+      "required_token_positions_per_second": 28757.98719082638,
+      "required_pages_per_second_upper_estimate": 0.002235807011069293
     },
     {
       "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": 28801.452835278425,
-      "required_pages_per_second_upper_estimate": 0.3694341123802725
+      "required_token_positions_per_second": 28757.98719082638,
+      "required_pages_per_second_upper_estimate": 0.3688765817630146
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 704566,
-    "producer_batch_seconds": 0.06766399788283906,
-    "producer_p95_seconds": 0.16793836199940415,
-    "gpu_update_seconds": 0.9599515746002375,
-    "producer_batches_per_second": 14.77890800557648,
-    "gpu_batches_per_second": 1.0417192142389478,
-    "producer_headroom": 14.187036010825196,
+    "producer_batch_seconds": 0.07497538868749132,
+    "producer_p95_seconds": 0.1706010130001232,
+    "gpu_update_seconds": 0.9614024728691561,
+    "producer_batches_per_second": 13.337710114023553,
+    "gpu_batches_per_second": 1.040147106149681,
+    "producer_headroom": 12.822907486034198,
     "sustainable": true,
-    "buffer_seconds": 60.47694919981496,
+    "buffer_seconds": 60.56835579075684,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 79,
+    "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 352969.2784105951,
-    "upload_bytes_per_second": 4907337.778785286,
+    "generation_bytes_per_second": 392232.3209518049,
+    "upload_bytes_per_second": 4862718.423654909,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 75.7389904426421,
+    "estimated_full_shards_per_hour": 84.16392537513362,
     "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": [
@@ -490,6 +490,12 @@
       "bytes": 11191562,
       "pages": 43,
       "shards": 3
+    },
+    {
+      "seconds": 1.1523795839993909,
+      "bytes": 2573171,
+      "pages": 37,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790824801708454555-41033672/train.jsonl

@@ -61,3 +61,4 @@
 {"step": 403200, "loss": 0.6168279647827148, "lr": 0.0003, "grad_norm": 0.17651831990613814, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 59017911201, "tokens_per_second": 153169.97719119757, "data_wait_s": 6.6350003180559725e-06, "gpu_peak_gb": 89.37861251831055}
 {"step": 403300, "loss": 0.5866488814353943, "lr": 0.0003, "grad_norm": 0.17743531496723028, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 59032566431, "tokens_per_second": 152560.7045594227, "data_wait_s": 8.998999874165747e-06, "gpu_peak_gb": 89.37861251831055}
 {"step": 403400, "loss": 0.5906795859336853, "lr": 0.0003, "grad_norm": 0.1922095941150668, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 59047255003, "tokens_per_second": 152823.18197802178, "data_wait_s": 1.4312000530480873e-05, "gpu_peak_gb": 89.37861251831055}
+{"step": 403500, "loss": 0.4693165123462677, "lr": 0.0003, "grad_norm": 0.18263407072278962, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 59061969613, "tokens_per_second": 153560.01743050016, "data_wait_s": 8.820999937597662e-06, "gpu_peak_gb": 89.37861251831055}

runs/void-byte-v1/events.out.tfevents.1790824801.afc92292-ba09-4e51-8d82-bc1983323308-sdcdh.469.0.1790824801708454555-41033672

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:5bc76c3094d31b6d1fd51ec0195614a9d0c90067753d70d77b44a387824fcc1e
-size 134382
+oid sha256:3aea1ceabc357a0193c4917ce01a292ffcbd112f24f50854334c97a7d6b7dc40
+size 136500