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": 76897.19231786462,
-      "required_pages_per_second_upper_estimate": 0.0053199261906183674
+      "required_token_positions_per_second": 77454.43613947435,
+      "required_pages_per_second_upper_estimate": 0.005358477611181132
     },
     {
       "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": 9612.149039733078,
-      "required_pages_per_second_upper_estimate": 0.004024406078439147
+      "required_token_positions_per_second": 9681.804517434293,
+      "required_pages_per_second_upper_estimate": 0.0040535693723808665
     },
     {
       "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": 19224.298079466156,
-      "required_pages_per_second_upper_estimate": 0.004003914311119595
+      "required_token_positions_per_second": 19363.609034868587,
+      "required_pages_per_second_upper_estimate": 0.00403292910925297
     },
     {
       "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": 28836.447119199238,
-      "required_pages_per_second_upper_estimate": 0.0017287009740907038
+      "required_token_positions_per_second": 29045.41355230288,
+      "required_pages_per_second_upper_estimate": 0.0017412281926819989
     },
     {
       "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": 28836.447119199238,
-      "required_pages_per_second_upper_estimate": 0.00224190692539153
+      "required_token_positions_per_second": 29045.41355230288,
+      "required_pages_per_second_upper_estimate": 0.0022581531464191374
     },
     {
       "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": 28836.447119199238,
-      "required_pages_per_second_upper_estimate": 0.36988298148047405
+      "required_token_positions_per_second": 29045.41355230288,
+      "required_pages_per_second_upper_estimate": 0.37256337851365273
     }
   ],
   "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": 63,
+    "prefetch_batches": 64,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.09684803360914884,
-    "producer_p95_seconds": 0.3354068600019673,
-    "gpu_update_seconds": 0.9587866315747348,
-    "producer_batches_per_second": 10.325454867113937,
-    "gpu_batches_per_second": 1.0429849218460372,
-    "producer_headroom": 9.899908091517121,
+    "producer_batch_seconds": 0.11728914106987531,
+    "producer_p95_seconds": 0.33895457800099393,
+    "gpu_update_seconds": 0.9518886673867969,
+    "producer_batches_per_second": 8.525938470333305,
+    "gpu_batches_per_second": 1.0505430248952141,
+    "producer_headroom": 8.115744208747396,
     "sustainable": true,
-    "buffer_seconds": 60.40355778920829,
+    "buffer_seconds": 60.920874712755,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 336622.4489792823,
-    "upload_bytes_per_second": 4482859.371013638,
+    "generation_bytes_per_second": 337199.2572232411,
+    "upload_bytes_per_second": 4515200.724149797,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 72.23134138139584,
+    "estimated_full_shards_per_hour": 72.35511100314068,
     "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": 1.1245584639982553,
-      "bytes": 2759850,
-      "pages": 40,
-      "shards": 2
-    },
     {
       "seconds": 1.2774146450028638,
       "bytes": 6152659,
@@ -562,6 +556,12 @@
       "bytes": 6150725,
       "pages": 33,
       "shards": 2
+    },
+    {
+      "seconds": 1.2803718870054581,
+      "bytes": 6573662,
+      "pages": 38,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791254470898546162-7724d9f6/train.jsonl

@@ -417,3 +417,4 @@
 {"step": 801700, "loss": 0.5376384854316711, "lr": 3e-05, "grad_norm": 0.25954194731941643, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 117345270836, "tokens_per_second": 151934.27009175546, "data_wait_s": 1.455999881727621e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 801800, "loss": 0.505940318107605, "lr": 3e-05, "grad_norm": 0.27344645578183735, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 117359953415, "tokens_per_second": 153007.89898270488, "data_wait_s": 9.234994649887085e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 801900, "loss": 0.5472398400306702, "lr": 3e-05, "grad_norm": 0.26347785421185776, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 117374697652, "tokens_per_second": 153732.81004057097, "data_wait_s": 1.4474004274234176e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 802000, "loss": 0.5652455687522888, "lr": 3e-05, "grad_norm": 0.289194650408052, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 117389459163, "tokens_per_second": 152971.83202043435, "data_wait_s": 1.1332995200064033e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791254470.1e0c9aa3-e42e-441e-9e17-2e25895f438d-9bhv5.366.0.1791254470898546162-7724d9f6

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:6d7cedbe0d78ee2c8150c47973cd5b9a2a2a564542a1acd6d6effd79d4740337
-size 893871
+oid sha256:6e56ef1b6ef2a20951f10dcad8c7abfef497508825d94066c4c6b945591fc711
+size 895989