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": 75235.47096800515,
-      "required_pages_per_second_upper_estimate": 0.005204964451910342
+      "required_token_positions_per_second": 76299.21578486163,
+      "required_pages_per_second_upper_estimate": 0.005278556786568504
     },
     {
       "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": 9404.433871000643,
-      "required_pages_per_second_upper_estimate": 0.003937440074876843
+      "required_token_positions_per_second": 9537.401973107704,
+      "required_pages_per_second_upper_estimate": 0.003993111042539349
     },
     {
       "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": 18808.867742001286,
-      "required_pages_per_second_upper_estimate": 0.003917391127460396
+      "required_token_positions_per_second": 19074.80394621541,
+      "required_pages_per_second_upper_estimate": 0.003972778625588875
     },
     {
       "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": 28213.301613001928,
-      "required_pages_per_second_upper_estimate": 0.0016913443524822702
+      "required_token_positions_per_second": 28612.20591932311,
+      "required_pages_per_second_upper_estimate": 0.0017152580565546168
     },
     {
       "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": 28213.301613001928,
-      "required_pages_per_second_upper_estimate": 0.0021934601032121007
+      "required_token_positions_per_second": 28612.20591932311,
+      "required_pages_per_second_upper_estimate": 0.00222447315843396
     },
     {
       "source": "cortex",
@@ -91,9 +91,9 @@
       "read_seconds": 0.04061180899998362,
       "tokens_per_second": 910315.184299418,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
-      "estimated_pages_for_buffer": 22,
-      "required_token_positions_per_second": 28213.301613001928,
-      "required_pages_per_second_upper_estimate": 0.3618899400084905
+      "estimated_pages_for_buffer": 23,
+      "required_token_positions_per_second": 28612.20591932311,
+      "required_pages_per_second_upper_estimate": 0.3670066561399047
     }
   ],
   "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": 62,
+    "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.0876339012343692,
-    "producer_p95_seconds": 0.23487457600276684,
-    "gpu_update_seconds": 0.9799632945921717,
-    "producer_batches_per_second": 11.411109010491128,
-    "gpu_batches_per_second": 1.0204463835721183,
-    "producer_headroom": 11.182467980871301,
+    "producer_batch_seconds": 0.13938924836699584,
+    "producer_p95_seconds": 0.2526524419990892,
+    "gpu_update_seconds": 0.9663008884375482,
+    "producer_batches_per_second": 7.174154475437841,
+    "gpu_batches_per_second": 1.0348743460403325,
+    "producer_headroom": 6.932391843403798,
     "sustainable": true,
-    "buffer_seconds": 60.757724264714646,
+    "buffer_seconds": 60.876955971565536,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 80,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 357533.31169293105,
-    "upload_bytes_per_second": 4265533.408157548,
+    "generation_bytes_per_second": 367602.2309508391,
+    "upload_bytes_per_second": 4259058.1787644,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 76.71832573977422,
+    "estimated_full_shards_per_hour": 78.87888142007714,
     "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": 1.0229502059992228,
-      "bytes": 2562647,
-      "pages": 37,
-      "shards": 2
-    },
     {
       "seconds": 1.0365147789998446,
       "bytes": 6395327,
@@ -562,6 +556,12 @@
       "bytes": 7346473,
       "pages": 31,
       "shards": 3
+    },
+    {
+      "seconds": 0.8391288720013108,
+      "bytes": 1133832,
+      "pages": 34,
+      "shards": 1
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791143623306220418-eb05a537/train.jsonl

@@ -173,3 +173,4 @@
 {"step": 702300, "loss": 0.5484373569488525, "lr": 8e-05, "grad_norm": 0.2223881358844163, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 102797076313, "tokens_per_second": 153476.14390037546, "data_wait_s": 2.1927000489085913e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 702400, "loss": 0.5829904079437256, "lr": 8e-05, "grad_norm": 0.22907028830171255, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 102811806027, "tokens_per_second": 152245.37009005426, "data_wait_s": 1.8148999515688047e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 702500, "loss": 0.5683314800262451, "lr": 8e-05, "grad_norm": 0.22214562817704783, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 102826531332, "tokens_per_second": 152768.12480804193, "data_wait_s": 1.2638000043807551e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 702600, "loss": 0.5840139389038086, "lr": 8e-05, "grad_norm": 0.22603571300496714, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 102841166411, "tokens_per_second": 152453.53514426816, "data_wait_s": 1.4353998267324641e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791143623.3eb6ddb3-ffce-42e5-bd48-c0d653189e98-8pb7d.369.0.1791143623306220418-eb05a537

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:9553d9947e6308e395dde8248439cfa3c501b35bcf8113e903c0564b1c1afd9e
-size 373164
+oid sha256:ed3c738e2d121418b772876087afa32a0dc4085ece7e28754a49430fa2dcec71
+size 375282