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": 77414.84927352374,
-      "required_pages_per_second_upper_estimate": 0.0053557388999404855
+      "required_token_positions_per_second": 77149.99603889415,
+      "required_pages_per_second_upper_estimate": 0.005337415738624639
     },
     {
       "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": 9676.856159190467,
-      "required_pages_per_second_upper_estimate": 0.004051497598117647
+      "required_token_positions_per_second": 9643.74950486177,
+      "required_pages_per_second_upper_estimate": 0.004037636533295779
     },
     {
       "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": 19353.712318380934,
-      "required_pages_per_second_upper_estimate": 0.004030867884202556
+      "required_token_positions_per_second": 19287.49900972354,
+      "required_pages_per_second_upper_estimate": 0.004017077398171595
     },
     {
       "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": 29030.5684775714,
-      "required_pages_per_second_upper_estimate": 0.0017403382531189632
+      "required_token_positions_per_second": 28931.248514585306,
+      "required_pages_per_second_upper_estimate": 0.0017343841729907494
     },
     {
       "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": 29030.5684775714,
-      "required_pages_per_second_upper_estimate": 0.0022569990071553475
+      "required_token_positions_per_second": 28931.248514585306,
+      "required_pages_per_second_upper_estimate": 0.002249277316895527
     },
     {
       "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": 29030.5684775714,
-      "required_pages_per_second_upper_estimate": 0.37237296183439667
+      "required_token_positions_per_second": 28931.248514585306,
+      "required_pages_per_second_upper_estimate": 0.37109899199067875
     }
   ],
   "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": 0.12020311989070365,
-    "producer_p95_seconds": 0.2332568879974133,
-    "gpu_update_seconds": 0.952375425281818,
-    "producer_batches_per_second": 8.319251621000053,
-    "gpu_batches_per_second": 1.05000609366216,
-    "producer_headroom": 7.9230508005763784,
+    "producer_batch_seconds": 0.34067203547633085,
+    "producer_p95_seconds": 0.23187028699976509,
+    "gpu_update_seconds": 0.9556448967648813,
+    "producer_batches_per_second": 2.9353744829738977,
+    "gpu_batches_per_second": 1.0464137917601746,
+    "producer_headroom": 2.805175644747857,
     "sustainable": true,
-    "buffer_seconds": 60.95202721803635,
+    "buffer_seconds": 60.20562849618752,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 87,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 454600.16813522583,
-    "upload_bytes_per_second": 5669323.015154715,
+    "generation_bytes_per_second": 405104.8895829964,
+    "upload_bytes_per_second": 5701781.212804185,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 97.54661353151876,
+    "estimated_full_shards_per_hour": 86.92607894532603,
     "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": 1.2687110730003042,
-      "bytes": 6191110,
-      "pages": 36,
-      "shards": 2
-    },
     {
       "seconds": 1.3943383940004423,
       "bytes": 6935244,
@@ -562,6 +556,12 @@
       "bytes": 1206466,
       "pages": 34,
       "shards": 1
+    },
+    {
+      "seconds": 1.7537132789984753,
+      "bytes": 11388396,
+      "pages": 34,
+      "shards": 4
     }
   ],
   "cache_key": {

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

@@ -195,3 +195,4 @@
 {"step": 704500, "loss": 0.6048893928527832, "lr": 8e-05, "grad_norm": 0.24395356854016637, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 103119160997, "tokens_per_second": 152173.85381440507, "data_wait_s": 1.1691001418512315e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 704600, "loss": 0.534127950668335, "lr": 8e-05, "grad_norm": 0.21791060682692254, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 103133713411, "tokens_per_second": 151894.78297041933, "data_wait_s": 1.5841000276850536e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 704700, "loss": 0.554182231426239, "lr": 8e-05, "grad_norm": 0.21033279899822088, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 103148434540, "tokens_per_second": 152963.37865608258, "data_wait_s": 1.5404002624563873e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 704800, "loss": 0.5306462049484253, "lr": 8e-05, "grad_norm": 0.20430684528132304, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 103163135910, "tokens_per_second": 152653.186544431, "data_wait_s": 1.72220024978742e-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 sha256:3361abd239e19d5bfbaa41bdb8f48b35de6efc11990a9d239dec649c298d6bc1
-size 419760
+oid sha256:463caf27be10e98813b7bc58c9c31010da472227afa641a64a1d038b396a872b
+size 421878