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": 77246.34714449453,
-      "required_pages_per_second_upper_estimate": 0.005344081531675439
+      "required_token_positions_per_second": 77219.7994208351,
+      "required_pages_per_second_upper_estimate": 0.005342244898553474
     },
     {
       "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": 9655.793393061816,
-      "required_pages_per_second_upper_estimate": 0.0040426790577801535
+      "required_token_positions_per_second": 9652.474927604388,
+      "required_pages_per_second_upper_estimate": 0.004041289685590567
     },
     {
       "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": 19311.586786123633,
-      "required_pages_per_second_upper_estimate": 0.004022094246758353
+      "required_token_positions_per_second": 19304.949855208775,
+      "required_pages_per_second_upper_estimate": 0.0040207119490764266
     },
     {
       "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": 28967.380179185446,
-      "required_pages_per_second_upper_estimate": 0.0017365502111136703
+      "required_token_positions_per_second": 28957.42478281316,
+      "required_pages_per_second_upper_estimate": 0.0017359534003021618
     },
     {
       "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": 28967.380179185446,
-      "required_pages_per_second_upper_estimate": 0.002252086394891791
+      "required_token_positions_per_second": 28957.42478281316,
+      "required_pages_per_second_upper_estimate": 0.0022513124066130064
     },
     {
       "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": 28967.380179185446,
-      "required_pages_per_second_upper_estimate": 0.37156245018901046
+      "required_token_positions_per_second": 28957.42478281316,
+      "required_pages_per_second_upper_estimate": 0.3714347530536186
     }
   ],
   "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": 0.12829153139844607,
-    "producer_p95_seconds": 0.3089481210008671,
-    "gpu_update_seconds": 0.9544528993984244,
-    "producer_batches_per_second": 7.794746770106078,
-    "gpu_batches_per_second": 1.0477206372679921,
-    "producer_headroom": 7.43971865480425,
+    "producer_batch_seconds": 0.08546301957781566,
+    "producer_p95_seconds": 0.30518934700012323,
+    "gpu_update_seconds": 0.9547810348249499,
+    "producer_batches_per_second": 11.700967329962891,
+    "gpu_batches_per_second": 1.0473605607209622,
+    "producer_headroom": 11.1718616957549,
     "sustainable": true,
-    "buffer_seconds": 60.130532662100734,
+    "buffer_seconds": 60.15120519397185,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 78,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 525886.5827572806,
-    "upload_bytes_per_second": 6133694.282782401,
+    "generation_bytes_per_second": 525631.0455944852,
+    "upload_bytes_per_second": 6080292.466293754,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 112.84301864661037,
+    "estimated_full_shards_per_hour": 112.78818632007521,
     "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": 1.1750758769994718,
-      "bytes": 10630556,
-      "pages": 35,
-      "shards": 3
-    },
     {
       "seconds": 1.1581052289984655,
       "bytes": 2607511,
@@ -562,6 +556,12 @@
       "bytes": 6852339,
       "pages": 35,
       "shards": 3
+    },
+    {
+      "seconds": 1.050476451000577,
+      "bytes": 6070361,
+      "pages": 29,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790914434605211575-090606d4/train.jsonl

@@ -300,3 +300,4 @@
 {"step": 515000, "loss": 0.6022728681564331, "lr": 0.0003, "grad_norm": 0.16161594023367803, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 75378632707, "tokens_per_second": 153146.02722571362, "data_wait_s": 1.647099998081103e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 515100, "loss": 0.6161287426948547, "lr": 0.0003, "grad_norm": 0.18326037244278115, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 75393312170, "tokens_per_second": 145675.3699440447, "data_wait_s": 1.7000998923322186e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 515200, "loss": 0.7914197444915771, "lr": 0.0003, "grad_norm": 0.1800201411212684, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 75408040149, "tokens_per_second": 153546.89705949405, "data_wait_s": 1.4504003047477454e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 515300, "loss": 0.6192801594734192, "lr": 0.0003, "grad_norm": 0.14641507097440298, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 75422702197, "tokens_per_second": 151600.634916967, "data_wait_s": 1.7659000150160864e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1790914434.4dec8a9d-3312-413a-9119-9058bec1adb7-bx5st.8214.0.1790914434605211575-090606d4

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:3b71a5afb8400e1ce96e836652d705cc89c7593b2e0d0883f001605a4c5765bb
-size 644499
+oid sha256:4454d089443351f72feb45a48ad84e39b3d5f5036bb741829e62e037aab627ca
+size 646617