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": 77369.58771953837,
-      "required_pages_per_second_upper_estimate": 0.005352607600614502
+      "required_token_positions_per_second": 76354.96370082689,
+      "required_pages_per_second_upper_estimate": 0.005282413556747967
     },
     {
       "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": 9671.198464942296,
-      "required_pages_per_second_upper_estimate": 0.004049128839682028
+      "required_token_positions_per_second": 9544.370462603361,
+      "required_pages_per_second_upper_estimate": 0.0039960286035725726
     },
     {
       "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": 19342.39692988459,
-      "required_pages_per_second_upper_estimate": 0.00402851118718563
+      "required_token_positions_per_second": 19088.740925206723,
+      "required_pages_per_second_upper_estimate": 0.003975681330769913
     },
     {
       "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": 29013.595394826887,
-      "required_pages_per_second_upper_estimate": 0.0017393207427248264
+      "required_token_positions_per_second": 28633.111387810084,
+      "required_pages_per_second_upper_estimate": 0.0017165113074696088
     },
     {
       "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": 29013.595394826887,
-      "required_pages_per_second_upper_estimate": 0.0022556794246286625
+      "required_token_positions_per_second": 28633.111387810084,
+      "required_pages_per_second_upper_estimate": 0.0022260984666553846
     },
     {
       "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": 29013.595394826887,
-      "required_pages_per_second_upper_estimate": 0.37215524935322647
+      "required_token_positions_per_second": 28633.111387810084,
+      "required_pages_per_second_upper_estimate": 0.3672748090431124
     }
   ],
   "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": 720750,
-    "producer_batch_seconds": 0.3458197621406214,
-    "producer_p95_seconds": 0.17324873300003674,
-    "gpu_update_seconds": 0.9529325691544464,
-    "producer_batches_per_second": 2.8916797403653525,
-    "gpu_batches_per_second": 1.0493921945466902,
-    "producer_headroom": 2.7555758041582177,
+    "producer_batch_seconds": 0.049891821398382774,
+    "producer_p95_seconds": 0.1716755050001666,
+    "gpu_update_seconds": 0.9655953775170423,
+    "producer_batches_per_second": 20.043365264520382,
+    "gpu_batches_per_second": 1.0356304755428993,
+    "producer_headroom": 19.35378084930653,
     "sustainable": true,
-    "buffer_seconds": 60.03475185673012,
+    "buffer_seconds": 60.832508783573665,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 85,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 355801.83317358646,
-    "upload_bytes_per_second": 5053461.385983229,
+    "generation_bytes_per_second": 356651.565915334,
+    "upload_bytes_per_second": 5182935.048544978,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 76.3467907562799,
+    "estimated_full_shards_per_hour": 76.52912362189306,
     "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": [
@@ -334,6 +334,12 @@
       "bytes": 806070,
       "pages": 24,
       "shards": 1
+    },
+    {
+      "seconds": 1.260675972000172,
+      "bytes": 10735591,
+      "pages": 41,
+      "shards": 3
     }
   ],
   "cache_key": {

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

@@ -66,3 +66,4 @@
 {"step": 766600, "loss": 0.5433411002159119, "lr": 3e-05, "grad_norm": 0.24735584486766685, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 112207648241, "tokens_per_second": 152666.7631765897, "data_wait_s": 1.5093999536475167e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 766700, "loss": 0.5982465744018555, "lr": 3e-05, "grad_norm": 0.24279012298017602, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 112222326615, "tokens_per_second": 153089.86859662223, "data_wait_s": 1.2883000636065844e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 766800, "loss": 0.5532582402229309, "lr": 3e-05, "grad_norm": 0.2568762408443731, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 112237081339, "tokens_per_second": 153914.97725973866, "data_wait_s": 9.608999789634254e-06, "gpu_peak_gb": 89.37899398803711}
+{"step": 766900, "loss": 0.5498709678649902, "lr": 3e-05, "grad_norm": 0.23755576228424988, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 112251765204, "tokens_per_second": 153039.06429852932, "data_wait_s": 1.1963999895669986e-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 sha256:3ad9f37904b74a7f7f899cfff2524fb58f3d6eda24dd041f389f4024fcf0d5d8
-size 144972
+oid sha256:f50c54288b506d6839366dc82c5b80465dea66dfc39c5e31c2c067a5e3947b32
+size 147090