Update TensorBoard training metrics

benchmark_cache/shard_scheduler.json

@@ -22,8 +22,8 @@
       "tokens_per_second": 1062967.6168812835,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 75297.13704684116,
-      "required_pages_per_second_upper_estimate": 0.006526565000371081
+      "required_token_positions_per_second": 75416.26173263152,
+      "required_pages_per_second_upper_estimate": 0.006536890426216633
     },
     {
       "source": "rewrite6",
@@ -36,8 +36,8 @@
       "tokens_per_second": 1206672.263377472,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 9412.142130855145,
-      "required_pages_per_second_upper_estimate": 0.0037546442200634854
+      "required_token_positions_per_second": 9427.03271657894,
+      "required_pages_per_second_upper_estimate": 0.0037605842973428036
     },
     {
       "source": "openmath",
@@ -50,8 +50,8 @@
       "tokens_per_second": 410042.7581134458,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 18824.28426171029,
-      "required_pages_per_second_upper_estimate": 0.004354666867858836
+      "required_token_positions_per_second": 18854.06543315788,
+      "required_pages_per_second_upper_estimate": 0.00436155621774252
     },
     {
       "source": "ultra_qa",
@@ -64,8 +64,8 @@
       "tokens_per_second": 2163486.5624068812,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28236.426392565434,
-      "required_pages_per_second_upper_estimate": 0.0017194502222272082
+      "required_token_positions_per_second": 28281.098149736823,
+      "required_pages_per_second_upper_estimate": 0.0017221704978643492
     },
     {
       "source": "ultra_style",
@@ -78,8 +78,8 @@
       "tokens_per_second": 1546694.2737633225,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28236.426392565434,
-      "required_pages_per_second_upper_estimate": 0.002173718267130624
+      "required_token_positions_per_second": 28281.098149736823,
+      "required_pages_per_second_upper_estimate": 0.002177157222657016
     },
     {
       "source": "cortex",
@@ -92,8 +92,8 @@
       "tokens_per_second": 1015177.8395898147,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 22,
-      "required_token_positions_per_second": 28236.426392565434,
-      "required_pages_per_second_upper_estimate": 0.35060191455561335
+      "required_token_positions_per_second": 28281.098149736823,
+      "required_pages_per_second_upper_estimate": 0.3511565882729283
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 690670,
-    "producer_batch_seconds": 0.08382809289840054,
-    "producer_p95_seconds": 0.2169164679999085,
-    "gpu_update_seconds": 0.9791607342804411,
-    "producer_batches_per_second": 11.929175118083597,
-    "gpu_batches_per_second": 1.0212827832959142,
-    "producer_headroom": 11.680579867982702,
+    "producer_batch_seconds": 0.1492102944296505,
+    "producer_p95_seconds": 0.19664406399897416,
+    "gpu_update_seconds": 0.9776140888736065,
+    "producer_batches_per_second": 6.701950450687428,
+    "gpu_batches_per_second": 1.0228985152537913,
+    "producer_headroom": 6.551921183524846,
     "sustainable": true,
-    "buffer_seconds": 60.70796552538735,
+    "buffer_seconds": 60.6120735101636,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 76,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 326864.79560704,
-    "upload_bytes_per_second": 5548334.800680078,
+    "generation_bytes_per_second": 349267.9366177436,
+    "upload_bytes_per_second": 5549040.418239836,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 70.13757611425781,
+    "estimated_full_shards_per_hour": 74.94476865672331,
     "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": [
@@ -496,6 +496,12 @@
       "bytes": 9581963,
       "pages": 33,
       "shards": 3
+    },
+    {
+      "seconds": 1.2597373989992775,
+      "bytes": 7034037,
+      "pages": 38,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790703713325746110-adad6d18/train.jsonl

@@ -63,3 +63,4 @@
 {"step": 286500, "loss": 0.6280013173818588, "lr": 0.0003, "grad_norm": 0.13263722039132297, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 41935690374, "tokens_per_second": 150288.80881129822, "data_wait_s": 1.2853999578510411e-05, "gpu_peak_gb": 85.46991729736328}
 {"step": 286600, "loss": 0.5946786962449551, "lr": 0.0003, "grad_norm": 0.17412314477500343, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 41950326801, "tokens_per_second": 149826.42959070986, "data_wait_s": 1.4883000403642654e-05, "gpu_peak_gb": 85.46991729736328}
 {"step": 286700, "loss": 0.5874792542308569, "lr": 0.0003, "grad_norm": 0.15030469061671303, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 41964796686, "tokens_per_second": 148058.06289482143, "data_wait_s": 1.3179000234231353e-05, "gpu_peak_gb": 85.46991729736328}
+{"step": 286800, "loss": 0.5679254066199064, "lr": 0.0003, "grad_norm": 0.15646814916094784, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 41979241507, "tokens_per_second": 147997.23661188598, "data_wait_s": 1.4410999938263558e-05, "gpu_peak_gb": 85.46991729736328}

runs/void-byte-v1/events.out.tfevents.1790703713.3e6d470a-8cc7-4d7e-a224-51a0f9a2ce55-85btc.38629.0.1790703713325746110-adad6d18

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:2e67baa153b053f4258665d2ae488ca1e170140546860c48c5aad13c794c6bc4
-size 138618
+oid sha256:9bafe676a06972d466055858d053003bcf8d04f4837dfca20008868f3c5e4b7a
+size 140736