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": 74385.04416725476,
-      "required_pages_per_second_upper_estimate": 0.0064475071012468
+      "required_token_positions_per_second": 74735.34281134255,
+      "required_pages_per_second_upper_estimate": 0.006477870099892533
     },
     {
       "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": 9298.130520906845,
-      "required_pages_per_second_upper_estimate": 0.0037091632842296334
+      "required_token_positions_per_second": 9341.917851417818,
+      "required_pages_per_second_upper_estimate": 0.003726630705049393
     },
     {
       "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": 18596.26104181369,
-      "required_pages_per_second_upper_estimate": 0.004301917709007365
+      "required_token_positions_per_second": 18683.835702835637,
+      "required_pages_per_second_upper_estimate": 0.0043221765655733985
     },
     {
       "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": 27894.391562720535,
-      "required_pages_per_second_upper_estimate": 0.0016986220956077185
+      "required_token_positions_per_second": 28025.753554253453,
+      "required_pages_per_second_upper_estimate": 0.0017066213516889678
     },
     {
       "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": 27894.391562720535,
-      "required_pages_per_second_upper_estimate": 0.0021473874791161558
+      "required_token_positions_per_second": 28025.753554253453,
+      "required_pages_per_second_upper_estimate": 0.002157500088857626
     },
     {
       "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": 21,
-      "required_token_positions_per_second": 27894.391562720535,
-      "required_pages_per_second_upper_estimate": 0.3463549866858777
+      "required_token_positions_per_second": 28025.753554253453,
+      "required_pages_per_second_upper_estimate": 0.3479860629804121
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 61,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.12566787948429692,
+    "producer_batch_seconds": 0.10762072320289917,
     "producer_p95_seconds": 0.3616767479979899,
-    "gpu_update_seconds": 0.991166985586815,
-    "producer_batches_per_second": 7.957482883483818,
-    "gpu_batches_per_second": 1.0089117318692322,
-    "producer_headroom": 7.887194322481333,
+    "gpu_update_seconds": 0.9865211990278091,
+    "producer_batches_per_second": 9.291890727352603,
+    "gpu_batches_per_second": 1.013662961308357,
+    "producer_headroom": 9.166647181583272,
     "sustainable": true,
-    "buffer_seconds": 60.461186120795716,
+    "buffer_seconds": 60.17779314069635,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 87,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 458147.7054689269,
-    "upload_bytes_per_second": 5997698.200850242,
+    "generation_bytes_per_second": 455559.01966431027,
+    "upload_bytes_per_second": 5979613.789024964,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 98.30783246088845,
+    "estimated_full_shards_per_hour": 97.75236074873906,
     "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": 0.9397628990009252,
-      "bytes": 7665294,
-      "pages": 38,
-      "shards": 3
-    },
     {
       "seconds": 1.2407093820002046,
       "bytes": 5771889,
@@ -562,6 +556,12 @@
       "bytes": 5944681,
       "pages": 28,
       "shards": 2
+    },
+    {
+      "seconds": 1.1792278270004317,
+      "bytes": 7823958,
+      "pages": 39,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790485060735632541-7579de36/train.jsonl

@@ -301,3 +301,4 @@
 {"step": 115100, "loss": 0.6090226266533136, "lr": 0.0003, "grad_norm": 0.12995592805320222, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 16843057721, "tokens_per_second": 133816.77263533513, "data_wait_s": 1.7527003365103155e-05, "gpu_peak_gb": 85.46991729736328}
 {"step": 115200, "loss": 0.5843383781611919, "lr": 0.0003, "grad_norm": 0.13943145174520966, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 16857707507, "tokens_per_second": 149062.55401111444, "data_wait_s": 1.1841999366879463e-05, "gpu_peak_gb": 85.46991729736328}
 {"step": 115300, "loss": 0.6172855384647846, "lr": 0.0003, "grad_norm": 0.12460532216136234, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 16872235692, "tokens_per_second": 148134.06387533378, "data_wait_s": 1.2730000889860094e-05, "gpu_peak_gb": 85.46991729736328}
+{"step": 115400, "loss": 0.6104896310716867, "lr": 0.0003, "grad_norm": 0.1571173659230503, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 16886971989, "tokens_per_second": 149986.69570470662, "data_wait_s": 1.5436999092344195e-05, "gpu_peak_gb": 85.46991729736328}

runs/void-byte-v1/events.out.tfevents.1790485060.0b09bbb2-f535-42d2-bd00-cbae7fb94f51-r2qrq.28895.0.1790485060735632541-7579de36

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:c6b258d98ae18acf24f6ff2f64e7b51d83e3a18bd0ab118f50cb8671d8d7c982
-size 646617
+oid sha256:4234fac0eb8d3ddf109791e438f460c955c5aae60a74c5064c2c0783b3a10089
+size 648735