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": 75489.75400094781,
-      "required_pages_per_second_upper_estimate": 0.006543260549769838
+      "required_token_positions_per_second": 75336.58356468836,
+      "required_pages_per_second_upper_estimate": 0.0065299841245617905
     },
     {
       "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": 9436.219250118476,
-      "required_pages_per_second_upper_estimate": 0.003764248942922641
+      "required_token_positions_per_second": 9417.072945586046,
+      "required_pages_per_second_upper_estimate": 0.0037566111957818915
     },
     {
       "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": 18872.438500236953,
-      "required_pages_per_second_upper_estimate": 0.004365806503456327
+      "required_token_positions_per_second": 18834.14589117209,
+      "required_pages_per_second_upper_estimate": 0.004356948182276073
     },
     {
       "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": 28308.657750355425,
-      "required_pages_per_second_upper_estimate": 0.0017238487329479725
+      "required_token_positions_per_second": 28251.218836758137,
+      "required_pages_per_second_upper_estimate": 0.0017203510044685871
     },
     {
       "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": 28308.657750355425,
-      "required_pages_per_second_upper_estimate": 0.002179278837002493
+      "required_token_positions_per_second": 28251.218836758137,
+      "required_pages_per_second_upper_estimate": 0.0021748570304326847
     },
     {
       "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": 28308.657750355425,
-      "required_pages_per_second_upper_estimate": 0.3514987862765614
+      "required_token_positions_per_second": 28251.218836758137,
+      "required_pages_per_second_upper_estimate": 0.35078558720536074
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 717798,
-    "producer_batch_seconds": 0.12226132703921166,
+    "producer_batch_seconds": 0.09920663029711818,
     "producer_p95_seconds": 0.3029395539997495,
-    "gpu_update_seconds": 0.9766623428005119,
-    "producer_batches_per_second": 8.179201258622687,
-    "gpu_batches_per_second": 1.0238953179382027,
-    "producer_headroom": 7.988317863483329,
+    "gpu_update_seconds": 0.9786480420457726,
+    "producer_batches_per_second": 10.079971439459815,
+    "gpu_batches_per_second": 1.021817810935986,
+    "producer_headroom": 9.864744313104655,
     "sustainable": true,
-    "buffer_seconds": 60.553065253631736,
+    "buffer_seconds": 60.6761786068379,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 79,
+    "upload_pages_per_commit": 84,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 256402.10482119222,
-    "upload_bytes_per_second": 4995265.12301872,
+    "generation_bytes_per_second": 277106.75297833997,
+    "upload_bytes_per_second": 5003427.660391841,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 55.017922959106684,
+    "estimated_full_shards_per_hour": 59.460658474089136,
     "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": 1.079595678998885,
-      "bytes": 1118845,
-      "pages": 34,
-      "shards": 1
-    },
     {
       "seconds": 1.1773780209987308,
       "bytes": 12342839,
@@ -562,6 +556,12 @@
       "bytes": 184838,
       "pages": 5,
       "shards": 1
+    },
+    {
+      "seconds": 0.9666055019988562,
+      "bytes": 1234734,
+      "pages": 35,
+      "shards": 1
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790782192342215557-281f6532/train.jsonl

@@ -250,3 +250,4 @@
 {"step": 380000, "loss": 0.5973047744482756, "lr": 0.0003, "grad_norm": 0.14794959448736272, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 55620842514, "tokens_per_second": 149050.32637906037, "data_wait_s": 1.7037000361597165e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 380100, "loss": 0.6261854413896799, "lr": 0.0003, "grad_norm": 0.18086174655739745, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 55635464482, "tokens_per_second": 134259.735991459, "data_wait_s": 1.5268000424839556e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 380200, "loss": 0.5697666611522436, "lr": 0.0003, "grad_norm": 0.17639096696056597, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 55649946284, "tokens_per_second": 147124.3801264208, "data_wait_s": 1.1393000022508204e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 380300, "loss": 0.6201922632753849, "lr": 0.0003, "grad_norm": 0.15635784513159612, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 55664393099, "tokens_per_second": 147345.95866581547, "data_wait_s": 1.7039001249941066e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790782192.062ccfe6-664f-43fe-9e0e-62f43f0c4870-64tpq.429.0.1790782192342215557-281f6532

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:37ef50f82629b30aafa86f035df2c2c712f2a6b0d7761d4af035aa53a9acb853
-size 537816
+oid sha256:a9a1d550b8432a7ee20b7ab661aea910c45ef67aa0227ac75085a6eb502791f9
+size 539934