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": 75586.73113578392,
-      "required_pages_per_second_upper_estimate": 0.0065516662820311305
+      "required_token_positions_per_second": 75805.32002225549,
+      "required_pages_per_second_upper_estimate": 0.006570613012702018
     },
     {
       "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": 9448.34139197299,
-      "required_pages_per_second_upper_estimate": 0.0037690846465505785
+      "required_token_positions_per_second": 9475.665002781936,
+      "required_pages_per_second_upper_estimate": 0.003779984443426654
     },
     {
       "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": 18896.68278394598,
-      "required_pages_per_second_upper_estimate": 0.00437141499180759
+      "required_token_positions_per_second": 18951.330005563872,
+      "required_pages_per_second_upper_estimate": 0.004384056664770637
     },
     {
       "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": 28345.024175918974,
-      "required_pages_per_second_upper_estimate": 0.0017260632574649035
+      "required_token_positions_per_second": 28426.995008345806,
+      "required_pages_per_second_upper_estimate": 0.0017310548510919774
     },
     {
       "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": 28345.024175918974,
-      "required_pages_per_second_upper_estimate": 0.002182078425111084
+      "required_token_positions_per_second": 28426.995008345806,
+      "required_pages_per_second_upper_estimate": 0.0021883887667010888
     },
     {
       "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": 28345.024175918974,
-      "required_pages_per_second_upper_estimate": 0.3519503355714637
+      "required_token_positions_per_second": 28426.995008345806,
+      "required_pages_per_second_upper_estimate": 0.35296813897147655
     }
   ],
   "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": 688946,
-    "producer_batch_seconds": 0.07033997385152446,
-    "producer_p95_seconds": 0.17801131200030795,
-    "gpu_update_seconds": 0.9754092932998398,
-    "producer_batches_per_second": 14.216667212740616,
-    "gpu_batches_per_second": 1.0252106545109583,
-    "producer_headroom": 13.867069319058327,
+    "producer_batch_seconds": 0.08609307993747706,
+    "producer_p95_seconds": 0.17751800499991077,
+    "gpu_update_seconds": 0.9725966459656709,
+    "producer_batches_per_second": 11.61533541053735,
+    "gpu_batches_per_second": 1.0281754560310261,
+    "producer_headroom": 11.297036262054915,
     "sustainable": true,
-    "buffer_seconds": 60.47537618459007,
+    "buffer_seconds": 60.300992049871596,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 80,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 419371.58761128766,
-    "upload_bytes_per_second": 4058615.1326010227,
+    "generation_bytes_per_second": 414850.2712968458,
+    "upload_bytes_per_second": 4215779.356272049,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 89.98738023046467,
+    "estimated_full_shards_per_hour": 89.01721100024253,
     "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": [
@@ -226,6 +226,12 @@
       "bytes": 5274454,
       "pages": 34,
       "shards": 2
+    },
+    {
+      "seconds": 1.0228857370002515,
+      "bytes": 6079721,
+      "pages": 34,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790563279934888633-82973089/train.jsonl

@@ -38,3 +38,4 @@
 {"step": 158800, "loss": 0.6244642548263073, "lr": 0.0003, "grad_norm": 0.14388045756002935, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 23241394389, "tokens_per_second": 149268.78273229898, "data_wait_s": 1.7186999684781767e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 158900, "loss": 0.5830402486026287, "lr": 0.0003, "grad_norm": 0.1417218542578867, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 23256080379, "tokens_per_second": 148964.90358281686, "data_wait_s": 1.2634000086109154e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 159000, "loss": 0.5832870006561279, "lr": 0.0003, "grad_norm": 0.14141191893168306, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 23270696232, "tokens_per_second": 149508.56279893636, "data_wait_s": 1.2864999916928355e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 159100, "loss": 0.6049460880458355, "lr": 0.0003, "grad_norm": 0.13986737619423734, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 23285302788, "tokens_per_second": 147677.8955632395, "data_wait_s": 1.3328000022738706e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790563279.bdfa290c-aca9-4e96-a5d2-8425921a3278-d85hq.1304.0.1790563279934888633-82973089

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:0091fdf0ecdd6a32077efd15af602d7d38425a1c444f9f9bdd653bf59862b550
-size 84885
+oid sha256:74cd9766697c758536f4d336a5b5d35ebcede91e131a2152f92f903a836fef32
+size 87003