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": 76358.10710230222,
-      "required_pages_per_second_upper_estimate": 0.005282631024555727
+      "required_token_positions_per_second": 77369.52953526781,
+      "required_pages_per_second_upper_estimate": 0.005352603575291659
     },
     {
       "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": 9544.763387787778,
-      "required_pages_per_second_upper_estimate": 0.003996193113142077
+      "required_token_positions_per_second": 9671.191191908476,
+      "required_pages_per_second_upper_estimate": 0.004049125794614646
     },
     {
       "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": 19089.526775575556,
-      "required_pages_per_second_upper_estimate": 0.003975845002677473
+      "required_token_positions_per_second": 19342.382383816952,
+      "required_pages_per_second_upper_estimate": 0.004028508157623346
     },
     {
       "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": 28634.29016336333,
-      "required_pages_per_second_upper_estimate": 0.0017165819732641399
+      "required_token_positions_per_second": 29013.573575725426,
+      "required_pages_per_second_upper_estimate": 0.0017393194347030097
     },
     {
       "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": 28634.29016336333,
-      "required_pages_per_second_upper_estimate": 0.0022261901112697637
+      "required_token_positions_per_second": 29013.573575725426,
+      "required_pages_per_second_upper_estimate": 0.002255677728289496
     },
     {
       "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": 28634.29016336333,
-      "required_pages_per_second_upper_estimate": 0.36728992911023883
+      "required_token_positions_per_second": 29013.573575725426,
+      "required_pages_per_second_upper_estimate": 0.37215496948122045
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 704566,
-    "producer_batch_seconds": 0.3415050684219523,
-    "producer_p95_seconds": 0.18518035499982943,
-    "gpu_update_seconds": 0.9655556272658975,
-    "producer_batches_per_second": 2.9282142271588,
-    "gpu_batches_per_second": 1.0356731106540558,
-    "producer_headroom": 2.82735372487324,
+    "producer_batch_seconds": 0.10570900743755374,
+    "producer_p95_seconds": 0.18889509799919324,
+    "gpu_update_seconds": 0.9529332857891056,
+    "producer_batches_per_second": 9.459931790493231,
+    "gpu_batches_per_second": 1.0493914053720133,
+    "producer_headroom": 9.014683884455533,
     "sustainable": true,
-    "buffer_seconds": 60.83000451775154,
+    "buffer_seconds": 60.03479700471365,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 82,
+    "upload_pages_per_commit": 81,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 395619.1308485946,
-    "upload_bytes_per_second": 5659318.024096263,
+    "generation_bytes_per_second": 394392.65071361064,
+    "upload_bytes_per_second": 5713211.929342723,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 84.89065593808535,
+    "estimated_full_shards_per_hour": 84.6274818521141,
     "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": 1.0666157699999985,
-      "bytes": 6099291,
-      "pages": 35,
-      "shards": 2
-    },
     {
       "seconds": 1.0668149449998054,
       "bytes": 5027383,
@@ -562,6 +556,12 @@
       "bytes": 5628518,
       "pages": 28,
       "shards": 2
+    },
+    {
+      "seconds": 1.4070167600002605,
+      "bytes": 12110556,
+      "pages": 40,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790989041640695387-36510d53/train.jsonl

@@ -87,3 +87,4 @@
 {"step": 565800, "loss": 0.5782396197319031, "lr": 0.0003, "grad_norm": 0.16403070827311558, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 82815974215, "tokens_per_second": 153332.08820165443, "data_wait_s": 1.2675000107265078e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 565900, "loss": 0.5894935131072998, "lr": 0.0003, "grad_norm": 0.1655009827884529, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 82830589934, "tokens_per_second": 152541.5745841334, "data_wait_s": 1.536899981147144e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 566000, "loss": 0.5413221716880798, "lr": 0.0003, "grad_norm": 0.1595919349606399, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 82845245541, "tokens_per_second": 151979.03032959043, "data_wait_s": 1.5294999684556387e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 566100, "loss": 0.5647397041320801, "lr": 0.0003, "grad_norm": 0.15198237748100768, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 82859990226, "tokens_per_second": 154040.11309169207, "data_wait_s": 1.4417999409488402e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1790989041.fb9d4967-10f3-4fa9-9622-1d5a28f8589b-vh8db.427.0.1790989041640695387-36510d53

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:fe9f481213b6eb0e84a349f8913f1d323401b27e7c00b2472c6d168e1013eda8
-size 190233
+oid sha256:692eacdf386bd2b28c74001984826504608efc2b21c9def588f4089c241f33dd
+size 192351