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": 76925.77774497677,
-      "required_pages_per_second_upper_estimate": 0.005321903796793315
+      "required_token_positions_per_second": 77333.51749116281,
+      "required_pages_per_second_upper_estimate": 0.0053501121785210215
     },
     {
       "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": 9615.722218122097,
-      "required_pages_per_second_upper_estimate": 0.004025902093614179
+      "required_token_positions_per_second": 9666.689686395352,
+      "required_pages_per_second_upper_estimate": 0.004047241108258426
     },
     {
       "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": 19231.444436244194,
-      "required_pages_per_second_upper_estimate": 0.004005402708774359
+      "required_token_positions_per_second": 19333.379372790703,
+      "required_pages_per_second_upper_estimate": 0.004026633067851945
     },
     {
       "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": 28847.16665436629,
-      "required_pages_per_second_upper_estimate": 0.0017293435938561862
+      "required_token_positions_per_second": 29000.06905918606,
+      "required_pages_per_second_upper_estimate": 0.0017385098595566774
     },
     {
       "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": 28847.16665436629,
-      "required_pages_per_second_upper_estimate": 0.002242740322100516
+      "required_token_positions_per_second": 29000.06905918606,
+      "required_pages_per_second_upper_estimate": 0.002254627811528651
     },
     {
       "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": 28847.16665436629,
-      "required_pages_per_second_upper_estimate": 0.3700204801678569
+      "required_token_positions_per_second": 29000.06905918606,
+      "required_pages_per_second_upper_estimate": 0.37198174804307355
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,27 +158,34 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.05520183825780833,
-    "producer_p95_seconds": 0.133867533999819,
-    "gpu_update_seconds": 0.9584303488542163,
-    "producer_batches_per_second": 18.11533875610654,
-    "gpu_batches_per_second": 1.0433726365149845,
-    "producer_headroom": 17.362290443627497,
+    "producer_batch_seconds": 0.04284780589061654,
+    "producer_p95_seconds": 0.1318608189999395,
+    "gpu_update_seconds": 0.953377040019228,
+    "producer_batches_per_second": 23.338417900623355,
+    "gpu_batches_per_second": 1.0489029607633846,
+    "producer_headroom": 22.25031177682806,
     "sustainable": true,
-    "buffer_seconds": 60.38111197781563,
+    "buffer_seconds": 60.06275352121136,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 64,
+    "upload_pages_per_commit": 68,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 0,
-    "upload_bytes_per_second": null,
-    "upload_sustainable": null,
+    "generation_bytes_per_second": 329631.9128633024,
+    "upload_bytes_per_second": 17370.243762308244,
+    "upload_sustainable": false,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 0.0,
+    "estimated_full_shards_per_hour": 70.73133506225875,
     "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": [],
+  "upload_measurements": [
+    {
+      "seconds": 1.296470003999957,
+      "bytes": 22520,
+      "pages": 1,
+      "shards": 1
+    }
+  ],
   "cache_key": {
     "version": 1,
     "kind": "data",

metrics/void-byte-v1/1791033297667653750-ba4975eb/train.jsonl

@@ -10,3 +10,4 @@
 {"step": 601000, "loss": 0.6038609147071838, "lr": 0.0003, "grad_norm": 0.14563901031957593, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 87968610536, "tokens_per_second": 152654.9884336665, "data_wait_s": 1.4574000033462653e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 601100, "loss": 0.5413586497306824, "lr": 0.0003, "grad_norm": 0.15647117915843692, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 87983109662, "tokens_per_second": 151496.7594740611, "data_wait_s": 1.5061000112837064e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 601200, "loss": 0.5303405523300171, "lr": 0.0003, "grad_norm": 0.13621012365771204, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 87997780680, "tokens_per_second": 153185.3052024277, "data_wait_s": 1.4321999969979515e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 601300, "loss": 0.5702246427536011, "lr": 0.0003, "grad_norm": 0.16696146154166297, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 88012437486, "tokens_per_second": 151797.32403943306, "data_wait_s": 1.4603999943574308e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791033297.83ac8725-d818-41af-bbb1-78dcd88792cf-7zwhs.737.0.1791033297667653750-ba4975eb

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:72973141f14c809880bde2a2656e2312996a2290d2a2199e660e4864ff930a25
-size 25581
+oid sha256:4b9de4747f3730c448d36d8c76ad398a96422024ca2e723b156e84723a60ee78
+size 27699