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": 76961.01291187802,
-      "required_pages_per_second_upper_estimate": 0.005324341447396394
+      "required_token_positions_per_second": 76599.27678474896,
+      "required_pages_per_second_upper_estimate": 0.005299315702778151
     },
     {
       "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": 9620.126613984752,
-      "required_pages_per_second_upper_estimate": 0.0040277461221876285
+      "required_token_positions_per_second": 9574.90959809362,
+      "required_pages_per_second_upper_estimate": 0.004008814701872677
     },
     {
       "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": 19240.253227969504,
-      "required_pages_per_second_upper_estimate": 0.004007237347787281
+      "required_token_positions_per_second": 19149.81919618724,
+      "required_pages_per_second_upper_estimate": 0.003988402323872832
     },
     {
       "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": 28860.379841954258,
-      "required_pages_per_second_upper_estimate": 0.0017301357042767156
+      "required_token_positions_per_second": 28724.728794280858,
+      "required_pages_per_second_upper_estimate": 0.001722003631095852
     },
     {
       "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": 28860.379841954258,
-      "required_pages_per_second_upper_estimate": 0.0022437675893168104
+      "required_token_positions_per_second": 28724.728794280858,
+      "required_pages_per_second_upper_estimate": 0.0022332213170261047
     },
     {
       "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": 28860.379841954258,
-      "required_pages_per_second_upper_estimate": 0.37018996475101984
+      "required_token_positions_per_second": 28724.728794280858,
+      "required_pages_per_second_upper_estimate": 0.36844997876221264
     }
   ],
   "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": 0.06026643610157123,
-    "producer_p95_seconds": 0.15971326699946076,
-    "gpu_update_seconds": 0.9579915493630536,
-    "producer_batches_per_second": 16.592983834561416,
-    "gpu_batches_per_second": 1.043850544052165,
-    "producer_headroom": 15.895938292227594,
+    "producer_batch_seconds": 0.04766273130464782,
+    "producer_p95_seconds": 0.15447732499978883,
+    "gpu_update_seconds": 0.9625156149604713,
+    "producer_batches_per_second": 20.980753150889722,
+    "gpu_batches_per_second": 1.0389441838209224,
+    "producer_headroom": 20.194302521362467,
     "sustainable": true,
-    "buffer_seconds": 60.35346760987238,
+    "buffer_seconds": 60.63848374250969,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 82,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 0,
-    "upload_bytes_per_second": null,
-    "upload_sustainable": null,
+    "generation_bytes_per_second": 341485.8116202284,
+    "upload_bytes_per_second": 30047.894409906512,
+    "upload_sustainable": false,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 0.0,
+    "estimated_full_shards_per_hour": 73.27490579085483,
     "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": 1.6723634379995929,
+      "bytes": 50251,
+      "pages": 1,
+      "shards": 1
+    }
+  ],
   "cache_key": {
     "version": 1,
     "kind": "data",

metrics/void-byte-v1/1791300814123050567-72be911d/train.jsonl

@@ -39,3 +39,4 @@
 {"step": 803900, "loss": 0.5207116007804871, "lr": 3e-05, "grad_norm": 0.2721884989782707, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 117667628222, "tokens_per_second": 153257.60957487288, "data_wait_s": 1.504299962107325e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 804000, "loss": 0.5551673769950867, "lr": 3e-05, "grad_norm": 0.26991183216104825, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 117682351354, "tokens_per_second": 153545.42914045625, "data_wait_s": 1.5083000107551925e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 804100, "loss": 0.584531843662262, "lr": 3e-05, "grad_norm": 0.27472927765275135, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 117697095316, "tokens_per_second": 152842.4900825801, "data_wait_s": 1.3640999895869754e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 804200, "loss": 0.5252413153648376, "lr": 3e-05, "grad_norm": 0.26306703657193725, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 117711729713, "tokens_per_second": 150651.58017005722, "data_wait_s": 1.5761999748065136e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791300814.0c9d201b-2dbc-40fb-b739-4ee2350a0f54-7b2tx.840.0.1791300814123050567-72be911d

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:a3478bd46a196a7fb119aa2b231c35a1356450c4c02d5ebc53f95923a12b3eb3
-size 87003
+oid sha256:4cd44925f6309624248935a00b863758598ccf84fe47526cf50d67bd7a6569da
+size 89121