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": 76581.95511166255,
-      "required_pages_per_second_upper_estimate": 0.005298117349242215
+      "required_token_positions_per_second": 77286.3711192274,
+      "required_pages_per_second_upper_estimate": 0.005346850483115876
     },
     {
       "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": 9572.744388957819,
-      "required_pages_per_second_upper_estimate": 0.004007908174022225
+      "required_token_positions_per_second": 9660.796389903426,
+      "required_pages_per_second_upper_estimate": 0.004044773708083281
     },
     {
       "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": 19145.488777915638,
-      "required_pages_per_second_upper_estimate": 0.003987500411947666
+      "required_token_positions_per_second": 19321.59277980685,
+      "required_pages_per_second_upper_estimate": 0.004024178231366769
     },
     {
       "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": 28718.233166873455,
-      "required_pages_per_second_upper_estimate": 0.001721614228151027
+      "required_token_positions_per_second": 28982.389169710274,
+      "required_pages_per_second_upper_estimate": 0.0017374499771782323
     },
     {
       "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": 28718.233166873455,
-      "required_pages_per_second_upper_estimate": 0.0022327163105664246
+      "required_token_positions_per_second": 28982.389169710274,
+      "required_pages_per_second_upper_estimate": 0.002253253277887524
     },
     {
       "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": 28718.233166873455,
-      "required_pages_per_second_upper_estimate": 0.36836665982829175
+      "required_token_positions_per_second": 28982.389169710274,
+      "required_pages_per_second_upper_estimate": 0.3717549694040645
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.3350609138358749,
+    "producer_batch_seconds": 0.106444388218776,
     "producer_p95_seconds": 0.20390691600005084,
-    "gpu_update_seconds": 0.9627333213483352,
-    "producer_batches_per_second": 2.984531942421182,
-    "gpu_batches_per_second": 1.0387092435935132,
-    "producer_headroom": 2.873308349597343,
+    "gpu_update_seconds": 0.953958620806015,
+    "producer_batches_per_second": 9.394576987418933,
+    "gpu_batches_per_second": 1.0482634971683404,
+    "producer_headroom": 8.962037705974092,
     "sustainable": true,
-    "buffer_seconds": 60.652199244945116,
+    "buffer_seconds": 60.09939311077895,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 376154.7163945556,
-    "upload_bytes_per_second": 3722486.2081569503,
+    "generation_bytes_per_second": 375549.2298314112,
+    "upload_bytes_per_second": 3808720.5720877824,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 80.71404570462704,
+    "estimated_full_shards_per_hour": 80.58412238318208,
     "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": [
@@ -448,6 +448,12 @@
       "bytes": 6395327,
       "pages": 37,
       "shards": 2
+    },
+    {
+      "seconds": 1.1455854699997872,
+      "bytes": 11238485,
+      "pages": 32,
+      "shards": 4
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791143623306220418-eb05a537/train.jsonl

@@ -97,3 +97,4 @@
 {"step": 694700, "loss": 0.5409365892410278, "lr": 8e-05, "grad_norm": 0.21010084934710088, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 101685371843, "tokens_per_second": 154221.52109970583, "data_wait_s": 1.30280004668748e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 694800, "loss": 0.534839928150177, "lr": 8e-05, "grad_norm": 0.19867366195370206, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 101700147190, "tokens_per_second": 154719.53195764706, "data_wait_s": 1.2542999684228562e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 694900, "loss": 0.5418973565101624, "lr": 8e-05, "grad_norm": 0.2038722056580159, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 101714792347, "tokens_per_second": 152167.68312968055, "data_wait_s": 1.5100999007700011e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 695000, "loss": 0.546134889125824, "lr": 8e-05, "grad_norm": 0.21493580285015115, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 101729356649, "tokens_per_second": 152240.4627242544, "data_wait_s": 1.3352000678423792e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791143623.3eb6ddb3-ffce-42e5-bd48-c0d653189e98-8pb7d.369.0.1791143623306220418-eb05a537

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:78435e200192360d0761f68a52b9d85079e3ae5e6364b2cb65f3efb2173bdbe2
-size 210630
+oid sha256:8616b6716875763f33e54b386ac09b54ab836931bd6274da5d7d393958e49d31
+size 212748