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": 77158.17895768624,
-      "required_pages_per_second_upper_estimate": 0.005337981851933662
+      "required_token_positions_per_second": 76688.90675774695,
+      "required_pages_per_second_upper_estimate": 0.005305516512280343
     },
     {
       "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": 9644.77236971078,
-      "required_pages_per_second_upper_estimate": 0.004038064785448213
+      "required_token_positions_per_second": 9586.11334471837,
+      "required_pages_per_second_upper_estimate": 0.004013505476623625
     },
     {
       "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": 19289.54473942156,
-      "required_pages_per_second_upper_estimate": 0.004017503469718172
+      "required_token_positions_per_second": 19172.22668943674,
+      "required_pages_per_second_upper_estimate": 0.003993069213791367
     },
     {
       "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": 28934.31710913234,
-      "required_pages_per_second_upper_estimate": 0.0017345681305483713
+      "required_token_positions_per_second": 28758.34003415511,
+      "required_pages_per_second_upper_estimate": 0.0017240185736049217
     },
     {
       "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": 28934.31710913234,
-      "required_pages_per_second_upper_estimate": 0.0022495158866242377
+      "required_token_positions_per_second": 28758.34003415511,
+      "required_pages_per_second_upper_estimate": 0.0022358344430861076
     },
     {
       "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": 28934.31710913234,
-      "required_pages_per_second_upper_estimate": 0.37113835262672795
+      "required_token_positions_per_second": 28758.34003415511,
+      "required_pages_per_second_upper_estimate": 0.3688811076583819
     }
   ],
   "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": 720750,
-    "producer_batch_seconds": 0.06708354450000797,
-    "producer_p95_seconds": 0.23201359899758245,
-    "gpu_update_seconds": 0.9555435469833036,
-    "producer_batches_per_second": 14.906785374167,
-    "gpu_batches_per_second": 1.0465247796995203,
-    "producer_headroom": 14.244082570550368,
+    "producer_batch_seconds": 0.0906656844611291,
+    "producer_p95_seconds": 0.24542762299824972,
+    "gpu_update_seconds": 0.9613906771796841,
+    "producer_batches_per_second": 11.029531249264739,
+    "gpu_batches_per_second": 1.0401598681335036,
+    "producer_headroom": 10.603688516705116,
     "sustainable": true,
-    "buffer_seconds": 60.19924345994812,
+    "buffer_seconds": 60.567612662320094,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 78,
+    "upload_pages_per_commit": 84,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 300709.39414568956,
-    "upload_bytes_per_second": 5658032.292965459,
+    "generation_bytes_per_second": 301772.86725704273,
+    "upload_bytes_per_second": 5629672.508276998,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 64.52523582723632,
+    "estimated_full_shards_per_hour": 64.7534324005457,
     "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": 1.034200502999738,
-      "bytes": 7563521,
-      "pages": 38,
-      "shards": 3
-    },
     {
       "seconds": 1.312004507999518,
       "bytes": 5834009,
@@ -562,6 +556,12 @@
       "bytes": 7703540,
       "pages": 39,
       "shards": 3
+    },
+    {
+      "seconds": 1.1081734670005972,
+      "bytes": 5790944,
+      "pages": 28,
+      "shards": 2
     }
   ],
   "cache_key": {

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

@@ -222,3 +222,4 @@
 {"step": 707200, "loss": 0.5109959244728088, "lr": 8e-05, "grad_norm": 0.20483624901390615, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 103514080865, "tokens_per_second": 150895.4630753194, "data_wait_s": 1.7610000213608146e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 707300, "loss": 0.5474433898925781, "lr": 8e-05, "grad_norm": 0.21330609410834672, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 103528584357, "tokens_per_second": 151187.28304752088, "data_wait_s": 1.585899735800922e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 707400, "loss": 0.5188702344894409, "lr": 8e-05, "grad_norm": 0.19953598797003128, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 103543182303, "tokens_per_second": 151694.7663148193, "data_wait_s": 1.6600999515503645e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 707500, "loss": 0.5464622974395752, "lr": 8e-05, "grad_norm": 0.19706057669124938, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 103557788498, "tokens_per_second": 152070.82317731014, "data_wait_s": 1.650600097491406e-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 sha256:034ece51f705cff7fb1009cb05c89f1b9c319d5c6d4bd068197d08e0cbe80b8f
-size 477729
+oid sha256:dafd4079d38b09d03b5dbce01c8b54843bc3dd221d9ce20d84b338dcb41487b7
+size 479847