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": 77188.76004772533,
-      "required_pages_per_second_upper_estimate": 0.005340097522700472
+      "required_token_positions_per_second": 76885.94299457752,
+      "required_pages_per_second_upper_estimate": 0.005319147936331339
     },
     {
       "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": 9648.595005965666,
-      "required_pages_per_second_upper_estimate": 0.004039665243422411
+      "required_token_positions_per_second": 9610.74287432219,
+      "required_pages_per_second_upper_estimate": 0.00402381734634568
     },
     {
       "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": 19297.190011931332,
-      "required_pages_per_second_upper_estimate": 0.004019095778362564
+      "required_token_positions_per_second": 19221.48574864438,
+      "required_pages_per_second_upper_estimate": 0.0040033285767755705
     },
     {
       "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": 28945.785017897,
-      "required_pages_per_second_upper_estimate": 0.0017352556141683301
+      "required_token_positions_per_second": 28832.22862296657,
+      "required_pages_per_second_upper_estimate": 0.0017284480816828261
     },
     {
       "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": 28945.785017897,
-      "required_pages_per_second_upper_estimate": 0.0022504074660886905
+      "required_token_positions_per_second": 28832.22862296657,
+      "required_pages_per_second_upper_estimate": 0.0022415789558646437
     },
     {
       "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": 28945.785017897,
-      "required_pages_per_second_upper_estimate": 0.37128545064707996
+      "required_token_positions_per_second": 28832.22862296657,
+      "required_pages_per_second_upper_estimate": 0.36982887114026974
     }
   ],
   "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": 0.13307103409368892,
-    "producer_p95_seconds": 0.34238415399886435,
-    "gpu_update_seconds": 0.9551649742062761,
-    "producer_batches_per_second": 7.514783414818495,
-    "gpu_batches_per_second": 1.0469395622792605,
-    "producer_headroom": 7.177857906580859,
+    "producer_batch_seconds": 0.09133434380504468,
+    "producer_p95_seconds": 0.3226876680055284,
+    "gpu_update_seconds": 0.9589269134047008,
+    "producer_batches_per_second": 10.94878397697282,
+    "gpu_batches_per_second": 1.0428323431339181,
+    "producer_headroom": 10.499083624573391,
     "sustainable": true,
-    "buffer_seconds": 60.175393374995394,
+    "buffer_seconds": 60.41239554449615,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 78,
+    "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 365812.13172245724,
-    "upload_bytes_per_second": 5723326.406535505,
+    "generation_bytes_per_second": 400426.601573972,
+    "upload_bytes_per_second": 5676427.9287501685,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 78.49476779704368,
+    "estimated_full_shards_per_hour": 85.9222272435605,
     "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": 0.9832762490004825,
-      "bytes": 2733834,
-      "pages": 37,
-      "shards": 2
-    },
     {
       "seconds": 1.1357609220031009,
       "bytes": 11893185,
@@ -562,6 +556,12 @@
       "bytes": 11947886,
       "pages": 38,
       "shards": 4
+    },
+    {
+      "seconds": 1.3009620130033,
+      "bytes": 995955,
+      "pages": 28,
+      "shards": 1
     }
   ],
   "cache_key": {

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

@@ -380,3 +380,4 @@
 {"step": 723000, "loss": 0.5880641341209412, "lr": 8e-05, "grad_norm": 0.2674153980514474, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 105827552658, "tokens_per_second": 152964.67979861106, "data_wait_s": 1.0994001058861613e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 723100, "loss": 0.5331712961196899, "lr": 8e-05, "grad_norm": 0.23303184121558437, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 105842314080, "tokens_per_second": 154526.20324963384, "data_wait_s": 1.3616998330689967e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 723200, "loss": 0.5436961054801941, "lr": 8e-05, "grad_norm": 0.22530942146442387, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 105856894818, "tokens_per_second": 151856.9309731779, "data_wait_s": 1.3899996702093631e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 723300, "loss": 0.5442550778388977, "lr": 8e-05, "grad_norm": 0.21169117058640777, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 105871498737, "tokens_per_second": 152329.35673679033, "data_wait_s": 1.446500391466543e-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:bac9b2d1f601e93be0674f9cb62aa34fec42befc79fbf990259181168f4f011d
-size 814722
+oid sha256:843172b1d5460e6c98b30418e480ffe1b8582730052eb658303d8812bb03652c
+size 816840