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": 76358.18746268797,
-      "required_pages_per_second_upper_estimate": 0.005282636584073674
+      "required_token_positions_per_second": 77188.76004772533,
+      "required_pages_per_second_upper_estimate": 0.005340097522700472
     },
     {
       "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": 9544.773432835997,
-      "required_pages_per_second_upper_estimate": 0.0039961973187940016
+      "required_token_positions_per_second": 9648.595005965666,
+      "required_pages_per_second_upper_estimate": 0.004039665243422411
     },
     {
       "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": 19089.546865671993,
-      "required_pages_per_second_upper_estimate": 0.003975849186914749
+      "required_token_positions_per_second": 19297.190011931332,
+      "required_pages_per_second_upper_estimate": 0.004019095778362564
     },
     {
       "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": 28634.32029850799,
-      "required_pages_per_second_upper_estimate": 0.0017165837798200486
+      "required_token_positions_per_second": 28945.785017897,
+      "required_pages_per_second_upper_estimate": 0.0017352556141683301
     },
     {
       "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": 28634.32029850799,
-      "required_pages_per_second_upper_estimate": 0.0022261924541447107
+      "required_token_positions_per_second": 28945.785017897,
+      "required_pages_per_second_upper_estimate": 0.0022504074660886905
     },
     {
       "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": 28634.32029850799,
-      "required_pages_per_second_upper_estimate": 0.367290315651518
+      "required_token_positions_per_second": 28945.785017897,
+      "required_pages_per_second_upper_estimate": 0.37128545064707996
     }
   ],
   "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": 0.26629807631229596,
-    "producer_p95_seconds": 0.3158541740049259,
-    "gpu_update_seconds": 0.9655546111021401,
-    "producer_batches_per_second": 3.755190476206329,
-    "gpu_batches_per_second": 1.0356742006115447,
-    "producer_headroom": 3.625841479867862,
+    "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,
     "sustainable": true,
-    "buffer_seconds": 60.82994049943483,
+    "buffer_seconds": 60.175393374995394,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 79,
+    "upload_pages_per_commit": 78,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 363398.81341471936,
-    "upload_bytes_per_second": 5645239.086805544,
+    "generation_bytes_per_second": 365812.13172245724,
+    "upload_bytes_per_second": 5723326.406535505,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 77.9769258673781,
+    "estimated_full_shards_per_hour": 78.49476779704368,
     "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": 1.2610901230000309,
-      "bytes": 5822650,
-      "pages": 29,
-      "shards": 2
-    },
     {
       "seconds": 0.9832762490004825,
       "bytes": 2733834,
@@ -562,6 +556,12 @@
       "bytes": 11847980,
       "pages": 36,
       "shards": 3
+    },
+    {
+      "seconds": 1.301661803998286,
+      "bytes": 11947886,
+      "pages": 38,
+      "shards": 4
     }
   ],
   "cache_key": {

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

@@ -379,3 +379,4 @@
 {"step": 722900, "loss": 0.5568122863769531, "lr": 8e-05, "grad_norm": 0.21909042152464192, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 105812839171, "tokens_per_second": 153428.21704980786, "data_wait_s": 1.3438999303616583e-05, "gpu_peak_gb": 89.37899398803711}
 {"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}

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 sha256:8a84707ae3b3f70c1ad77e5ad598237363b2489a93077bbfd3674daed902ff39
-size 812604
+oid sha256:bac9b2d1f601e93be0674f9cb62aa34fec42befc79fbf990259181168f4f011d
+size 814722