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": 77408.12428617418,
-      "required_pages_per_second_upper_estimate": 0.005355273649711522
+      "required_token_positions_per_second": 77453.72648399828,
+      "required_pages_per_second_upper_estimate": 0.00535842851556867
     },
     {
       "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": 9676.015535771772,
-      "required_pages_per_second_upper_estimate": 0.0040511456466464525
+      "required_token_positions_per_second": 9681.715810499785,
+      "required_pages_per_second_upper_estimate": 0.004053532232639799
     },
     {
       "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": 19352.031071543544,
-      "required_pages_per_second_upper_estimate": 0.004030517724823789
+      "required_token_positions_per_second": 19363.43162099957,
+      "required_pages_per_second_upper_estimate": 0.004032892158622772
     },
     {
       "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": 29028.04660731532,
-      "required_pages_per_second_upper_estimate": 0.001740187070847784
+      "required_token_positions_per_second": 29045.147431499354,
+      "required_pages_per_second_upper_estimate": 0.0017412122391461703
     },
     {
       "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": 29028.04660731532,
-      "required_pages_per_second_upper_estimate": 0.002256802942835472
+      "required_token_positions_per_second": 29045.147431499354,
+      "required_pages_per_second_upper_estimate": 0.0022581324566972047
     },
     {
       "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": 29028.04660731532,
-      "required_pages_per_second_upper_estimate": 0.372340613990525
+      "required_token_positions_per_second": 29045.147431499354,
+      "required_pages_per_second_upper_estimate": 0.37255996500172334
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,36 +155,30 @@
   "shard_benchmark_scope": "Local packing/read measurements; upload capacity measured only on real data commits. Targets larger than available pages are sample-limited.",
   "estimated_fresh_generation_batch_seconds": 2.558590814491007,
   "selected": {
-    "prefetch_batches": 63,
+    "prefetch_batches": 64,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 704566,
-    "producer_batch_seconds": 0.40975772554685364,
-    "producer_p95_seconds": 0.31547553400014294,
-    "gpu_update_seconds": 0.9524581648229979,
-    "producer_batches_per_second": 2.4404664943544923,
-    "gpu_batches_per_second": 1.0499148801835692,
-    "producer_headroom": 2.324442238524895,
+    "producer_batch_seconds": 0.11005713519494975,
+    "producer_p95_seconds": 0.3142559350017109,
+    "gpu_update_seconds": 0.9518973888910561,
+    "producer_batches_per_second": 9.086189625312795,
+    "gpu_batches_per_second": 1.0505333995767996,
+    "producer_headroom": 8.649120179304253,
     "sustainable": true,
-    "buffer_seconds": 60.00486438384887,
+    "buffer_seconds": 60.92143288902759,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 84,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 313528.7732393986,
-    "upload_bytes_per_second": 5602161.381612419,
+    "generation_bytes_per_second": 314172.3132429037,
+    "upload_bytes_per_second": 5685149.198085611,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 67.27597616087407,
+    "estimated_full_shards_per_hour": 67.4140648647817,
     "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": 0.9898015409999061,
-      "bytes": 867435,
-      "pages": 29,
-      "shards": 1
-    },
     {
       "seconds": 1.4056065479999234,
       "bytes": 17665783,
@@ -562,6 +556,12 @@
       "bytes": 2126873,
       "pages": 22,
       "shards": 2
+    },
+    {
+      "seconds": 1.4999150360017666,
+      "bytes": 10166456,
+      "pages": 43,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790914434605211575-090606d4/train.jsonl

@@ -362,3 +362,4 @@
 {"step": 521200, "loss": 0.6419379711151123, "lr": 0.0003, "grad_norm": 0.1925108938238185, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 76286146838, "tokens_per_second": 154367.94719962295, "data_wait_s": 9.009003406390548e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 521300, "loss": 0.6548756957054138, "lr": 0.0003, "grad_norm": 0.16964390841660704, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 76300837391, "tokens_per_second": 149669.21883760186, "data_wait_s": 1.5433004591614008e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 521400, "loss": 0.5528781414031982, "lr": 0.0003, "grad_norm": 0.14258201655071912, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 76315482106, "tokens_per_second": 152911.77912135163, "data_wait_s": 1.5258003259077668e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 521500, "loss": 0.57232266664505, "lr": 0.0003, "grad_norm": 0.16306147595004902, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 76330023305, "tokens_per_second": 151847.95394713708, "data_wait_s": 1.5318000805564225e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1790914434.4dec8a9d-3312-413a-9119-9058bec1adb7-bx5st.8214.0.1790914434605211575-090606d4

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:b2f1558bec44ca39cd8b46d8cc57ec908c20c230bbbddc585b1a9a715be0eb52
-size 776598
+oid sha256:d129a7c28ef083c7e859f4b3c389d6610f48a7aabe8a0a0c839a5dc73748cf0e
+size 778716