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": 76978.54197178235,
-      "required_pages_per_second_upper_estimate": 0.0053255541484335985
+      "required_token_positions_per_second": 77196.84089797227,
+      "required_pages_per_second_upper_estimate": 0.005340656574670711
     },
     {
       "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": 9622.317746472794,
-      "required_pages_per_second_upper_estimate": 0.004028663503604322
+      "required_token_positions_per_second": 9649.605112246534,
+      "required_pages_per_second_upper_estimate": 0.004040088153828793
     },
     {
       "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": 19244.63549294559,
-      "required_pages_per_second_upper_estimate": 0.0040081500580137
+      "required_token_positions_per_second": 19299.21022449307,
+      "required_pages_per_second_upper_estimate": 0.004019516535362585
     },
     {
       "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": 28866.953239418384,
-      "required_pages_per_second_upper_estimate": 0.0017305297694176932
+      "required_token_positions_per_second": 28948.815336739604,
+      "required_pages_per_second_upper_estimate": 0.0017354372771559152
     },
     {
       "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": 28866.953239418384,
-      "required_pages_per_second_upper_estimate": 0.0022442786420563262
+      "required_token_positions_per_second": 28948.815336739604,
+      "required_pages_per_second_upper_estimate": 0.0022506430600497387
     },
     {
       "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": 28866.953239418384,
-      "required_pages_per_second_upper_estimate": 0.37027428123572537
+      "required_token_positions_per_second": 28948.815336739604,
+      "required_pages_per_second_upper_estimate": 0.37132432032349
     }
   ],
   "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": 0.10807373273441101,
-    "producer_p95_seconds": 0.17119580500002485,
-    "gpu_update_seconds": 0.957773401671158,
-    "producer_batches_per_second": 9.25294217844293,
-    "gpu_batches_per_second": 1.0440882971433154,
-    "producer_headroom": 8.86222190571382,
+    "producer_batch_seconds": 0.3312597744219019,
+    "producer_p95_seconds": 0.16357444999994186,
+    "gpu_update_seconds": 0.9550649889604046,
+    "producer_batches_per_second": 3.018778847341638,
+    "gpu_batches_per_second": 1.0470491658253618,
+    "producer_headroom": 2.8831299865102444,
     "sustainable": true,
-    "buffer_seconds": 60.33972430528296,
+    "buffer_seconds": 60.16909430450549,
     "buffer_drain_seconds": null,
     "shard_target_mib": 32,
-    "upload_pages_per_commit": 173,
+    "upload_pages_per_commit": 157,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 635813.5617594672,
-    "upload_bytes_per_second": 5238957.15017186,
+    "generation_bytes_per_second": 567187.552687345,
+    "upload_bytes_per_second": 5356088.063990163,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 68.21539468568808,
+    "estimated_full_shards_per_hour": 60.85262267811423,
     "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": [
@@ -394,6 +394,12 @@
       "bytes": 2451346,
       "pages": 32,
       "shards": 2
+    },
+    {
+      "seconds": 1.0700342689997342,
+      "bytes": 10599218,
+      "pages": 33,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791039222312944374-63c10d34/train.jsonl

@@ -41,3 +41,4 @@
 {"step": 608500, "loss": 0.5443984866142273, "lr": 0.0001, "grad_norm": 0.1289834412303883, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 89066479770, "tokens_per_second": 150885.8250828299, "data_wait_s": 1.8201999409939162e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 608600, "loss": 0.5503295063972473, "lr": 0.0001, "grad_norm": 0.12712442751930556, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 89081010922, "tokens_per_second": 151129.97094509134, "data_wait_s": 1.75280001712963e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 608700, "loss": 0.5426244139671326, "lr": 0.0001, "grad_norm": 0.14481412817115388, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 89095599368, "tokens_per_second": 151382.33120894, "data_wait_s": 1.3783999747829512e-05, "gpu_peak_gb": 89.37861251831055}
+{"step": 608800, "loss": 0.5435750484466553, "lr": 0.0001, "grad_norm": 0.15740634315374563, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 89110419345, "tokens_per_second": 153860.36729184573, "data_wait_s": 1.3574000149674248e-05, "gpu_peak_gb": 89.37861251831055}

runs/void-byte-v1/events.out.tfevents.1791039222.f2a393d4-e1e4-407b-a6aa-034aa2ada5ac-vwfgg.353.0.1791039222312944374-63c10d34

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:3205facd6fd767377153c0a6b4a0a1b4800fb24892a667563a6aac339f3a0d7f
-size 92022
+oid sha256:0e4dd27f2e58c45f80403cef8293d2e3f140fe767e0fa94f218d8e7b8ba5cf08
+size 94140