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": 77521.91031653603,
-      "required_pages_per_second_upper_estimate": 0.00536314563131192
+      "required_token_positions_per_second": 77647.0132695985,
+      "required_pages_per_second_upper_estimate": 0.005371800543883625
     },
     {
       "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": 9690.238789567004,
-      "required_pages_per_second_upper_estimate": 0.004057100625995202
+      "required_token_positions_per_second": 9705.876658699812,
+      "required_pages_per_second_upper_estimate": 0.004063647875245267
     },
     {
       "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": 19380.477579134007,
-      "required_pages_per_second_upper_estimate": 0.0040364423821700295
+      "required_token_positions_per_second": 19411.753317399623,
+      "required_pages_per_second_upper_estimate": 0.0040429562936540745
     },
     {
       "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": 29070.716368701014,
-      "required_pages_per_second_upper_estimate": 0.0017427450578899033
+      "required_token_positions_per_second": 29117.629976099437,
+      "required_pages_per_second_upper_estimate": 0.001745557457020508
     },
     {
       "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": 29070.716368701014,
-      "required_pages_per_second_upper_estimate": 0.0022601203290987647
+      "required_token_positions_per_second": 29117.629976099437,
+      "required_pages_per_second_upper_estimate": 0.002263767655722841
     },
     {
       "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": 29070.716368701014,
-      "required_pages_per_second_upper_estimate": 0.3728879358743604
+      "required_token_positions_per_second": 29117.629976099437,
+      "required_pages_per_second_upper_estimate": 0.37348969325816034
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 64,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 704566,
-    "producer_batch_seconds": 0.09989942612499192,
-    "producer_p95_seconds": 0.17660916199929488,
-    "gpu_update_seconds": 0.9510601544641404,
-    "producer_batches_per_second": 10.010067512788536,
-    "gpu_batches_per_second": 1.0514582019929475,
-    "producer_headroom": 9.52017635490914,
+    "producer_batch_seconds": 0.051475819140677004,
+    "producer_p95_seconds": 0.16822228800083394,
+    "gpu_update_seconds": 0.9495278297957028,
+    "producer_batches_per_second": 19.426597122565926,
+    "gpu_batches_per_second": 1.0531550193901706,
+    "producer_headroom": 18.446094606105472,
     "sustainable": true,
-    "buffer_seconds": 60.867849885704985,
+    "buffer_seconds": 60.76978110692498,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 79,
+    "upload_pages_per_commit": 86,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 373238.09022738825,
-    "upload_bytes_per_second": 5829879.354611598,
+    "generation_bytes_per_second": 391921.9848607955,
+    "upload_bytes_per_second": 5837199.948025468,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 80.08820562473522,
+    "estimated_full_shards_per_hour": 84.09733447425745,
     "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": [
@@ -436,6 +436,12 @@
       "bytes": 1202765,
       "pages": 34,
       "shards": 1
+    },
+    {
+      "seconds": 1.2456794589998026,
+      "bytes": 7629808,
+      "pages": 38,
+      "shards": 3
     }
   ],
   "cache_key": {

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

@@ -56,3 +56,4 @@
 {"step": 490600, "loss": 0.5570622086524963, "lr": 0.0003, "grad_norm": 0.17094227700346842, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 71807382888, "tokens_per_second": 153244.69705833748, "data_wait_s": 1.0742000085883774e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 490700, "loss": 0.5813391208648682, "lr": 0.0003, "grad_norm": 0.1400664954006451, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 71822127687, "tokens_per_second": 154452.1648729404, "data_wait_s": 1.6877998859854415e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 490800, "loss": 0.6053390502929688, "lr": 0.0003, "grad_norm": 0.15017771599203275, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 71836843887, "tokens_per_second": 153448.9073051394, "data_wait_s": 1.5513000107603148e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 490900, "loss": 0.6020042300224304, "lr": 0.0003, "grad_norm": 0.19061114434149673, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 71851442957, "tokens_per_second": 116923.8574456793, "data_wait_s": 1.2448999768821523e-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 sha256:928bd3b6e54f3e4a26fb0a4e90c71933b5aa6d5726fc1bc05b2cac396fbb11d6
-size 123792
+oid sha256:bfa40da8b7ab0d41669b7dc39c3371ffbba540ea910b00a756136d0f412b402d
+size 125910