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": 76981.40347872314,
-      "required_pages_per_second_upper_estimate": 0.005325752114123373
+      "required_token_positions_per_second": 76485.21437128789,
+      "required_pages_per_second_upper_estimate": 0.0052914246003536525
     },
     {
       "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": 9622.675434840392,
-      "required_pages_per_second_upper_estimate": 0.004028813260254453
+      "required_token_positions_per_second": 9560.651796410986,
+      "required_pages_per_second_upper_estimate": 0.004002845258044913
     },
     {
       "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": 19245.350869680784,
-      "required_pages_per_second_upper_estimate": 0.004008299052121888
+      "required_token_positions_per_second": 19121.303592821972,
+      "required_pages_per_second_upper_estimate": 0.003982463275698877
     },
     {
       "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": 28868.026304521176,
-      "required_pages_per_second_upper_estimate": 0.00173059409803213
+      "required_token_positions_per_second": 28681.95538923296,
+      "required_pages_per_second_upper_estimate": 0.0017194394307744384
     },
     {
       "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": 28868.026304521176,
-      "required_pages_per_second_upper_estimate": 0.002244362068148154
+      "required_token_positions_per_second": 28681.95538923296,
+      "required_pages_per_second_upper_estimate": 0.002229895872923956
     },
     {
       "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": 28868.026304521176,
-      "required_pages_per_second_upper_estimate": 0.37028804536269644
+      "required_token_positions_per_second": 28681.95538923296,
+      "required_pages_per_second_upper_estimate": 0.36790132744876236
     }
   ],
   "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": 0.08202366232052327,
-    "producer_p95_seconds": 0.2177247530016757,
-    "gpu_update_seconds": 0.9577377998879646,
-    "producer_batches_per_second": 12.191603882454153,
-    "gpu_batches_per_second": 1.0441271088151467,
-    "producer_headroom": 11.676359879487208,
+    "producer_batch_seconds": 0.08642769467951439,
+    "producer_p95_seconds": 0.21377978899909067,
+    "gpu_update_seconds": 0.9639510146640455,
+    "producer_batches_per_second": 11.57036530602992,
+    "gpu_batches_per_second": 1.0373971133258448,
+    "producer_headroom": 11.153265376781212,
     "sustainable": true,
-    "buffer_seconds": 60.33748139294177,
+    "buffer_seconds": 60.72891392383487,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 505373.34597003605,
-    "upload_bytes_per_second": 5184122.314564994,
+    "generation_bytes_per_second": 505389.291658463,
+    "upload_bytes_per_second": 5114228.923640908,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 108.44135555578052,
+    "estimated_full_shards_per_hour": 108.44477712932031,
     "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": 1.4404141169998184,
-      "bytes": 12120623,
-      "pages": 39,
-      "shards": 3
-    },
     {
       "seconds": 0.8672846369991021,
       "bytes": 2691618,
@@ -562,6 +556,12 @@
       "bytes": 5395680,
       "pages": 34,
       "shards": 2
+    },
+    {
+      "seconds": 1.439521827000135,
+      "bytes": 6360408,
+      "pages": 36,
+      "shards": 2
     }
   ],
   "cache_key": {

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

@@ -165,3 +165,4 @@
 {"step": 620900, "loss": 0.5788757801055908, "lr": 0.0001, "grad_norm": 0.1803991948934275, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 90883023139, "tokens_per_second": 151778.5397582285, "data_wait_s": 9.416999091627076e-06, "gpu_peak_gb": 89.37861251831055}
 {"step": 621000, "loss": 0.5698689818382263, "lr": 0.0001, "grad_norm": 0.15833012025703624, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 90897686835, "tokens_per_second": 152322.75978387456, "data_wait_s": 1.432100179954432e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 621100, "loss": 0.5663334131240845, "lr": 0.0001, "grad_norm": 0.1626163194744334, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 90912295393, "tokens_per_second": 152126.60911954296, "data_wait_s": 1.0592997568892315e-05, "gpu_peak_gb": 89.37861251831055}
+{"step": 621200, "loss": 0.5962516069412231, "lr": 0.0001, "grad_norm": 0.15721248328626372, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 90927049456, "tokens_per_second": 153052.80672774845, "data_wait_s": 1.9547998817870393e-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 sha256:4b6effc9ccd8e405322e096b781a9f244c1f172dbf6edbcf19b7bc2f41128325
-size 357003
+oid sha256:832ba5a344bd874a9f4f47482d9b5ece9e22d2ac6b8ee7221e0866f84b326580
+size 359121