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": 76268.24350531543,
-      "required_pages_per_second_upper_estimate": 0.005276414052403908
+      "required_token_positions_per_second": 76619.41625708391,
+      "required_pages_per_second_upper_estimate": 0.005300708998204297
     },
     {
       "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": 9533.53043816443,
-      "required_pages_per_second_upper_estimate": 0.003991490111705443
+      "required_token_positions_per_second": 9577.427032135489,
+      "required_pages_per_second_upper_estimate": 0.004009868698936006
     },
     {
       "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": 19067.06087632886,
-      "required_pages_per_second_upper_estimate": 0.0039711659483299905
+      "required_token_positions_per_second": 19154.854064270978,
+      "required_pages_per_second_upper_estimate": 0.003989450954116273
     },
     {
       "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": 28600.591314493286,
-      "required_pages_per_second_upper_estimate": 0.0017145617787295436
+      "required_token_positions_per_second": 28732.281096406467,
+      "required_pages_per_second_upper_estimate": 0.0017224563800765846
     },
     {
       "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": 28600.591314493286,
-      "required_pages_per_second_upper_estimate": 0.002223570173995689
+      "required_token_positions_per_second": 28732.281096406467,
+      "required_pages_per_second_upper_estimate": 0.0022338084752973044
     },
     {
       "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": 28600.591314493286,
-      "required_pages_per_second_upper_estimate": 0.3668576764599388
+      "required_token_positions_per_second": 28732.281096406467,
+      "required_pages_per_second_upper_estimate": 0.3685468515848497
     }
   ],
   "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": 0.10028361990634949,
-    "producer_p95_seconds": 0.3005844260005688,
-    "gpu_update_seconds": 0.9666933000084316,
-    "producer_batches_per_second": 9.971718222116998,
-    "gpu_batches_per_second": 1.0344542576133278,
-    "producer_headroom": 9.63959319489249,
+    "producer_batch_seconds": 0.11286741078126283,
+    "producer_p95_seconds": 0.29360367600020254,
+    "gpu_update_seconds": 0.9622626169927706,
+    "producer_batches_per_second": 8.859953400880269,
+    "gpu_batches_per_second": 1.039217342896646,
+    "producer_headroom": 8.525601945965045,
     "sustainable": true,
-    "buffer_seconds": 60.90167790053119,
+    "buffer_seconds": 60.622544870544544,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 80,
+    "upload_pages_per_commit": 87,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 284965.9048300221,
-    "upload_bytes_per_second": 3197138.3493447485,
+    "generation_bytes_per_second": 324023.5411767437,
+    "upload_bytes_per_second": 3175563.9322219808,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 61.14704950976846,
+    "estimated_full_shards_per_hour": 69.52790905453428,
     "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": 1.5388405060002697,
-      "bytes": 6065532,
-      "pages": 36,
-      "shards": 2
-    },
     {
       "seconds": 1.3961634170009347,
       "bytes": 973137,
@@ -562,6 +556,12 @@
       "bytes": 5855746,
       "pages": 36,
       "shards": 2
+    },
+    {
+      "seconds": 1.3942693540011533,
+      "bytes": 2663289,
+      "pages": 34,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791300814123050567-72be911d/train.jsonl

@@ -303,3 +303,4 @@
 {"step": 830300, "loss": 0.511318027973175, "lr": 3e-05, "grad_norm": 0.2898206888323544, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 121532931866, "tokens_per_second": 151441.41417938436, "data_wait_s": 1.4012999599799514e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 830400, "loss": 0.6924585700035095, "lr": 3e-05, "grad_norm": 0.404697670905524, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 121547435539, "tokens_per_second": 150176.2794399814, "data_wait_s": 1.3621000107377768e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 830500, "loss": 0.5266762375831604, "lr": 3e-05, "grad_norm": 0.29181261907840583, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 121561936684, "tokens_per_second": 150365.47665462186, "data_wait_s": 1.2590000551426783e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 830600, "loss": 0.527016818523407, "lr": 3e-05, "grad_norm": 0.2842183373083332, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 121576469306, "tokens_per_second": 150481.12139268251, "data_wait_s": 1.2268003047211096e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791300814.0c9d201b-2dbc-40fb-b739-4ee2350a0f54-7b2tx.840.0.1791300814123050567-72be911d

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:e5ae221eed46bcb08101210a8995e02c8421732051ae304b14d91618662d83ef
-size 650853
+oid sha256:8af222b2d6bbea048b6b60e99d236c45f33e208f36c02d5805d9d457026e1487
+size 652971