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": 76800.61010024272,
-      "required_pages_per_second_upper_estimate": 0.005313244408701664
+      "required_token_positions_per_second": 76410.17242326398,
+      "required_pages_per_second_upper_estimate": 0.00528623302426805
     },
     {
       "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": 9600.07626253034,
-      "required_pages_per_second_upper_estimate": 0.004019351458732616
+      "required_token_positions_per_second": 9551.271552907998,
+      "required_pages_per_second_upper_estimate": 0.003998917945971971
     },
     {
       "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": 19200.15252506068,
-      "required_pages_per_second_upper_estimate": 0.003998885428898024
+      "required_token_positions_per_second": 19102.543105815996,
+      "required_pages_per_second_upper_estimate": 0.003978555961002845
     },
     {
       "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": 28800.228787591022,
-      "required_pages_per_second_upper_estimate": 0.0017265297404129817
+      "required_token_positions_per_second": 28653.814658723993,
+      "required_pages_per_second_upper_estimate": 0.0017177524369488334
     },
     {
       "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": 28800.228787591022,
-      "required_pages_per_second_upper_estimate": 0.0022390911094165955
+      "required_token_positions_per_second": 28653.814658723993,
+      "required_pages_per_second_upper_estimate": 0.0022277080549048763
     },
     {
       "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": 28800.228787591022,
-      "required_pages_per_second_upper_estimate": 0.3694184116108185
+      "required_token_positions_per_second": 28653.814658723993,
+      "required_pages_per_second_upper_estimate": 0.3675403683729556
     }
   ],
   "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": 0.10769118219531038,
-    "producer_p95_seconds": 0.3239337299964973,
-    "gpu_update_seconds": 0.9599923738075485,
-    "producer_batches_per_second": 9.285811332132882,
-    "gpu_batches_per_second": 1.0416749416808095,
-    "producer_headroom": 8.914308063463281,
+    "producer_batch_seconds": 0.2376439335392888,
+    "producer_p95_seconds": 0.34044266199634876,
+    "gpu_update_seconds": 0.9648977048709373,
+    "producer_batches_per_second": 4.207976130956752,
+    "gpu_batches_per_second": 1.0363792917651906,
+    "producer_headroom": 4.060266510911856,
     "sustainable": true,
-    "buffer_seconds": 60.479519549875555,
+    "buffer_seconds": 60.78855540686905,
     "buffer_drain_seconds": null,
     "shard_target_mib": 32,
-    "upload_pages_per_commit": 160,
+    "upload_pages_per_commit": 158,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 713215.2180970903,
+    "generation_bytes_per_second": 661683.7518257533,
     "upload_bytes_per_second": 3833977.5236239615,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 76.51969150154366,
+    "estimated_full_shards_per_hour": 70.99096496619914,
     "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": [

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

@@ -345,3 +345,4 @@
 {"step": 834500, "loss": 0.5453035235404968, "lr": 3e-05, "grad_norm": 0.3164018617622983, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 122147057444, "tokens_per_second": 149809.1086840898, "data_wait_s": 9.306997526437044e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 834600, "loss": 0.5199332237243652, "lr": 3e-05, "grad_norm": 0.27532710486133033, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 122161579239, "tokens_per_second": 150612.0426535751, "data_wait_s": 1.3378004950936884e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 834700, "loss": 0.48385611176490784, "lr": 3e-05, "grad_norm": 0.28114723383564727, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 122176097630, "tokens_per_second": 150496.268789307, "data_wait_s": 9.189003321807832e-06, "gpu_peak_gb": 89.37899398803711}
+{"step": 834800, "loss": 0.5412554144859314, "lr": 3e-05, "grad_norm": 0.2698154589368708, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 122190688973, "tokens_per_second": 150956.17025571744, "data_wait_s": 7.982001989148557e-06, "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:afad077e1c952c3fc44ad0f284103bd2a6757ee098f74df288d0caa9edc7f088
-size 739809
+oid sha256:82ad61534163a414abd109cdb9a17f8dc93721c3994491ec1550f6e63856c94f
+size 741927