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": 76614.54911222005,
-      "required_pages_per_second_upper_estimate": 0.005300372277829283
+      "required_token_positions_per_second": 76476.03866312363,
+      "required_pages_per_second_upper_estimate": 0.005290789803572273
     },
     {
       "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": 9576.818639027506,
-      "required_pages_per_second_upper_estimate": 0.004009613977446386
+      "required_token_positions_per_second": 9559.504832890454,
+      "required_pages_per_second_upper_estimate": 0.004002365048370189
     },
     {
       "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": 19153.637278055012,
-      "required_pages_per_second_upper_estimate": 0.003989197529636298
+      "required_token_positions_per_second": 19119.009665780908,
+      "required_pages_per_second_upper_estimate": 0.003981985511191148
     },
     {
       "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": 28730.455917082516,
-      "required_pages_per_second_upper_estimate": 0.0017223469633630012
+      "required_token_positions_per_second": 28678.514498671364,
+      "required_pages_per_second_upper_estimate": 0.0017192331546392063
     },
     {
       "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": 28730.455917082516,
-      "required_pages_per_second_upper_estimate": 0.0022336665756329827
+      "required_token_positions_per_second": 28678.514498671364,
+      "required_pages_per_second_upper_estimate": 0.00222962835881767
     },
     {
       "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": 28730.455917082516,
-      "required_pages_per_second_upper_estimate": 0.3685234401442069
+      "required_token_positions_per_second": 28678.514498671364,
+      "required_pages_per_second_upper_estimate": 0.3678571913991786
     }
   ],
   "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": 0.11559550772682314,
-    "producer_p95_seconds": 0.3330327630028478,
-    "gpu_update_seconds": 0.9623237473082037,
-    "producer_batches_per_second": 8.650855207654033,
-    "gpu_batches_per_second": 1.0391513280194775,
-    "producer_headroom": 8.324923400850318,
+    "producer_batch_seconds": 0.8526361567656409,
+    "producer_p95_seconds": 0.3472557479981333,
+    "gpu_update_seconds": 0.9640666709316793,
+    "producer_batches_per_second": 1.172833209177246,
+    "gpu_batches_per_second": 1.0372726598188426,
+    "producer_headroom": 1.1306894075296254,
     "sustainable": true,
-    "buffer_seconds": 60.62639608041683,
+    "buffer_seconds": 60.7362002686958,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 86,
+    "upload_pages_per_commit": 78,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 310446.57486348075,
-    "upload_bytes_per_second": 3983167.1614108244,
+    "generation_bytes_per_second": 265600.2828675304,
+    "upload_bytes_per_second": 4087731.5521492357,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 66.61460813930813,
+    "estimated_full_shards_per_hour": 56.99163784522471,
     "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": 1.3146963539984426,
-      "bytes": 5075870,
-      "pages": 30,
-      "shards": 2
-    },
     {
       "seconds": 1.278188054999191,
       "bytes": 7607915,
@@ -562,6 +556,12 @@
       "bytes": 1006990,
       "pages": 33,
       "shards": 1
+    },
+    {
+      "seconds": 1.5516541759934626,
+      "bytes": 17080870,
+      "pages": 28,
+      "shards": 5
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791342976705245012-6808dd50/train.jsonl

@@ -348,3 +348,4 @@
 {"step": 877300, "loss": 0.5222134590148926, "lr": 3e-05, "grad_norm": 0.2997342212671748, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 128412512155, "tokens_per_second": 152827.05755084864, "data_wait_s": 1.974500628421083e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 877400, "loss": 0.5608643293380737, "lr": 3e-05, "grad_norm": 0.32331489554311, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 128427011320, "tokens_per_second": 150429.5205340763, "data_wait_s": 8.13999940874055e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 877500, "loss": 0.49823275208473206, "lr": 3e-05, "grad_norm": 0.3163390680099127, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 128441726918, "tokens_per_second": 119242.43539023101, "data_wait_s": 1.7575999663677067e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 877600, "loss": 0.5132738947868347, "lr": 3e-05, "grad_norm": 0.34010128533589473, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 128456356294, "tokens_per_second": 151670.18337014448, "data_wait_s": 1.408400567015633e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791342976.f74d737f-395e-43a1-ae69-439b0359d002-2qp5c.418.0.1791342976705245012-6808dd50

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:0bd3ba24a59ec0b0705a86aa64dbac2fed2d69822afb16b5825db787935d364a
-size 746946
+oid sha256:c0e58ee595d3b18defa0a74aaba96e80a63e592973bf759e684fa795842efd29
+size 749064