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": 77155.68582227238,
-      "required_pages_per_second_upper_estimate": 0.005337809371040861
+      "required_token_positions_per_second": 77022.2359438986,
+      "required_pages_per_second_upper_estimate": 0.005328576998808583
     },
     {
       "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": 9644.460727784048,
-      "required_pages_per_second_upper_estimate": 0.004037934307481314
+      "required_token_positions_per_second": 9627.779492987325,
+      "required_pages_per_second_upper_estimate": 0.00403095022281572
     },
     {
       "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": 19288.921455568096,
-      "required_pages_per_second_upper_estimate": 0.004017373656128596
+      "required_token_positions_per_second": 19255.55898597465,
+      "required_pages_per_second_upper_estimate": 0.004010425133539771
     },
     {
       "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": 28933.382183352143,
-      "required_pages_per_second_upper_estimate": 0.0017345120831753977
+      "required_token_positions_per_second": 28883.338478961974,
+      "required_pages_per_second_upper_estimate": 0.0017315120395095193
     },
     {
       "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": 28933.382183352143,
-      "required_pages_per_second_upper_estimate": 0.0022494432002571315
+      "required_token_positions_per_second": 28883.338478961974,
+      "required_pages_per_second_upper_estimate": 0.0022455525223597883
     },
     {
       "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": 28933.382183352143,
-      "required_pages_per_second_upper_estimate": 0.3711263604026647
+      "required_token_positions_per_second": 28883.338478961974,
+      "required_pages_per_second_upper_estimate": 0.37048445349549103
     }
   ],
   "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": 0.3744751613436961,
-    "producer_p95_seconds": 0.30265087600128027,
-    "gpu_update_seconds": 0.9555744235082294,
-    "producer_batches_per_second": 2.6704040834425133,
-    "gpu_batches_per_second": 1.0464909643862899,
-    "producer_headroom": 2.5517698425696014,
+    "producer_batch_seconds": 0.08721789763274046,
+    "producer_p95_seconds": 0.30104131199914264,
+    "gpu_update_seconds": 0.9572300660513406,
+    "producer_batches_per_second": 11.465536628856015,
+    "gpu_batches_per_second": 1.0446809345689372,
+    "producer_headroom": 10.975156384553909,
     "sustainable": true,
-    "buffer_seconds": 60.20118868101845,
+    "buffer_seconds": 60.30549416123446,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 76,
+    "upload_pages_per_commit": 81,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 409028.67553819035,
-    "upload_bytes_per_second": 5639558.632833212,
+    "generation_bytes_per_second": 394480.1719436193,
+    "upload_bytes_per_second": 5803467.822931092,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 87.76803207024844,
+    "estimated_full_shards_per_hour": 84.646261870684,
     "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": 1.516641152997181,
-      "bytes": 1120266,
-      "pages": 34,
-      "shards": 1
-    },
     {
       "seconds": 1.2580970729977707,
       "bytes": 5952993,
@@ -562,6 +556,12 @@
       "bytes": 6398380,
       "pages": 33,
       "shards": 3
+    },
+    {
+      "seconds": 1.3141280880008708,
+      "bytes": 12092914,
+      "pages": 40,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791143623306220418-eb05a537/train.jsonl

@@ -355,3 +355,4 @@
 {"step": 720500, "loss": 0.5453026294708252, "lr": 8e-05, "grad_norm": 0.2268957195603477, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 105461179941, "tokens_per_second": 153552.91413386018, "data_wait_s": 2.015000063693151e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 720600, "loss": 0.5363257527351379, "lr": 8e-05, "grad_norm": 0.23791260345480333, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 105475828610, "tokens_per_second": 152334.94725441295, "data_wait_s": 1.4915000065229833e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 720700, "loss": 0.5245054364204407, "lr": 8e-05, "grad_norm": 0.2286034159329082, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 105490324667, "tokens_per_second": 150841.09056894577, "data_wait_s": 1.5284000255633146e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 720800, "loss": 0.5011938214302063, "lr": 8e-05, "grad_norm": 0.2403712256520014, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 105504891856, "tokens_per_second": 151785.59926973566, "data_wait_s": 9.578994649928063e-06, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791143623.3eb6ddb3-ffce-42e5-bd48-c0d653189e98-8pb7d.369.0.1791143623306220418-eb05a537

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:bf77bbcee400cbb4b43a726003221864ed8247d22328fb3644c0ca4ad2f8c0d6
-size 761772
+oid sha256:3c49cedf04d791d8f51f93554d86fe4176c17491e0f0021a858d8529a0598000
+size 763890