Update TensorBoard training metrics

benchmark_cache/shard_scheduler.json

@@ -22,8 +22,8 @@
       "tokens_per_second": 1062967.6168812835,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 75362.25864704886,
-      "required_pages_per_second_upper_estimate": 0.006532209575627897
+      "required_token_positions_per_second": 75321.4321744952,
+      "required_pages_per_second_upper_estimate": 0.006528670840460694
     },
     {
       "source": "rewrite6",
@@ -36,8 +36,8 @@
       "tokens_per_second": 1206672.263377472,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 9420.282330881108,
-      "required_pages_per_second_upper_estimate": 0.003757891467560678
+      "required_token_positions_per_second": 9415.1790218119,
+      "required_pages_per_second_upper_estimate": 0.003755855681271701
     },
     {
       "source": "openmath",
@@ -50,8 +50,8 @@
       "tokens_per_second": 410042.7581134458,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 18840.564661762215,
-      "required_pages_per_second_upper_estimate": 0.004358433051885594
+      "required_token_positions_per_second": 18830.3580436238,
+      "required_pages_per_second_upper_estimate": 0.004356071930409616
     },
     {
       "source": "ultra_qa",
@@ -64,8 +64,8 @@
       "tokens_per_second": 2163486.5624068812,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28260.84699264332,
-      "required_pages_per_second_upper_estimate": 0.0017209373086469648
+      "required_token_positions_per_second": 28245.5370654357,
+      "required_pages_per_second_upper_estimate": 0.001720005014405001
     },
     {
       "source": "ultra_style",
@@ -78,8 +78,8 @@
       "tokens_per_second": 1546694.2737633225,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28260.84699264332,
-      "required_pages_per_second_upper_estimate": 0.0021755982325251675
+      "required_token_positions_per_second": 28245.5370654357,
+      "required_pages_per_second_upper_estimate": 0.0021744196319481335
     },
     {
       "source": "cortex",
@@ -92,8 +92,8 @@
       "tokens_per_second": 1015177.8395898147,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 22,
-      "required_token_positions_per_second": 28260.84699264332,
-      "required_pages_per_second_upper_estimate": 0.3509051366780898
+      "required_token_positions_per_second": 28245.5370654357,
+      "required_pages_per_second_upper_estimate": 0.3507150386212014
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 690670,
-    "producer_batch_seconds": 0.09776206959355704,
-    "producer_p95_seconds": 0.3090543500002241,
-    "gpu_update_seconds": 0.978314627555117,
-    "producer_batches_per_second": 10.228916021903698,
-    "gpu_batches_per_second": 1.0221660515278979,
-    "producer_headroom": 10.007098168261285,
+    "producer_batch_seconds": 0.1257646073591161,
+    "producer_p95_seconds": 0.3114236850015004,
+    "gpu_update_seconds": 0.9788449033894664,
+    "producer_batches_per_second": 7.951362636902588,
+    "gpu_batches_per_second": 1.0216123070542427,
+    "producer_headroom": 7.783150792133527,
     "sustainable": true,
-    "buffer_seconds": 60.65550690841725,
+    "buffer_seconds": 60.68838401014692,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 80,
+    "upload_pages_per_commit": 75,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 325314.7535407333,
-    "upload_bytes_per_second": 4143614.3058857257,
+    "generation_bytes_per_second": 338371.85013694776,
+    "upload_bytes_per_second": 4192393.36356571,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 69.80497316996097,
+    "estimated_full_shards_per_hour": 72.60672214585614,
     "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": 0.8558966809978301,
-      "bytes": 4770134,
-      "pages": 28,
-      "shards": 2
-    },
     {
       "seconds": 1.0199386870008311,
       "bytes": 6036267,
@@ -562,6 +556,12 @@
       "bytes": 6670501,
       "pages": 27,
       "shards": 2
+    },
+    {
+      "seconds": 1.1155471289966954,
+      "bytes": 10846938,
+      "pages": 39,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790738827394728502-c744ae20/train.jsonl

@@ -281,3 +281,4 @@
 {"step": 343100, "loss": 0.5893463902175426, "lr": 0.0003, "grad_norm": 0.16156630115794485, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 50220664206, "tokens_per_second": 133755.84668161237, "data_wait_s": 1.5747998986626044e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 343200, "loss": 0.5764463171362877, "lr": 0.0003, "grad_norm": 0.13561009381328162, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 50235274808, "tokens_per_second": 148330.61295416395, "data_wait_s": 1.5412002539960667e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 343300, "loss": 0.5748323947191238, "lr": 0.0003, "grad_norm": 0.17213277630921067, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 50249887136, "tokens_per_second": 148665.9782444599, "data_wait_s": 1.631099803489633e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 343400, "loss": 0.5832381173968315, "lr": 0.0003, "grad_norm": 0.1490679566560398, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 50264490100, "tokens_per_second": 149142.0098490905, "data_wait_s": 1.70599996636156e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790738827.1a46e860-412e-40b5-8c01-42d24543f8c9-dfnph.319.0.1790738827394728502-c744ae20

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:e91c0d663e0cb77a56403ef0aee201a052d5e3bb82c74f1bfddc7bf9b9a754d3
-size 603474
+oid sha256:6858a6b260226a741825aac254fe0099f444807c8eca3171e47e5af278eb1142
+size 605592