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": 75420.9647685022,
-      "required_pages_per_second_upper_estimate": 0.006537298073446168
+      "required_token_positions_per_second": 75446.38401579497,
+      "required_pages_per_second_upper_estimate": 0.006539501349377012
     },
     {
       "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": 9427.620596062776,
-      "required_pages_per_second_upper_estimate": 0.0037608188112584874
+      "required_token_positions_per_second": 9430.798001974372,
+      "required_pages_per_second_upper_estimate": 0.0037620863259830747
     },
     {
       "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": 18855.24119212555,
-      "required_pages_per_second_upper_estimate": 0.0043618282088870395
+      "required_token_positions_per_second": 18861.596003948744,
+      "required_pages_per_second_upper_estimate": 0.004363298282761467
     },
     {
       "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": 28282.86178818833,
-      "required_pages_per_second_upper_estimate": 0.0017222778941929503
+      "required_token_positions_per_second": 28292.394005923117,
+      "required_pages_per_second_upper_estimate": 0.0017228583562413175
     },
     {
       "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": 28282.86178818833,
-      "required_pages_per_second_upper_estimate": 0.0021772929924270767
+      "required_token_positions_per_second": 28292.394005923117,
+      "required_pages_per_second_upper_estimate": 0.002178026808935168
     },
     {
       "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": 28282.86178818833,
-      "required_pages_per_second_upper_estimate": 0.35117848675997776
+      "required_token_positions_per_second": 28292.394005923117,
+      "required_pages_per_second_upper_estimate": 0.3512968450019633
     }
   ],
   "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": 0.5026795613593151,
-    "producer_p95_seconds": 0.20843738300027326,
-    "gpu_update_seconds": 0.9775531276522568,
-    "producer_batches_per_second": 1.9893388887661585,
-    "gpu_batches_per_second": 1.0229623042602838,
-    "producer_headroom": 1.9446844526736233,
+    "producer_batch_seconds": 0.06980385129682531,
+    "producer_p95_seconds": 0.20166413100014324,
+    "gpu_update_seconds": 0.9772237723754232,
+    "producer_batches_per_second": 14.32585711850945,
+    "gpu_batches_per_second": 1.0233070748670108,
+    "producer_headroom": 13.999568135861114,
     "sustainable": true,
-    "buffer_seconds": 60.60829391443992,
+    "buffer_seconds": 60.58787388727624,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 78,
+    "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 284788.4705175513,
-    "upload_bytes_per_second": 3726836.968530263,
+    "generation_bytes_per_second": 291273.4057928204,
+    "upload_bytes_per_second": 3694007.150635728,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 61.108976236771625,
+    "estimated_full_shards_per_hour": 62.500492385277354,
     "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": 1.4821450189999723,
-      "bytes": 5748205,
-      "pages": 24,
-      "shards": 2
-    },
     {
       "seconds": 1.3858465779999278,
       "bytes": 775414,
@@ -562,6 +556,12 @@
       "bytes": 6877765,
       "pages": 35,
       "shards": 3
+    },
+    {
+      "seconds": 1.007327716999498,
+      "bytes": 296016,
+      "pages": 9,
+      "shards": 1
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790651964532391049-1e9227c4/train.jsonl

@@ -77,3 +77,4 @@
 {"step": 244800, "loss": 0.633926959708333, "lr": 0.0003, "grad_norm": 0.1565239988882782, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 35831509297, "tokens_per_second": 148656.90972094, "data_wait_s": 1.440599953639321e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 244900, "loss": 0.6207933761179447, "lr": 0.0003, "grad_norm": 0.14457842399768928, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 35846077747, "tokens_per_second": 149183.63465190583, "data_wait_s": 1.7278998711844906e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 245000, "loss": 0.6086236331611872, "lr": 0.0003, "grad_norm": 0.15498698273521785, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 35860739606, "tokens_per_second": 148805.88176483728, "data_wait_s": 1.2215001333970577e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 245100, "loss": 0.6002775095403194, "lr": 0.0003, "grad_norm": 0.16302515660379907, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 35875346044, "tokens_per_second": 137269.7995699142, "data_wait_s": 1.375700048811268e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790651964.96f3217e-af24-484e-a73b-abea9eb37bd2-c4mbg.442.0.1790651964532391049-1e9227c4

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:248c4dcf572e5d69067e08c3d2cd0012f2d2144c1623133733009f712d4bc203
-size 169053
+oid sha256:b3dbfbcd7aebaf202185bbf9600c478b9858fa0037bb93dde2ec39f9b277f684
+size 171171