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": 75926.31583868862,
-      "required_pages_per_second_upper_estimate": 0.006581100623409349
+      "required_token_positions_per_second": 75590.28506419691,
+      "required_pages_per_second_upper_estimate": 0.006551974327538619
     },
     {
       "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": 9490.789479836078,
-      "required_pages_per_second_upper_estimate": 0.0037860178234546344
+      "required_token_positions_per_second": 9448.785633024614,
+      "required_pages_per_second_upper_estimate": 0.003769261860948067
     },
     {
       "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": 18981.578959672155,
-      "required_pages_per_second_upper_estimate": 0.004391054227940178
+      "required_token_positions_per_second": 18897.571266049228,
+      "required_pages_per_second_upper_estimate": 0.004371620526505425
     },
     {
       "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": 28472.36843950823,
-      "required_pages_per_second_upper_estimate": 0.0017338178549937736
+      "required_token_positions_per_second": 28346.356899073842,
+      "required_pages_per_second_upper_estimate": 0.001726144413312777
     },
     {
       "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": 28472.36843950823,
-      "required_pages_per_second_upper_estimate": 0.0021918817390336686
+      "required_token_positions_per_second": 28346.356899073842,
+      "required_pages_per_second_upper_estimate": 0.0021821810218287598
     },
     {
       "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": 28472.36843950823,
-      "required_pages_per_second_upper_estimate": 0.3535315251314083
+      "required_token_positions_per_second": 28346.356899073842,
+      "required_pages_per_second_upper_estimate": 0.3519668835327097
     }
   ],
   "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": 688946,
-    "producer_batch_seconds": 0.08110531017183575,
-    "producer_p95_seconds": 0.3143580930009193,
-    "gpu_update_seconds": 0.9710467205684112,
-    "producer_batches_per_second": 12.329648920413787,
-    "gpu_batches_per_second": 1.029816566822491,
-    "producer_headroom": 11.972665149927659,
+    "producer_batch_seconds": 0.13663193298444298,
+    "producer_p95_seconds": 0.358894037999562,
+    "gpu_update_seconds": 0.975363433771743,
+    "producer_batches_per_second": 7.318933269529757,
+    "gpu_batches_per_second": 1.0252588577500668,
+    "producer_headroom": 7.138619885314793,
     "sustainable": true,
-    "buffer_seconds": 60.2048966752415,
+    "buffer_seconds": 60.47253289384806,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 84,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 297497.60232783377,
-    "upload_bytes_per_second": 2775794.8291046056,
+    "generation_bytes_per_second": 275765.6774725651,
+    "upload_bytes_per_second": 2838905.760038997,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 63.83606006981144,
+    "estimated_full_shards_per_hour": 59.172894889189855,
     "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": 0.8121612659997481,
-      "bytes": 2238306,
-      "pages": 27,
-      "shards": 2
-    },
     {
       "seconds": 1.133708954002941,
       "bytes": 6148950,
@@ -562,6 +556,12 @@
       "bytes": 5451054,
       "pages": 32,
       "shards": 2
+    },
+    {
+      "seconds": 1.2875092799986305,
+      "bytes": 13122964,
+      "pages": 36,
+      "shards": 4
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790606918076636751-3952f1c5/train.jsonl

@@ -278,3 +278,4 @@
 {"step": 222800, "loss": 0.5395330935716629, "lr": 0.0003, "grad_norm": 0.15129146284377312, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 32611032732, "tokens_per_second": 149619.1039673062, "data_wait_s": 1.046399847837165e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 222900, "loss": 0.5923827104270458, "lr": 0.0003, "grad_norm": 0.14402803898302785, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 32625667614, "tokens_per_second": 149895.4228098872, "data_wait_s": 1.2005999451503158e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 223000, "loss": 0.6031157094985247, "lr": 0.0003, "grad_norm": 0.14561991782955186, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 32640289553, "tokens_per_second": 147825.07847140328, "data_wait_s": 1.4483000995824113e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 223100, "loss": 0.6118058785796165, "lr": 0.0003, "grad_norm": 0.13831426155139062, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 32654974765, "tokens_per_second": 148573.64504332296, "data_wait_s": 1.252899892278947e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790606918.0a7f8bdc-23d3-4ecb-bd0c-4a4bf460026e-rl2cx.1192.0.1790606918076636751-3952f1c5

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:33fe76381c9a82e284b557bc9e473b0b99bcab64280da08dd435a609241ebf90
-size 597120
+oid sha256:fa8df88201fba845d22dd86cd58801dce69cf3b27d992a657ed2249598fadc96
+size 599238