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": 76240.37723507191,
-      "required_pages_per_second_upper_estimate": 0.0066083226692665205
+      "required_token_positions_per_second": 76351.05237726738,
+      "required_pages_per_second_upper_estimate": 0.0066179157100884406
     },
     {
       "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": 9530.047154383989,
-      "required_pages_per_second_upper_estimate": 0.0038016782967863366
+      "required_token_positions_per_second": 9543.881547158422,
+      "required_pages_per_second_upper_estimate": 0.0038071970429066626
     },
     {
       "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": 19060.094308767977,
-      "required_pages_per_second_upper_estimate": 0.004409217372130548
+      "required_token_positions_per_second": 19087.763094316844,
+      "required_pages_per_second_upper_estimate": 0.004415618058713284
     },
     {
       "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": 28590.141463151966,
-      "required_pages_per_second_upper_estimate": 0.0017409896142263764
+      "required_token_positions_per_second": 28631.644641475265,
+      "required_pages_per_second_upper_estimate": 0.0017435169400359172
     },
     {
       "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": 28590.141463151966,
-      "required_pages_per_second_upper_estimate": 0.0022009482324103586
+      "required_token_positions_per_second": 28631.644641475265,
+      "required_pages_per_second_upper_estimate": 0.002204143261965837
     },
     {
       "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": 28590.141463151966,
-      "required_pages_per_second_upper_estimate": 0.354993871924109
+      "required_token_positions_per_second": 28631.644641475265,
+      "required_pages_per_second_upper_estimate": 0.35550920249668183
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,17 +158,17 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 690670,
-    "producer_batch_seconds": 0.052247876749923705,
-    "producer_p95_seconds": 0.15275691499937238,
-    "gpu_update_seconds": 0.9670466316381738,
-    "producer_batches_per_second": 19.13953374194216,
-    "gpu_batches_per_second": 1.0340762971336794,
-    "producer_headroom": 18.508821636270337,
+    "producer_batch_seconds": 0.052516312531160736,
+    "producer_p95_seconds": 0.13618497799961915,
+    "gpu_update_seconds": 0.965644843186885,
+    "producer_batches_per_second": 19.04170250732373,
+    "gpu_batches_per_second": 1.0355774248218774,
+    "producer_headroom": 18.38752183169594,
     "sustainable": true,
-    "buffer_seconds": 60.92393779320495,
+    "buffer_seconds": 60.835625120773756,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 84,
     "max_pending_pages": 4096,
     "generation_bytes_per_second": 0,
     "upload_bytes_per_second": null,

metrics/void-byte-v1/1790696019647648547-3fdea0a8/train.jsonl

@@ -37,3 +37,4 @@
 {"step": 278700, "loss": 0.5746679967269301, "lr": 0.0003, "grad_norm": 0.15670635691392462, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 40794039350, "tokens_per_second": 148513.58695770724, "data_wait_s": 1.127800078393193e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 278800, "loss": 0.6013021990656853, "lr": 0.0003, "grad_norm": 0.16295544019408925, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 40808645024, "tokens_per_second": 150309.74284495253, "data_wait_s": 1.3943999874754809e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 278900, "loss": 0.5618883538991213, "lr": 0.0003, "grad_norm": 0.1863066547620779, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 40823246291, "tokens_per_second": 150619.76547717387, "data_wait_s": 1.0720999853219837e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 279000, "loss": 0.6319178156554699, "lr": 0.0003, "grad_norm": 0.17287498253791164, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 40838042164, "tokens_per_second": 152059.73354522864, "data_wait_s": 1.053100004355656e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790696019.3e6d470a-8cc7-4d7e-a224-51a0f9a2ce55-85btc.22407.0.1790696019647648547-3fdea0a8

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:d667597cd22edc35726295996819816779b81ea3c0876bf020b655f6e4df5161
-size 82767
+oid sha256:6d5726fcc45b6549860802fa78efc5eb84b48c4b910a8a713e67c82f44f79ab7
+size 84885