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": 75528.1408991966,
-      "required_pages_per_second_upper_estimate": 0.006546587828819339
+      "required_token_positions_per_second": 75438.99544344687,
+      "required_pages_per_second_upper_estimate": 0.006538860926652044
     },
     {
       "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": 9441.017612399575,
-      "required_pages_per_second_upper_estimate": 0.0037661630813784807
+      "required_token_positions_per_second": 9429.874430430858,
+      "required_pages_per_second_upper_estimate": 0.0037617178994857423
     },
     {
       "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": 18882.03522479915,
-      "required_pages_per_second_upper_estimate": 0.004368026536787207
+      "required_token_positions_per_second": 18859.748860861717,
+      "required_pages_per_second_upper_estimate": 0.004362870978716891
     },
     {
       "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": 28323.052837198728,
-      "required_pages_per_second_upper_estimate": 0.001724725318211546
+      "required_token_positions_per_second": 28289.623291292577,
+      "required_pages_per_second_upper_estimate": 0.0017226896342571337
     },
     {
       "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": 28323.052837198728,
-      "required_pages_per_second_upper_estimate": 0.0021803870106358414
+      "required_token_positions_per_second": 28289.623291292577,
+      "required_pages_per_second_upper_estimate": 0.00217781351165308
     },
     {
       "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": 28323.052837198728,
-      "required_pages_per_second_upper_estimate": 0.3516775250778987
+      "required_token_positions_per_second": 28289.623291292577,
+      "required_pages_per_second_upper_estimate": 0.35126244199923734
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,17 +158,17 @@
     "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.04154882489058309,
-    "producer_p95_seconds": 0.13499132100150746,
-    "gpu_update_seconds": 0.9761659577772587,
-    "producer_batches_per_second": 24.068069377014965,
-    "gpu_batches_per_second": 1.0244159735676623,
-    "producer_headroom": 23.49442999526332,
+    "producer_batch_seconds": 0.04033760042965184,
+    "producer_p95_seconds": 0.13031000000046333,
+    "gpu_update_seconds": 0.9773194826708751,
+    "producer_batches_per_second": 24.790765671448025,
+    "gpu_batches_per_second": 1.0232068609408485,
+    "producer_headroom": 24.228498281034472,
     "sustainable": true,
-    "buffer_seconds": 60.52228938219004,
+    "buffer_seconds": 60.593807925594255,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 84,
+    "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
     "generation_bytes_per_second": 0,
     "upload_bytes_per_second": null,

metrics/void-byte-v1/1790485060735632541-7579de36/train.jsonl

@@ -24,3 +24,4 @@
 {"step": 87400, "loss": 0.6537573337554932, "lr": 0.0003, "grad_norm": 0.13270455503542447, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 12790268643, "tokens_per_second": 148972.17118350157, "data_wait_s": 1.534100010758266e-05, "gpu_peak_gb": 85.46991729736328}
 {"step": 87500, "loss": 0.6125136204063892, "lr": 0.0003, "grad_norm": 0.13232944352760084, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 12804833533, "tokens_per_second": 149206.4226659045, "data_wait_s": 8.506000085617416e-06, "gpu_peak_gb": 85.46991729736328}
 {"step": 87600, "loss": 0.6151273623108864, "lr": 0.0003, "grad_norm": 0.1302308393090945, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 12819477179, "tokens_per_second": 149952.52009716092, "data_wait_s": 1.0099000064656138e-05, "gpu_peak_gb": 85.46991729736328}
+{"step": 87700, "loss": 0.5613355282694101, "lr": 0.0003, "grad_norm": 0.13900986651536845, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 12834214746, "tokens_per_second": 150770.54922077234, "data_wait_s": 1.0669000403140672e-05, "gpu_peak_gb": 85.46991729736328}

runs/void-byte-v1/events.out.tfevents.1790485060.0b09bbb2-f535-42d2-bd00-cbae7fb94f51-r2qrq.28895.0.1790485060735632541-7579de36

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:2e818fc85159936ced6046d8007b708c41d1c46b68afb7910484ab911bb8b99b
-size 55233
+oid sha256:2b5f7c735f1a7c0574679da5770b31aaaf6bd0e87f19e5e2bc87610529180d26
+size 57351