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": 75567.30063787976,
-      "required_pages_per_second_upper_estimate": 0.006549982095718944
+      "required_token_positions_per_second": 75415.26384765975,
+      "required_pages_per_second_upper_estimate": 0.006536803932076396
     },
     {
       "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": 9445.91257973497,
-      "required_pages_per_second_upper_estimate": 0.003768115757034853
+      "required_token_positions_per_second": 9426.907980957469,
+      "required_pages_per_second_upper_estimate": 0.0037605345384384347
     },
     {
       "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": 18891.82515946994,
-      "required_pages_per_second_upper_estimate": 0.004370291265876329
+      "required_token_positions_per_second": 18853.815961914937,
+      "required_pages_per_second_upper_estimate": 0.004361498506961009
     },
     {
       "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": 28337.73773920491,
-      "required_pages_per_second_upper_estimate": 0.001725619551697996
+      "required_token_positions_per_second": 28280.72394287241,
+      "required_pages_per_second_upper_estimate": 0.0017221477106296183
     },
     {
       "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": 28337.73773920491,
-      "required_pages_per_second_upper_estimate": 0.002181517495042688
+      "required_token_positions_per_second": 28280.72394287241,
+      "required_pages_per_second_upper_estimate": 0.0021771284151767224
     },
     {
       "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": 28337.73773920491,
-      "required_pages_per_second_upper_estimate": 0.35185986241361
+      "required_token_positions_per_second": 28280.72394287241,
+      "required_pages_per_second_upper_estimate": 0.3511519418760621
     }
   ],
   "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": 0.039901346640689894,
-    "producer_p95_seconds": 0.1367718750007043,
-    "gpu_update_seconds": 0.9756600987152667,
-    "producer_batches_per_second": 25.061810795634592,
-    "gpu_batches_per_second": 1.0249471115163813,
-    "producer_headroom": 24.45180879485218,
+    "producer_batch_seconds": 0.04111052343752419,
+    "producer_p95_seconds": 0.13983913300035056,
+    "gpu_update_seconds": 0.9776270245361992,
+    "producer_batches_per_second": 24.32467203974436,
+    "gpu_batches_per_second": 1.022884980572642,
+    "producer_headroom": 23.78045674903416,
     "sustainable": true,
-    "buffer_seconds": 60.490926120346536,
+    "buffer_seconds": 60.61287552124435,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 82,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
     "generation_bytes_per_second": 0,
     "upload_bytes_per_second": null,

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

@@ -39,3 +39,4 @@
 {"step": 88900, "loss": 0.6370924413204193, "lr": 0.0003, "grad_norm": 0.13101865207589453, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 13010034356, "tokens_per_second": 149930.09538276383, "data_wait_s": 1.265399987460114e-05, "gpu_peak_gb": 85.46991729736328}
 {"step": 89000, "loss": 0.642592053860426, "lr": 0.0003, "grad_norm": 0.14786466122497507, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 13024704107, "tokens_per_second": 150067.41889543916, "data_wait_s": 9.576000593369827e-06, "gpu_peak_gb": 85.46991729736328}
 {"step": 89100, "loss": 0.5991051606833935, "lr": 0.0003, "grad_norm": 0.13633313821056686, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 13039297606, "tokens_per_second": 149385.12140472018, "data_wait_s": 1.4216999261407182e-05, "gpu_peak_gb": 85.46991729736328}
+{"step": 89200, "loss": 0.6012036129832268, "lr": 0.0003, "grad_norm": 0.18671982782268762, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 13053837626, "tokens_per_second": 148796.3517379957, "data_wait_s": 1.532500027678907e-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:ede9b6fa58c1236175d3fd6253541f5f95b86e604af5cca051c7c95e8378742e
-size 87003
+oid sha256:093a0e666effcfa9b4f9d3a640d5da62a914d3483ad94c1f52edd456d95e52d6
+size 89121