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": 76084.96331764552,
-      "required_pages_per_second_upper_estimate": 0.006594851784796974
+      "required_token_positions_per_second": 74991.3450559129,
+      "required_pages_per_second_upper_estimate": 0.006500059725620134
     },
     {
       "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": 9510.62041470569,
-      "required_pages_per_second_upper_estimate": 0.0037939286798730214
+      "required_token_positions_per_second": 9373.918131989112,
+      "required_pages_per_second_upper_estimate": 0.0037393960954161132
     },
     {
       "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": 19021.24082941138,
-      "required_pages_per_second_upper_estimate": 0.004400229303479281
+      "required_token_positions_per_second": 18747.836263978224,
+      "required_pages_per_second_upper_estimate": 0.00433698196902233
     },
     {
       "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": 28531.861244117074,
-      "required_pages_per_second_upper_estimate": 0.0017374406546598294
+      "required_token_positions_per_second": 28121.754395967335,
+      "required_pages_per_second_upper_estimate": 0.001712467299271853
     },
     {
       "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": 28531.861244117074,
-      "required_pages_per_second_upper_estimate": 0.0021964616598190807
+      "required_token_positions_per_second": 28121.754395967335,
+      "required_pages_per_second_upper_estimate": 0.002164890499400099
     },
     {
       "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": 28531.861244117074,
-      "required_pages_per_second_upper_estimate": 0.35427022665504143
+      "required_token_positions_per_second": 28121.754395967335,
+      "required_pages_per_second_upper_estimate": 0.34917807214034957
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 690670,
-    "producer_batch_seconds": 0.08052424352349874,
-    "producer_p95_seconds": 0.20305411600020307,
-    "gpu_update_seconds": 0.9690219563121101,
-    "producer_batches_per_second": 12.418620234639995,
-    "gpu_batches_per_second": 1.0319683609706696,
-    "producer_headroom": 12.033915674468004,
+    "producer_batch_seconds": 0.8869138885626029,
+    "producer_p95_seconds": 0.20045495699923777,
+    "gpu_update_seconds": 0.9831534551757812,
+    "producer_batches_per_second": 1.1275051759767487,
+    "gpu_batches_per_second": 1.017135213974513,
+    "producer_headroom": 1.1085106094901176,
     "sustainable": true,
-    "buffer_seconds": 60.07936129135082,
+    "buffer_seconds": 60.955514220898436,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 87,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 320856.4701423033,
-    "upload_bytes_per_second": 5502546.710543686,
+    "generation_bytes_per_second": 264763.40548813896,
+    "upload_bytes_per_second": 5579095.736845195,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 68.84832933618378,
+    "estimated_full_shards_per_hour": 56.81206344111563,
     "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": [
@@ -382,6 +382,18 @@
       "bytes": 6298017,
       "pages": 35,
       "shards": 2
+    },
+    {
+      "seconds": 0.9624441889991431,
+      "bytes": 1027701,
+      "pages": 30,
+      "shards": 1
+    },
+    {
+      "seconds": 1.7364580140001635,
+      "bytes": 17104134,
+      "pages": 27,
+      "shards": 5
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790703713325746110-adad6d18/train.jsonl

@@ -41,3 +41,4 @@
 {"step": 284300, "loss": 0.564514322206378, "lr": 0.0003, "grad_norm": 0.15623138993272118, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 41613899464, "tokens_per_second": 148956.48427583487, "data_wait_s": 1.3965000107418746e-05, "gpu_peak_gb": 85.46991729736328}
 {"step": 284400, "loss": 0.6265847701579332, "lr": 0.0003, "grad_norm": 0.16860107929567475, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 41628432200, "tokens_per_second": 148841.90353205925, "data_wait_s": 1.3670998669113033e-05, "gpu_peak_gb": 85.46991729736328}
 {"step": 284500, "loss": 0.5872170403599739, "lr": 0.0003, "grad_norm": 0.16318429725297312, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 41642989966, "tokens_per_second": 150177.27312181768, "data_wait_s": 1.2517999493866228e-05, "gpu_peak_gb": 85.46991729736328}
+{"step": 284600, "loss": 0.6178310494869947, "lr": 0.0003, "grad_norm": 0.1553548230141656, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 41657633602, "tokens_per_second": 96489.93389584642, "data_wait_s": 9.039000360644422e-06, "gpu_peak_gb": 85.46991729736328}

runs/void-byte-v1/events.out.tfevents.1790703713.3e6d470a-8cc7-4d7e-a224-51a0f9a2ce55-85btc.38629.0.1790703713325746110-adad6d18

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:fa672f2694f22a94f9219d9d132675027695f41d35631ed0b0eefc1097193691
-size 91239
+oid sha256:665db270ba1fc9e7382fb0e3add8fcedc7763633fac1ceb75979f90617070a3e
+size 93357