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": 75493.8686528651,
-      "required_pages_per_second_upper_estimate": 0.006543617197369538
+      "required_token_positions_per_second": 75152.24271030424,
+      "required_pages_per_second_upper_estimate": 0.006514005926511399
     },
     {
       "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": 9436.733581608138,
-      "required_pages_per_second_upper_estimate": 0.003764454117443808
+      "required_token_positions_per_second": 9394.03033878803,
+      "required_pages_per_second_upper_estimate": 0.0037474191554124903
     },
     {
       "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": 18873.467163216275,
-      "required_pages_per_second_upper_estimate": 0.0043660444665327425
+      "required_token_positions_per_second": 18788.06067757606,
+      "required_pages_per_second_upper_estimate": 0.0043462871791826885
     },
     {
       "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": 28310.20074482441,
-      "required_pages_per_second_upper_estimate": 0.0017239426932156674
+      "required_token_positions_per_second": 28182.091016364087,
+      "required_pages_per_second_upper_estimate": 0.0017161414828922334
     },
     {
       "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": 28310.20074482441,
-      "required_pages_per_second_upper_estimate": 0.0021793976209879976
+      "required_token_positions_per_second": 28182.091016364087,
+      "required_pages_per_second_upper_estimate": 0.002169535379460696
     },
     {
       "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": 28310.20074482441,
-      "required_pages_per_second_upper_estimate": 0.3515179451037959
+      "required_token_positions_per_second": 28182.091016364087,
+      "required_pages_per_second_upper_estimate": 0.3499272510319988
     }
   ],
   "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": 0.14357713656238502,
-    "producer_p95_seconds": 0.3675366429997666,
-    "gpu_update_seconds": 0.9766091116487235,
-    "producer_batches_per_second": 6.9648972248829795,
-    "gpu_batches_per_second": 1.0239511264765773,
-    "producer_headroom": 6.801982091517626,
+    "producer_batch_seconds": 0.08793713420297422,
+    "producer_p95_seconds": 0.3571137830003863,
+    "gpu_update_seconds": 0.9810485667634113,
+    "producer_batches_per_second": 11.371760167801533,
+    "gpu_batches_per_second": 1.0193175280802984,
+    "producer_headroom": 11.156249014198943,
     "sustainable": true,
-    "buffer_seconds": 60.54976492222086,
+    "buffer_seconds": 60.8250111393315,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 78,
+    "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 307690.19375241845,
-    "upload_bytes_per_second": 6310512.311635319,
+    "generation_bytes_per_second": 333913.20401496394,
+    "upload_bytes_per_second": 6315527.755867869,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 66.02315291814246,
+    "estimated_full_shards_per_hour": 71.65000048004806,
     "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": 1.1784233340003993,
-      "bytes": 7683890,
-      "pages": 38,
-      "shards": 3
-    },
     {
       "seconds": 1.1247381559987844,
       "bytes": 11305256,
@@ -562,6 +556,12 @@
       "bytes": 7158906,
       "pages": 42,
       "shards": 3
+    },
+    {
+      "seconds": 0.8130094469997857,
+      "bytes": 5715102,
+      "pages": 28,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790390190216476996-4d8dc439/train.jsonl

@@ -300,3 +300,4 @@
 {"step": 30000, "loss": 0.6875663865357637, "lr": 0.0003, "grad_norm": 0.14237463414510249, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 4389888866, "tokens_per_second": 149914.74994791884, "data_wait_s": 9.903000318445265e-06, "gpu_peak_gb": 85.43514251708984}
 {"step": 30100, "loss": 0.6164825251325965, "lr": 0.0003, "grad_norm": 0.13171608209137095, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 4404659166, "tokens_per_second": 127771.60330143584, "data_wait_s": 1.5090001397766173e-05, "gpu_peak_gb": 85.43514251708984}
 {"step": 30200, "loss": 0.6555165983736515, "lr": 0.0003, "grad_norm": 0.1274789077023394, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 4419506132, "tokens_per_second": 150744.9750210262, "data_wait_s": 1.0255997040076181e-05, "gpu_peak_gb": 85.43514251708984}
+{"step": 30300, "loss": 0.6104271803051233, "lr": 0.0003, "grad_norm": 0.14621686984475263, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 4434240188, "tokens_per_second": 149897.92644218702, "data_wait_s": 1.0044997907243669e-05, "gpu_peak_gb": 85.43514251708984}

runs/void-byte-v1/events.out.tfevents.1790390190.c8125a87-1345-4bfd-b42f-0e32ed8b7bb7-6w295.451.0.1790390190216476996-4d8dc439

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:a12b3be235871e7a074ab5d5e8014955d2d6ff12a93a8dbfe9ac95cc80576245
-size 638725
+oid sha256:831296bd26c15cdc770005800df709f03cc395877e5191911a0b2a02c45d36ce
+size 640843