Update TensorBoard training metrics

benchmark_cache/shard_scheduler.json

@@ -22,8 +22,8 @@
       "tokens_per_second": 598315.063954628,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 75948.0229487074,
-      "required_pages_per_second_upper_estimate": 0.005254260451283705
+      "required_token_positions_per_second": 76013.25269845844,
+      "required_pages_per_second_upper_estimate": 0.005258773196725354
     },
     {
       "source": "rewrite6",
@@ -36,8 +36,8 @@
       "tokens_per_second": 8533.957213384612,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 9493.502868588424,
-      "required_pages_per_second_upper_estimate": 0.003974731404194673
+      "required_token_positions_per_second": 9501.656587307305,
+      "required_pages_per_second_upper_estimate": 0.0039781451959532595
     },
     {
       "source": "openmath",
@@ -50,8 +50,8 @@
       "tokens_per_second": 380144.35761751566,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 18987.00573717685,
-      "required_pages_per_second_upper_estimate": 0.003954492574040618
+      "required_token_positions_per_second": 19003.31317461461,
+      "required_pages_per_second_upper_estimate": 0.003957888983202859
     },
     {
       "source": "ultra_qa",
@@ -64,8 +64,8 @@
       "tokens_per_second": 43742.16556775435,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28480.508605765273,
-      "required_pages_per_second_upper_estimate": 0.0017073630036969789
+      "required_token_positions_per_second": 28504.969761921915,
+      "required_pages_per_second_upper_estimate": 0.0017088294126585467
     },
     {
       "source": "ultra_style",
@@ -78,8 +78,8 @@
       "tokens_per_second": 22018.86276523721,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28480.508605765273,
-      "required_pages_per_second_upper_estimate": 0.002214234271580101
+      "required_token_positions_per_second": 28504.969761921915,
+      "required_pages_per_second_upper_estimate": 0.0022161360188780246
     },
     {
       "source": "cortex",
@@ -92,8 +92,8 @@
       "tokens_per_second": 910315.184299418,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 22,
-      "required_token_positions_per_second": 28480.508605765273,
-      "required_pages_per_second_upper_estimate": 0.36531738440714295
+      "required_token_positions_per_second": 28504.969761921915,
+      "required_pages_per_second_upper_estimate": 0.36563114585397716
     }
   ],
   "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": 704566,
-    "producer_batch_seconds": 0.14875617208599579,
-    "producer_p95_seconds": 0.35381016099563567,
-    "gpu_update_seconds": 0.970769180519594,
-    "producer_batches_per_second": 6.722410142564714,
-    "gpu_batches_per_second": 1.030110988345098,
-    "producer_headroom": 6.525908585214155,
+    "producer_batch_seconds": 0.11744215000038594,
+    "producer_p95_seconds": 0.35164239600271685,
+    "gpu_update_seconds": 0.9699361280127828,
+    "producer_batches_per_second": 8.514830493112685,
+    "gpu_batches_per_second": 1.0309957234491434,
+    "producer_headroom": 8.25884171917489,
     "sustainable": true,
-    "buffer_seconds": 60.18768919221483,
+    "buffer_seconds": 60.136039936792535,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 80,
+    "upload_pages_per_commit": 87,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 352449.9817351,
-    "upload_bytes_per_second": 5306492.160540054,
+    "generation_bytes_per_second": 363664.2218677372,
+    "upload_bytes_per_second": 5249834.478697952,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 75.62756146468877,
+    "estimated_full_shards_per_hour": 78.03387634300314,
     "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": 1.3809676959972421,
-      "bytes": 13424902,
-      "pages": 35,
-      "shards": 4
-    },
     {
       "seconds": 0.9754727400031697,
       "bytes": 994247,
@@ -562,6 +556,12 @@
       "bytes": 904800,
       "pages": 27,
       "shards": 1
+    },
+    {
+      "seconds": 1.1419830579980044,
+      "bytes": 7590743,
+      "pages": 31,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790868774355361490-de221e1d/train.jsonl

@@ -396,3 +396,4 @@
 {"step": 479800, "loss": 0.5866878628730774, "lr": 0.0003, "grad_norm": 0.14783031110429193, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 70227140101, "tokens_per_second": 149996.468889314, "data_wait_s": 9.069000952877104e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 479900, "loss": 0.5699813365936279, "lr": 0.0003, "grad_norm": 0.1999884384445087, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 70241720642, "tokens_per_second": 150707.2263292181, "data_wait_s": 6.436996045522392e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 480000, "loss": 0.6134145259857178, "lr": 0.0003, "grad_norm": 0.15646532583825484, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 70256414402, "tokens_per_second": 152872.12864308245, "data_wait_s": 1.1411000741645694e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 480100, "loss": 0.5854529738426208, "lr": 0.0003, "grad_norm": 0.144068183056573, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 70271254957, "tokens_per_second": 134565.61914146633, "data_wait_s": 1.5833000361453742e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1790868774.f92f6aaf-743a-4c24-a54a-850bd75954cb-8js8b.612.0.1790868774355361490-de221e1d

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:9c446fd4985aeeb09ea9e5d662ea42c5d6e217ca6b49ecc83029d9003e5862a9
-size 849393
+oid sha256:9836c1a1909fbf786618d1ceeb1af526556ac6294e342589b55c3bb8145b47eb
+size 851511