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": 77103.56069166814,
-      "required_pages_per_second_upper_estimate": 0.005334203233558693
+      "required_token_positions_per_second": 77250.41187602695,
+      "required_pages_per_second_upper_estimate": 0.005344362739234328
     },
     {
       "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": 9637.945086458518,
-      "required_pages_per_second_upper_estimate": 0.004035206344520377
+      "required_token_positions_per_second": 9656.301484503369,
+      "required_pages_per_second_upper_estimate": 0.004042891785056575
     },
     {
       "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": 19275.890172917036,
-      "required_pages_per_second_upper_estimate": 0.004014659583610414
+      "required_token_positions_per_second": 19312.602969006737,
+      "required_pages_per_second_upper_estimate": 0.0040223058908543585
     },
     {
       "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": 28913.835259375555,
-      "required_pages_per_second_upper_estimate": 0.001733340274929426
+      "required_token_positions_per_second": 28968.904453510106,
+      "required_pages_per_second_upper_estimate": 0.0017366415890320015
     },
     {
       "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": 28913.835259375555,
-      "required_pages_per_second_upper_estimate": 0.0022479235128957838
+      "required_token_positions_per_second": 28968.904453510106,
+      "required_pages_per_second_upper_estimate": 0.002252204900515959
     },
     {
       "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": 23,
-      "required_token_positions_per_second": 28913.835259375555,
-      "required_pages_per_second_upper_estimate": 0.37087563344974483
+      "required_token_positions_per_second": 28968.904453510106,
+      "required_pages_per_second_upper_estimate": 0.37158200194340896
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.0792886286171921,
-    "producer_p95_seconds": 0.19790217399895482,
-    "gpu_update_seconds": 0.9562204305302218,
-    "producer_batches_per_second": 12.612149023639068,
-    "gpu_batches_per_second": 1.0457839720549609,
-    "producer_headroom": 12.059994569295466,
+    "producer_batch_seconds": 0.22206120985160283,
+    "producer_p95_seconds": 0.2133292939997773,
+    "gpu_update_seconds": 0.9544026783743265,
+    "producer_batches_per_second": 4.503262864632105,
+    "gpu_batches_per_second": 1.0477757687178135,
+    "producer_headroom": 4.297926139428523,
     "sustainable": true,
-    "buffer_seconds": 60.24188712340397,
+    "buffer_seconds": 60.127368737582565,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 84,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 468867.36875870574,
-    "upload_bytes_per_second": 3460226.9915787955,
+    "generation_bytes_per_second": 436880.55377648794,
+    "upload_bytes_per_second": 3442951.749367663,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 100.60802266188506,
+    "estimated_full_shards_per_hour": 93.74439678164462,
     "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": 1.4442174710002291,
-      "bytes": 7300139,
-      "pages": 36,
-      "shards": 3
-    },
     {
       "seconds": 1.0513844460001565,
       "bytes": 1119612,
@@ -562,6 +556,12 @@
       "bytes": 6072497,
       "pages": 36,
       "shards": 2
+    },
+    {
+      "seconds": 1.7697906350003905,
+      "bytes": 6241903,
+      "pages": 36,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791463982175382982-d4a3b372/train.jsonl

@@ -127,3 +127,4 @@
 {"step": 937700, "loss": 0.5223715901374817, "lr": 3e-05, "grad_norm": 0.33024077437116944, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 137252705765, "tokens_per_second": 154431.41308355043, "data_wait_s": 1.3341999874683097e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 937800, "loss": 0.5168602466583252, "lr": 3e-05, "grad_norm": 0.33473344251680687, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 137267246678, "tokens_per_second": 152357.613139161, "data_wait_s": 2.5258001187467016e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 937900, "loss": 0.5590638518333435, "lr": 3e-05, "grad_norm": 0.36152371841712266, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 137281887062, "tokens_per_second": 153148.5988647969, "data_wait_s": 1.3499000488081947e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 938000, "loss": 0.5621144771575928, "lr": 3e-05, "grad_norm": 0.33235588155988943, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 137296379882, "tokens_per_second": 151453.7891966275, "data_wait_s": 1.6631000107736327e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791463982.2e076f0c-3fe2-4316-8f17-4e30b2e706e2-b5972.360.0.1791463982175382982-d4a3b372

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:08e7bc33c3a8440c7929bdf1fbaf6cd61601c42fe1663bff8860738a92a1a19b
-size 274953
+oid sha256:e7c742f27f289394622e17ada9ea41ac25774cffbe6f5fc6a329d5c4209eefcb
+size 277071