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": 76585.1676045291,
-      "required_pages_per_second_upper_estimate": 0.005298339596952733
+      "required_token_positions_per_second": 76659.91354249345,
+      "required_pages_per_second_upper_estimate": 0.005303510694375578
     },
     {
       "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": 9573.145950566137,
-      "required_pages_per_second_upper_estimate": 0.004008076299482068
+      "required_token_positions_per_second": 9582.489192811681,
+      "required_pages_per_second_upper_estimate": 0.0040119881199011925
     },
     {
       "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": 19146.291901132274,
-      "required_pages_per_second_upper_estimate": 0.003987667681333908
+      "required_token_positions_per_second": 19164.978385623363,
+      "required_pages_per_second_upper_estimate": 0.0039915595832576665
     },
     {
       "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": 28719.43785169841,
-      "required_pages_per_second_upper_estimate": 0.0017216864471668357
+      "required_token_positions_per_second": 28747.467578435047,
+      "required_pages_per_second_upper_estimate": 0.0017233667864857826
     },
     {
       "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": 28719.43785169841,
-      "required_pages_per_second_upper_estimate": 0.0022328099695127163
+      "required_token_positions_per_second": 28747.467578435047,
+      "required_pages_per_second_upper_estimate": 0.0022349891574767514
     },
     {
       "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": 28719.43785169841,
-      "required_pages_per_second_upper_estimate": 0.3683821122317365
+      "required_token_positions_per_second": 28747.467578435047,
+      "required_pages_per_second_upper_estimate": 0.36874164747033833
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 704566,
-    "producer_batch_seconds": 0.06954895614057932,
-    "producer_p95_seconds": 0.15909554599966214,
-    "gpu_update_seconds": 0.9626929378899717,
-    "producer_batches_per_second": 14.378361020670098,
-    "gpu_batches_per_second": 1.0387528158166381,
-    "producer_headroom": 13.84194661303155,
+    "producer_batch_seconds": 0.04720387348437072,
+    "producer_p95_seconds": 0.1600050300003204,
+    "gpu_update_seconds": 0.9617542806010567,
+    "producer_batches_per_second": 21.18470214803668,
+    "gpu_batches_per_second": 1.0397666224839064,
+    "producer_headroom": 20.374477974132677,
     "sustainable": true,
-    "buffer_seconds": 60.64965508706822,
+    "buffer_seconds": 60.590519677866574,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 506516.5493044301,
-    "upload_bytes_per_second": 3753733.724657172,
+    "generation_bytes_per_second": 506877.3010835644,
+    "upload_bytes_per_second": 3710302.153971347,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 108.68666037892987,
+    "estimated_full_shards_per_hour": 108.76406931286049,
     "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 +382,12 @@
       "bytes": 8046612,
       "pages": 38,
       "shards": 3
+    },
+    {
+      "seconds": 1.0096394150004926,
+      "bytes": 1048802,
+      "pages": 30,
+      "shards": 1
     }
   ],
   "cache_key": {

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

@@ -43,3 +43,4 @@
 {"step": 444500, "loss": 0.58826744556427, "lr": 0.0003, "grad_norm": 0.17988697609140758, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 65061485024, "tokens_per_second": 152900.03011147288, "data_wait_s": 1.3471999409375712e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 444600, "loss": 0.5913227796554565, "lr": 0.0003, "grad_norm": 0.1416641029051212, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 65076195746, "tokens_per_second": 152711.02291267272, "data_wait_s": 1.5451999388460536e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 444700, "loss": 0.5960000157356262, "lr": 0.0003, "grad_norm": 0.1548911055979621, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 65090874113, "tokens_per_second": 152302.72494620606, "data_wait_s": 1.0871999620576389e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 444800, "loss": 0.5636985301971436, "lr": 0.0003, "grad_norm": 0.16352203214652344, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 65105497414, "tokens_per_second": 151367.44674651648, "data_wait_s": 1.705900012893835e-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:d760ac61f1f65b52df20c9b7998a39bab7baf09c7ca2bc9cd4fbca63fe61f1c6
-size 95475
+oid sha256:cd1a1c12cf492c6a2b3e50ad1ce9b350ccecb5cafcf122b9563b2a7ea55c135a
+size 97593