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": 77558.49822163687,
-      "required_pages_per_second_upper_estimate": 0.005365676867482432
+      "required_token_positions_per_second": 76781.04325529202,
+      "required_pages_per_second_upper_estimate": 0.005311890728966639
     },
     {
       "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": 9694.812277704608,
-      "required_pages_per_second_upper_estimate": 0.004059015449973124
+      "required_token_positions_per_second": 9597.630406911503,
+      "required_pages_per_second_upper_estimate": 0.004018327430060283
     },
     {
       "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": 19389.624555409217,
-      "required_pages_per_second_upper_estimate": 0.00403834745610617
+      "required_token_positions_per_second": 19195.260813823006,
+      "required_pages_per_second_upper_estimate": 0.003997866614450317
     },
     {
       "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": 29084.43683311383,
-      "required_pages_per_second_upper_estimate": 0.0017435675787815155
+      "required_token_positions_per_second": 28792.89122073451,
+      "required_pages_per_second_upper_estimate": 0.0017260898644837494
     },
     {
       "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": 29084.43683311383,
-      "required_pages_per_second_upper_estimate": 0.002261187034856916
+      "required_token_positions_per_second": 28792.89122073451,
+      "required_pages_per_second_upper_estimate": 0.0022385206458680473
     },
     {
       "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": 29084.43683311383,
-      "required_pages_per_second_upper_estimate": 0.3730639272599611
+      "required_token_positions_per_second": 28792.89122073451,
+      "required_pages_per_second_upper_estimate": 0.3693242931816486
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,36 +155,30 @@
   "shard_benchmark_scope": "Local packing/read measurements; upload capacity measured only on real data commits. Targets larger than available pages are sample-limited.",
   "estimated_fresh_generation_batch_seconds": 2.558590814491007,
   "selected": {
-    "prefetch_batches": 64,
+    "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 704566,
-    "producer_batch_seconds": 0.1131204803749597,
-    "producer_p95_seconds": 0.24706060900280136,
-    "gpu_update_seconds": 0.950611495716555,
-    "producer_batches_per_second": 8.840132190787262,
-    "gpu_batches_per_second": 1.0519544572162118,
-    "producer_headroom": 8.403531284216346,
+    "producer_batch_seconds": 0.07085522641401099,
+    "producer_p95_seconds": 0.27302171000337694,
+    "gpu_update_seconds": 0.9602370178125758,
+    "producer_batches_per_second": 14.113284941846702,
+    "gpu_batches_per_second": 1.0414095493610571,
+    "producer_headroom": 13.55209864409801,
     "sustainable": true,
-    "buffer_seconds": 60.83913572585952,
+    "buffer_seconds": 60.494932122192274,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 82,
+    "upload_pages_per_commit": 78,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 324945.88880192785,
-    "upload_bytes_per_second": 4937128.21067252,
+    "generation_bytes_per_second": 291637.6241046646,
+    "upload_bytes_per_second": 4947991.2683024155,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 69.72582338374497,
+    "estimated_full_shards_per_hour": 62.57864515643075,
     "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": 0.9514035880001757,
-      "bytes": 6117975,
-      "pages": 37,
-      "shards": 2
-    },
     {
       "seconds": 0.8122407450009632,
       "bytes": 1177705,
@@ -562,6 +556,12 @@
       "bytes": 6085764,
       "pages": 28,
       "shards": 2
+    },
+    {
+      "seconds": 1.1963462499988964,
+      "bytes": 8278514,
+      "pages": 33,
+      "shards": 3
     }
   ],
   "cache_key": {

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

@@ -227,3 +227,4 @@
 {"step": 462900, "loss": 0.5904766917228699, "lr": 0.0003, "grad_norm": 0.18589227614191275, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 67754596057, "tokens_per_second": 154393.3384983321, "data_wait_s": 1.169900133390911e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 463000, "loss": 0.5758638381958008, "lr": 0.0003, "grad_norm": 0.16824982275587075, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 67769284022, "tokens_per_second": 152527.59354923086, "data_wait_s": 1.737300044624135e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 463100, "loss": 0.6063556671142578, "lr": 0.0003, "grad_norm": 0.1748745241501398, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 67783875003, "tokens_per_second": 132804.81875591702, "data_wait_s": 1.2006003089481965e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 463200, "loss": 0.5854924321174622, "lr": 0.0003, "grad_norm": 0.14913088991432086, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 67798604936, "tokens_per_second": 152690.07047614496, "data_wait_s": 1.2446002074284479e-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:e5de9ca1b9af78e9af8457184a9af1d36d58e8ccde82eebbdf97838126c73c6d
-size 488319
+oid sha256:6deee833e79066a1869a2f3a248c384305363d83abb7a264e1db334633aab5d8
+size 490437