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": 77008.88924755574,
-      "required_pages_per_second_upper_estimate": 0.0053276536433869825
+      "required_token_positions_per_second": 76905.14364040564,
+      "required_pages_per_second_upper_estimate": 0.005320476281561365
     },
     {
       "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": 9626.111155944467,
-      "required_pages_per_second_upper_estimate": 0.004030251724934714
+      "required_token_positions_per_second": 9613.142955050705,
+      "required_pages_per_second_upper_estimate": 0.004024822210027325
     },
     {
       "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": 19252.222311888934,
-      "required_pages_per_second_upper_estimate": 0.004009730192321729
+      "required_token_positions_per_second": 19226.28591010141,
+      "required_pages_per_second_upper_estimate": 0.004004328323818299
     },
     {
       "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": 28878.333467833403,
-      "required_pages_per_second_upper_estimate": 0.0017312119967345708
+      "required_token_positions_per_second": 28839.428865152113,
+      "required_pages_per_second_upper_estimate": 0.0017288797252077162
     },
     {
       "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": 28878.333467833403,
-      "required_pages_per_second_upper_estimate": 0.0022451634047592596
+      "required_token_positions_per_second": 28839.428865152113,
+      "required_pages_per_second_upper_estimate": 0.0022421387430240514
     },
     {
       "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": 28878.333467833403,
-      "required_pages_per_second_upper_estimate": 0.370420254586696
+      "required_token_positions_per_second": 28839.428865152113,
+      "required_pages_per_second_upper_estimate": 0.3699212281160082
     }
   ],
   "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": 0.10245465230485706,
-    "producer_p95_seconds": 0.3182921309999074,
-    "gpu_update_seconds": 0.9573959671459633,
-    "producer_batches_per_second": 9.760415730312259,
-    "gpu_batches_per_second": 1.044499908414113,
-    "producer_headroom": 9.344582657868978,
+    "producer_batch_seconds": 0.13296261607854376,
+    "producer_p95_seconds": 0.3148557239983347,
+    "gpu_update_seconds": 0.9586875013814241,
+    "producer_batches_per_second": 7.520910986057008,
+    "gpu_batches_per_second": 1.0430927685601894,
+    "producer_headroom": 7.210203361335096,
     "sustainable": true,
-    "buffer_seconds": 60.31594593019569,
+    "buffer_seconds": 60.39731258702972,
     "buffer_drain_seconds": null,
-    "shard_target_mib": 16,
-    "upload_pages_per_commit": 85,
+    "shard_target_mib": 32,
+    "upload_pages_per_commit": 165,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 526637.8420752296,
-    "upload_bytes_per_second": 3837858.195923135,
+    "generation_bytes_per_second": 739001.5516179414,
+    "upload_bytes_per_second": 3816169.204784275,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 113.00422140782038,
+    "estimated_full_shards_per_hour": 79.2862649507698,
     "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": 2.121729968999716,
-      "bytes": 11759509,
-      "pages": 29,
-      "shards": 3
-    },
     {
       "seconds": 1.7052124439978797,
       "bytes": 6177071,
@@ -562,6 +556,12 @@
       "bytes": 987981,
       "pages": 29,
       "shards": 1
+    },
+    {
+      "seconds": 1.6518487119974452,
+      "bytes": 7534957,
+      "pages": 37,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791300814123050567-72be911d/train.jsonl

@@ -342,3 +342,4 @@
 {"step": 834200, "loss": 0.5487772822380066, "lr": 3e-05, "grad_norm": 0.2698189661515686, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 122103315124, "tokens_per_second": 152552.07153830057, "data_wait_s": 1.068100391421467e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 834300, "loss": 0.5587754845619202, "lr": 3e-05, "grad_norm": 0.2948067504570078, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 122118046312, "tokens_per_second": 153036.59667053417, "data_wait_s": 1.3212003977969289e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 834400, "loss": 0.5076045393943787, "lr": 3e-05, "grad_norm": 0.2948557973666099, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 122132533332, "tokens_per_second": 150243.17243182645, "data_wait_s": 1.3503005902748555e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 834500, "loss": 0.5453035235404968, "lr": 3e-05, "grad_norm": 0.3164018617622983, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 122147057444, "tokens_per_second": 149809.1086840898, "data_wait_s": 9.306997526437044e-06, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791300814.0c9d201b-2dbc-40fb-b739-4ee2350a0f54-7b2tx.840.0.1791300814123050567-72be911d

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:5834d3587439062fad86f216481884b911761223fac6df5245536043abd72546
-size 733455
+oid sha256:4ca16c31e6b049df81b3403c450408e2f0c27d5ca105e0fd59aae95ee8afa9ce
+size 735573