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": 76995.42557316499,
-      "required_pages_per_second_upper_estimate": 0.005326722195152601
+      "required_token_positions_per_second": 76862.35480694346,
+      "required_pages_per_second_upper_estimate": 0.005317516050778679
     },
     {
       "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": 9624.428196645624,
-      "required_pages_per_second_upper_estimate": 0.004029547105020475
+      "required_token_positions_per_second": 9607.794350867933,
+      "required_pages_per_second_upper_estimate": 0.0040225828611475545
     },
     {
       "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": 19248.856393291248,
-      "required_pages_per_second_upper_estimate": 0.004009029160243074
+      "required_token_positions_per_second": 19215.588701735865,
+      "required_pages_per_second_upper_estimate": 0.004002100377420111
     },
     {
       "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": 28873.284589936873,
-      "required_pages_per_second_upper_estimate": 0.0017309093242138642
+      "required_token_positions_per_second": 28823.383052603796,
+      "required_pages_per_second_upper_estimate": 0.0017279178032459817
     },
     {
       "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": 28873.284589936873,
-      "required_pages_per_second_upper_estimate": 0.0022447708767104712
+      "required_token_positions_per_second": 28823.383052603796,
+      "required_pages_per_second_upper_estimate": 0.0022408912516764905
     },
     {
       "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": 28873.284589936873,
-      "required_pages_per_second_upper_estimate": 0.37035549300210197
+      "required_token_positions_per_second": 28823.383052603796,
+      "required_pages_per_second_upper_estimate": 0.3697154096612896
     }
   ],
   "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": 0.10759108114817195,
-    "producer_p95_seconds": 0.2604964589991141,
-    "gpu_update_seconds": 0.9575633805665491,
-    "producer_batches_per_second": 9.29445070472731,
-    "gpu_batches_per_second": 1.0443172956429714,
-    "producer_headroom": 8.900025637327827,
+    "producer_batch_seconds": 0.27122722857023973,
+    "producer_p95_seconds": 0.26114974099982646,
+    "gpu_update_seconds": 0.9592211972321681,
+    "producer_batches_per_second": 3.686945463666934,
+    "gpu_batches_per_second": 1.042512407863274,
+    "producer_headroom": 3.536596241788308,
     "sustainable": true,
-    "buffer_seconds": 60.326492975692595,
+    "buffer_seconds": 60.43093542562659,
     "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": 116686.85495183895,
-    "upload_bytes_per_second": 4702536.762644612,
+    "generation_bytes_per_second": 119965.27414969122,
+    "upload_bytes_per_second": 4687622.147728078,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 25.038282741702808,
+    "estimated_full_shards_per_hour": 25.741755183868907,
     "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": 1.14655600399783,
-      "bytes": 10568739,
-      "pages": 40,
-      "shards": 3
-    },
     {
       "seconds": 0.9451345529996615,
       "bytes": 2348924,
@@ -562,6 +556,12 @@
       "bytes": 5644531,
       "pages": 20,
       "shards": 2
+    },
+    {
+      "seconds": 1.038082987000962,
+      "bytes": 8819669,
+      "pages": 48,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791143623306220418-eb05a537/train.jsonl

@@ -273,3 +273,4 @@
 {"step": 712300, "loss": 0.5719985365867615, "lr": 8e-05, "grad_norm": 0.22361118293217075, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 104260546465, "tokens_per_second": 152077.1954910395, "data_wait_s": 1.4656998246209696e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 712400, "loss": 0.5470711588859558, "lr": 8e-05, "grad_norm": 0.21405512876181312, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 104275265509, "tokens_per_second": 129775.96559413782, "data_wait_s": 1.205899752676487e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 712500, "loss": 0.52162104845047, "lr": 8e-05, "grad_norm": 0.22757465109347105, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 104289867153, "tokens_per_second": 151846.83034889368, "data_wait_s": 1.735600017127581e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 712600, "loss": 0.5429339408874512, "lr": 8e-05, "grad_norm": 0.22336768670284796, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 104304513287, "tokens_per_second": 151675.49010240284, "data_wait_s": 1.801699909265153e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791143623.3eb6ddb3-ffce-42e5-bd48-c0d653189e98-8pb7d.369.0.1791143623306220418-eb05a537

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:549775449422aee1c830a8f489111ff12238a3384f7c009cab72741e495d9031
-size 586530
+oid sha256:4c6d86f42ce0954c6965e34439e9ee29ee958afee0647213a9b07697c684569f
+size 588648