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": 76235.28940132579,
-      "required_pages_per_second_upper_estimate": 0.0052741342110258485
+      "required_token_positions_per_second": 75680.7864820368,
+      "required_pages_per_second_upper_estimate": 0.005235772412445401
     },
     {
       "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": 9529.411175165724,
-      "required_pages_per_second_upper_estimate": 0.003989765462307878
+      "required_token_positions_per_second": 9460.0983102546,
+      "required_pages_per_second_upper_estimate": 0.003960745613186801
     },
     {
       "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": 19058.822350331448,
-      "required_pages_per_second_upper_estimate": 0.003969450080629271
+      "required_token_positions_per_second": 18920.1966205092,
+      "required_pages_per_second_upper_estimate": 0.003940577996913635
     },
     {
       "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": 28588.23352549717,
-      "required_pages_per_second_upper_estimate": 0.0017138209481484775
+      "required_token_positions_per_second": 28380.294930763797,
+      "required_pages_per_second_upper_estimate": 0.001701355346898062
     },
     {
       "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": 28588.23352549717,
-      "required_pages_per_second_upper_estimate": 0.00222260941025741
+      "required_token_positions_per_second": 28380.294930763797,
+      "required_pages_per_second_upper_estimate": 0.0022064430991421084
     },
     {
       "source": "cortex",
@@ -91,9 +91,9 @@
       "read_seconds": 0.04061180899998362,
       "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": 28588.23352549717,
-      "required_pages_per_second_upper_estimate": 0.36669916401145664
+      "estimated_pages_for_buffer": 22,
+      "required_token_positions_per_second": 28380.294930763797,
+      "required_pages_per_second_upper_estimate": 0.3640319509852849
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,42 +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": 63,
+    "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.06858718393758068,
-    "producer_p95_seconds": 0.22669146000043838,
-    "gpu_update_seconds": 0.9671111709417584,
-    "producer_batches_per_second": 14.579983352430284,
-    "gpu_batches_per_second": 1.0340072889719751,
-    "producer_headroom": 14.100464772280196,
+    "producer_batch_seconds": 0.12105938257801085,
+    "producer_p95_seconds": 0.25936367900067125,
+    "gpu_update_seconds": 0.9741970641055601,
+    "producer_batches_per_second": 8.260408889460496,
+    "gpu_batches_per_second": 1.0264863617897786,
+    "producer_headroom": 8.047266088423886,
     "sustainable": true,
-    "buffer_seconds": 60.92800376933078,
+    "buffer_seconds": 60.400217974544724,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 84,
+    "upload_pages_per_commit": 87,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 389341.2469068264,
-    "upload_bytes_per_second": 3447999.215230209,
+    "generation_bytes_per_second": 420245.6094063423,
+    "upload_bytes_per_second": 3423798.4880936784,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 83.54356818584054,
+    "estimated_full_shards_per_hour": 90.17492496149733,
     "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": 1.182235641999796,
-      "bytes": 6603515,
-      "pages": 27,
-      "shards": 2
-    },
-    {
-      "seconds": 1.285518773000149,
-      "bytes": 6002809,
-      "pages": 36,
-      "shards": 2
-    },
     {
       "seconds": 1.7619018259993027,
       "bytes": 7674402,
@@ -562,6 +550,18 @@
       "bytes": 9744014,
       "pages": 39,
       "shards": 3
+    },
+    {
+      "seconds": 1.1809931960015092,
+      "bytes": 7571529,
+      "pages": 39,
+      "shards": 3
+    },
+    {
+      "seconds": 0.7855844450023142,
+      "bytes": 302035,
+      "pages": 9,
+      "shards": 1
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791342976705245012-6808dd50/train.jsonl

@@ -173,3 +173,4 @@
 {"step": 859800, "loss": 0.5487866997718811, "lr": 3e-05, "grad_norm": 0.299327830126915, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 125850669999, "tokens_per_second": 151724.1533317882, "data_wait_s": 1.35770023916848e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 859900, "loss": 0.5440846085548401, "lr": 3e-05, "grad_norm": 0.3083464456228531, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 125865178178, "tokens_per_second": 150252.7886356925, "data_wait_s": 8.582999726058915e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 860000, "loss": 0.5167421102523804, "lr": 3e-05, "grad_norm": 0.3017758321480747, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 125879763724, "tokens_per_second": 150859.77013761847, "data_wait_s": 1.5797999367350712e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 860100, "loss": 0.47436317801475525, "lr": 3e-05, "grad_norm": 0.30331582675117236, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 125894241221, "tokens_per_second": 144668.14853932388, "data_wait_s": 1.6124002286233008e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791342976.f74d737f-395e-43a1-ae69-439b0359d002-2qp5c.418.0.1791342976705245012-6808dd50

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:2caee313594381a54a51af75063321025c693cd30401746f4ee1f6797ab2985d
-size 373947
+oid sha256:c8a4f350b3af3aa004b1e7968fa9bc928bd24b5f93e6a70cc83be2864411dabe
+size 376065