Update TensorBoard training metrics

benchmark_cache/shard_scheduler.json

@@ -22,8 +22,8 @@
       "tokens_per_second": 1062967.6168812835,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 74881.61806750661,
-      "required_pages_per_second_upper_estimate": 0.0064905488683655865
+      "required_token_positions_per_second": 75794.43379357354,
+      "required_pages_per_second_upper_estimate": 0.006569669421990761
     },
     {
       "source": "rewrite6",
@@ -36,8 +36,8 @@
       "tokens_per_second": 1206672.263377472,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 9360.202258438327,
-      "required_pages_per_second_upper_estimate": 0.0037339246283861205
+      "required_token_positions_per_second": 9474.304224196692,
+      "required_pages_per_second_upper_estimate": 0.003779441608503547
     },
     {
       "source": "openmath",
@@ -50,8 +50,8 @@
       "tokens_per_second": 410042.7581134458,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 18720.404516876653,
-      "required_pages_per_second_upper_estimate": 0.004330636117112641
+      "required_token_positions_per_second": 18948.608448393385,
+      "required_pages_per_second_upper_estimate": 0.004383427080417015
     },
     {
       "source": "ultra_qa",
@@ -64,8 +64,8 @@
       "tokens_per_second": 2163486.5624068812,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28080.606775314984,
-      "required_pages_per_second_upper_estimate": 0.001709961625059006
+      "required_token_positions_per_second": 28422.912672590075,
+      "required_pages_per_second_upper_estimate": 0.0017308062582628136
     },
     {
       "source": "ultra_style",
@@ -78,8 +78,8 @@
       "tokens_per_second": 1546694.2737633225,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28080.606775314984,
-      "required_pages_per_second_upper_estimate": 0.002161722841658377
+      "required_token_positions_per_second": 28422.912672590075,
+      "required_pages_per_second_upper_estimate": 0.002188074497193984
     },
     {
       "source": "cortex",
@@ -91,9 +91,9 @@
       "read_seconds": 0.042780299000014566,
       "tokens_per_second": 1015177.8395898147,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
-      "estimated_pages_for_buffer": 21,
-      "required_token_positions_per_second": 28080.606775314984,
-      "required_pages_per_second_upper_estimate": 0.3486671564040749
+      "estimated_pages_for_buffer": 22,
+      "required_token_positions_per_second": 28422.912672590075,
+      "required_pages_per_second_upper_estimate": 0.35291745002408925
     }
   ],
   "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": 0.1434574083060457,
   "selected": {
-    "prefetch_batches": 61,
+    "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.08709631311688781,
-    "producer_p95_seconds": 0.3280964990008215,
-    "gpu_update_seconds": 0.9845941087108104,
-    "producer_batches_per_second": 11.481542262965226,
-    "gpu_batches_per_second": 1.015646946445131,
-    "producer_headroom": 11.304658871029748,
+    "producer_batch_seconds": 0.10196691082015263,
+    "producer_p95_seconds": 0.3479446619967348,
+    "gpu_update_seconds": 0.9727363384070989,
+    "producer_batches_per_second": 9.80710302937177,
+    "gpu_batches_per_second": 1.0280278021046758,
+    "producer_headroom": 9.539725491172263,
     "sustainable": true,
-    "buffer_seconds": 60.060240631359434,
+    "buffer_seconds": 60.309652981240134,
     "buffer_drain_seconds": null,
     "shard_target_mib": 32,
-    "upload_pages_per_commit": 156,
+    "upload_pages_per_commit": 160,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 609736.3837363385,
-    "upload_bytes_per_second": 5229181.144667102,
+    "generation_bytes_per_second": 614655.9751940828,
+    "upload_bytes_per_second": 5226424.745693594,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 65.4176170066243,
+    "estimated_full_shards_per_hour": 65.94543190892631,
     "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": 1.3641967180010397,
-      "bytes": 7009240,
-      "pages": 38,
-      "shards": 2
-    },
     {
       "seconds": 1.4704428500008362,
       "bytes": 11028707,
@@ -562,6 +556,12 @@
       "bytes": 2492218,
       "pages": 31,
       "shards": 2
+    },
+    {
+      "seconds": 1.2180680759993265,
+      "bytes": 6022567,
+      "pages": 36,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790519810984249790-cb9c9513/train.jsonl

@@ -307,3 +307,4 @@
 {"step": 145700, "loss": 0.5823377333581448, "lr": 0.0003, "grad_norm": 0.14650722565116236, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 21322804932, "tokens_per_second": 147950.65339411545, "data_wait_s": 1.0825999197550118e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 145800, "loss": 0.630335159599781, "lr": 0.0003, "grad_norm": 0.17256094263642674, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 21337331824, "tokens_per_second": 148205.72856089415, "data_wait_s": 1.3564000255428255e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 145900, "loss": 0.6064195018261671, "lr": 0.0003, "grad_norm": 0.12443164934563086, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 21351993301, "tokens_per_second": 150695.9610669303, "data_wait_s": 1.3901000784244388e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 146000, "loss": 0.6049474477767944, "lr": 0.0003, "grad_norm": 0.14668302956766735, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 21366663299, "tokens_per_second": 149947.5760015606, "data_wait_s": 1.5970999811543152e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790519810.cad0b19a-59f2-43f0-8fb4-2a94097843e2-tjh7c.1336.0.1790519810984249790-cb9c9513

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:fb94bb73c971e417fa9b358001d42bc839ed445f31daced3d46b275cd002ab96
-size 659325
+oid sha256:bb6eecaaf83fc47b491b54e26117fe27e02ed8a02fe68527d00f1619edd6c720
+size 661443