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": 75807.15230105139,
-      "required_pages_per_second_upper_estimate": 0.006570771829984179
+      "required_token_positions_per_second": 75916.76969289294,
+      "required_pages_per_second_upper_estimate": 0.006580273187686264
     },
     {
       "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": 9475.894037631424,
-      "required_pages_per_second_upper_estimate": 0.003780075808852491
+      "required_token_positions_per_second": 9489.596211611617,
+      "required_pages_per_second_upper_estimate": 0.0037855418109189473
     },
     {
       "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": 18951.788075262848,
-      "required_pages_per_second_upper_estimate": 0.004384162631133744
+      "required_token_positions_per_second": 18979.192423223234,
+      "required_pages_per_second_upper_estimate": 0.004390502144734327
     },
     {
       "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": 28427.682112894272,
-      "required_pages_per_second_upper_estimate": 0.00173109669215402
+      "required_token_positions_per_second": 28468.78863483485,
+      "required_pages_per_second_upper_estimate": 0.0017335998636709483
     },
     {
       "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": 28427.682112894272,
-      "required_pages_per_second_upper_estimate": 0.002188441661911257
+      "required_token_positions_per_second": 28468.78863483485,
+      "required_pages_per_second_upper_estimate": 0.002191606155760378
     },
     {
       "source": "cortex",
@@ -92,8 +92,8 @@
       "tokens_per_second": 1015177.8395898147,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 22,
-      "required_token_positions_per_second": 28427.682112894272,
-      "required_pages_per_second_upper_estimate": 0.3529766705103775
+      "required_token_positions_per_second": 28468.78863483485,
+      "required_pages_per_second_upper_estimate": 0.35348707593820045
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.10379333235923127,
-    "producer_p95_seconds": 0.3525290469988249,
-    "gpu_update_seconds": 0.9725731380491052,
-    "producer_batches_per_second": 9.634530246499606,
-    "gpu_batches_per_second": 1.0282003079027153,
-    "producer_headroom": 9.370285315467141,
+    "producer_batch_seconds": 0.09779975890620562,
+    "producer_p95_seconds": 0.3297770409990335,
+    "gpu_update_seconds": 0.9711688247307256,
+    "producer_batches_per_second": 10.224974081572585,
+    "gpu_batches_per_second": 1.029687088933552,
+    "producer_headroom": 9.930176061702976,
     "sustainable": true,
-    "buffer_seconds": 60.29953455904452,
+    "buffer_seconds": 60.21246713330499,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 80,
+    "upload_pages_per_commit": 87,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 314395.0017212032,
+    "generation_bytes_per_second": 317544.44149353827,
     "upload_bytes_per_second": 3908129.4436399788,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 67.4618486283023,
+    "estimated_full_shards_per_hour": 68.13764508823978,
     "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": [

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

@@ -249,3 +249,4 @@
 {"step": 139900, "loss": 0.5581844821572304, "lr": 0.0003, "grad_norm": 0.15211728827907198, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 20474365976, "tokens_per_second": 147882.59859125703, "data_wait_s": 1.2952001270605251e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 140000, "loss": 0.6059972457587719, "lr": 0.0003, "grad_norm": 0.12777334099873983, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 20489021311, "tokens_per_second": 149128.76249265246, "data_wait_s": 1.251399953616783e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 140100, "loss": 0.564550356939435, "lr": 0.0003, "grad_norm": 0.13439425516967296, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 20503478898, "tokens_per_second": 137768.5755227633, "data_wait_s": 1.0812000255100429e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 140200, "loss": 0.5690125618129969, "lr": 0.0003, "grad_norm": 0.13399954106283535, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 20518052449, "tokens_per_second": 148214.7831573339, "data_wait_s": 1.2761996913468465e-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:f38304f076f3408643caf3135b7009be7de71f784026dbe85ec431657daca802
-size 535698
+oid sha256:4b394175d8c9bd9e446753369d3e6a37fc9c6fd929501dfa473cb5aa30ec7393
+size 537816