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": 75598.05647987999,
-      "required_pages_per_second_upper_estimate": 0.006552647934153556
+      "required_token_positions_per_second": 76165.90640865364,
+      "required_pages_per_second_upper_estimate": 0.006601867726777169
     },
     {
       "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": 9449.757059984999,
-      "required_pages_per_second_upper_estimate": 0.0037696493776866917
+      "required_token_positions_per_second": 9520.738301081705,
+      "required_pages_per_second_upper_estimate": 0.0037979648560242957
     },
     {
       "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": 18899.514119969997,
-      "required_pages_per_second_upper_estimate": 0.004372069971566934
+      "required_token_positions_per_second": 19041.47660216341,
+      "required_pages_per_second_upper_estimate": 0.004404910493368026
     },
     {
       "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": 28349.271179954998,
-      "required_pages_per_second_upper_estimate": 0.0017263218777283894
+      "required_token_positions_per_second": 28562.214903245116,
+      "required_pages_per_second_upper_estimate": 0.0017392890332473855
     },
     {
       "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": 28349.271179954998,
-      "required_pages_per_second_upper_estimate": 0.002182405371238237
+      "required_token_positions_per_second": 28562.214903245116,
+      "required_pages_per_second_upper_estimate": 0.0021987983685230637
     },
     {
       "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": 28349.271179954998,
-      "required_pages_per_second_upper_estimate": 0.35200306914778295
+      "required_token_positions_per_second": 28562.214903245116,
+      "required_pages_per_second_upper_estimate": 0.35464711751424954
     }
   ],
   "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": 0.09760435868736295,
-    "producer_p95_seconds": 0.27602307699999074,
-    "gpu_update_seconds": 0.9752631672432254,
-    "producer_batches_per_second": 10.245444091314667,
-    "gpu_batches_per_second": 1.025364264321289,
-    "producer_headroom": 9.992004254308931,
+    "producer_batch_seconds": 0.08015877382023007,
+    "producer_p95_seconds": 0.21611261699945317,
+    "gpu_update_seconds": 0.967992156548712,
+    "producer_batches_per_second": 12.47524072963832,
+    "gpu_batches_per_second": 1.033066221905567,
+    "producer_headroom": 12.075935177346924,
     "sustainable": true,
-    "buffer_seconds": 60.466316369079976,
+    "buffer_seconds": 60.015513706020144,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 82,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 483820.2943805587,
+    "generation_bytes_per_second": 502623.33866337617,
     "upload_bytes_per_second": 5205137.642691322,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 103.81657241404125,
+    "estimated_full_shards_per_hour": 107.85126800466503,
     "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-v1/1790563279934888633-82973089/train.jsonl

@@ -140,3 +140,4 @@
 {"step": 169000, "loss": 0.6521192621439695, "lr": 0.0003, "grad_norm": 0.16166802150130857, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 24733838653, "tokens_per_second": 149149.88780075763, "data_wait_s": 1.673299993854016e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 169100, "loss": 0.6279681622982025, "lr": 0.0003, "grad_norm": 0.12974266857108036, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 24748531140, "tokens_per_second": 150551.60547993824, "data_wait_s": 1.5487999917240813e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 169200, "loss": 0.5250497795641422, "lr": 0.0003, "grad_norm": 0.1483682590171452, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 24763144161, "tokens_per_second": 150077.83161036504, "data_wait_s": 8.96399978955742e-06, "gpu_peak_gb": 85.46427917480469}
+{"step": 169300, "loss": 0.6024130657315254, "lr": 0.0003, "grad_norm": 0.15711487218539658, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 24777769244, "tokens_per_second": 149740.9886463385, "data_wait_s": 1.4848999853711575e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790563279.bdfa290c-aca9-4e96-a5d2-8425921a3278-d85hq.1304.0.1790563279934888633-82973089

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:db5e44740b078520edccd0e9c1f563fe3e90ddf5f2c196a3ed475d626f46bf1e
-size 302487
+oid sha256:d17a75ea74ec522fdd1ae838a3b61e9e45421b23980a064c8b2029fa619b70a1
+size 304605