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": 75157.17042322725,
-      "required_pages_per_second_upper_estimate": 0.00651443304817839
+      "required_token_positions_per_second": 75845.33990980581,
+      "required_pages_per_second_upper_estimate": 0.006574081835125403
     },
     {
       "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": 9394.646302903406,
-      "required_pages_per_second_upper_estimate": 0.0037476648727075975
+      "required_token_positions_per_second": 9480.667488725727,
+      "required_pages_per_second_upper_estimate": 0.003781980009863462
     },
     {
       "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": 18789.29260580681,
-      "required_pages_per_second_upper_estimate": 0.004346572164097676
+      "required_token_positions_per_second": 18961.334977451454,
+      "required_pages_per_second_upper_estimate": 0.004386371138935337
     },
     {
       "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": 28183.938908710217,
-      "required_pages_per_second_upper_estimate": 0.0017162540098409707
+      "required_token_positions_per_second": 28442.00246617718,
+      "required_pages_per_second_upper_estimate": 0.0017319687265358618
     },
     {
       "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": 28183.938908710217,
-      "required_pages_per_second_upper_estimate": 0.002169677635328795
+      "required_token_positions_per_second": 28442.00246617718,
+      "required_pages_per_second_upper_estimate": 0.00218954408234825
     },
     {
       "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": 28183.938908710217,
-      "required_pages_per_second_upper_estimate": 0.3499501956704399
+      "required_token_positions_per_second": 28442.00246617718,
+      "required_pages_per_second_upper_estimate": 0.3531544813710118
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.3413639609687493,
-    "producer_p95_seconds": 0.3689845970002352,
-    "gpu_update_seconds": 0.9809842438827957,
-    "producer_batches_per_second": 2.929424644482452,
-    "gpu_batches_per_second": 1.0193843644643452,
-    "producer_headroom": 2.8737194198792455,
+    "producer_batch_seconds": 0.45224296575784706,
+    "producer_p95_seconds": 0.36191675800000667,
+    "gpu_update_seconds": 0.9720834541407063,
+    "producer_batches_per_second": 2.211200783022126,
+    "gpu_batches_per_second": 1.028718260495413,
+    "producer_headroom": 2.1494716949587827,
     "sustainable": true,
-    "buffer_seconds": 60.82102312073334,
+    "buffer_seconds": 60.26917415672379,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 82,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 341581.6420395341,
-    "upload_bytes_per_second": 6088898.701746191,
+    "generation_bytes_per_second": 306643.53757303464,
+    "upload_bytes_per_second": 6224430.471288741,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 73.29546876801984,
+    "estimated_full_shards_per_hour": 65.79856486695556,
     "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": 0.808165985999949,
-      "bytes": 1135299,
-      "pages": 34,
-      "shards": 1
-    },
     {
       "seconds": 1.9975283729982038,
       "bytes": 6018894,
@@ -562,6 +556,12 @@
       "bytes": 2436827,
       "pages": 30,
       "shards": 2
+    },
+    {
+      "seconds": 1.0387542439966637,
+      "bytes": 11801776,
+      "pages": 24,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790390190216476996-4d8dc439/train.jsonl

@@ -298,3 +298,4 @@
 {"step": 29800, "loss": 0.6372893694788218, "lr": 0.0003, "grad_norm": 0.13709524730172185, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 4360541926, "tokens_per_second": 148383.01489362915, "data_wait_s": 1.182200139737688e-05, "gpu_peak_gb": 85.43514251708984}
 {"step": 29900, "loss": 0.7031157780438662, "lr": 0.0003, "grad_norm": 0.13538670995417385, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 4375163291, "tokens_per_second": 148461.46273434168, "data_wait_s": 1.149200033978559e-05, "gpu_peak_gb": 85.43514251708984}
 {"step": 30000, "loss": 0.6875663865357637, "lr": 0.0003, "grad_norm": 0.14237463414510249, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 4389888866, "tokens_per_second": 149914.74994791884, "data_wait_s": 9.903000318445265e-06, "gpu_peak_gb": 85.43514251708984}
+{"step": 30100, "loss": 0.6164825251325965, "lr": 0.0003, "grad_norm": 0.13171608209137095, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 4404659166, "tokens_per_second": 127771.60330143584, "data_wait_s": 1.5090001397766173e-05, "gpu_peak_gb": 85.43514251708984}

runs/void-byte-v1/events.out.tfevents.1790390190.c8125a87-1345-4bfd-b42f-0e32ed8b7bb7-6w295.451.0.1790390190216476996-4d8dc439

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:5f35f6bb9f8e5192bf8add50646aae3cd96b6a764979323e903e39d25efc42a7
-size 634489
+oid sha256:7d883ea2c8ee4dbbfb662667229e8f27cac491aababad92656165dab1e753e9c
+size 636607