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": 76499.76095720792,
-      "required_pages_per_second_upper_estimate": 0.00529243096692033
+      "required_token_positions_per_second": 76656.37107606677,
+      "required_pages_per_second_upper_estimate": 0.005303265618328526
     },
     {
       "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": 9562.47011965099,
-      "required_pages_per_second_upper_estimate": 0.004003606552014596
+      "required_token_positions_per_second": 9582.046384508347,
+      "required_pages_per_second_upper_estimate": 0.004011802725311475
     },
     {
       "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": 19124.94023930198,
-      "required_pages_per_second_upper_estimate": 0.003983220693255846
+      "required_token_positions_per_second": 19164.092769016694,
+      "required_pages_per_second_upper_estimate": 0.003991375132673778
     },
     {
       "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": 28687.41035895297,
-      "required_pages_per_second_upper_estimate": 0.0017197664478798965
+      "required_token_positions_per_second": 28746.13915352504,
+      "required_pages_per_second_upper_estimate": 0.001723287149440815
     },
     {
       "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": 28687.41035895297,
-      "required_pages_per_second_upper_estimate": 0.002230319972825811
+      "required_token_positions_per_second": 28746.13915352504,
+      "required_pages_per_second_upper_estimate": 0.0022348858782831223
     },
     {
       "source": "cortex",
@@ -92,8 +92,8 @@
       "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": 28687.41035895297,
-      "required_pages_per_second_upper_estimate": 0.36797129794324046
+      "required_token_positions_per_second": 28746.13915352504,
+      "required_pages_per_second_upper_estimate": 0.36872460786194433
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,39 +158,27 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 704566,
-    "producer_batch_seconds": 0.08245735881254745,
-    "producer_p95_seconds": 0.19908109500102,
-    "gpu_update_seconds": 0.9637677174081841,
-    "producer_batches_per_second": 12.127480365619363,
-    "gpu_batches_per_second": 1.03759441402463,
-    "producer_headroom": 11.688074069885543,
+    "producer_batch_seconds": 0.06829020395311147,
+    "producer_p95_seconds": 0.17971769600080734,
+    "gpu_update_seconds": 0.9617987254684816,
+    "producer_batches_per_second": 14.643388687001242,
+    "gpu_batches_per_second": 1.039718574707937,
+    "producer_headroom": 14.083992575697376,
     "sustainable": true,
-    "buffer_seconds": 60.7173661967156,
+    "buffer_seconds": 60.593319704514336,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 87,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 322888.0851025507,
-    "upload_bytes_per_second": 4628498.063736584,
+    "generation_bytes_per_second": 322466.1908349695,
+    "upload_bytes_per_second": 4617155.832011391,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 69.28426661307708,
+    "estimated_full_shards_per_hour": 69.19373792445006,
     "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": 1.011093388000063,
-      "bytes": 1199201,
-      "pages": 34,
-      "shards": 1
-    },
-    {
-      "seconds": 1.0240878049999083,
-      "bytes": 7205273,
-      "pages": 28,
-      "shards": 4
-    },
     {
       "seconds": 1.1896957620001558,
       "bytes": 10758587,
@@ -562,6 +550,18 @@
       "bytes": 7565247,
       "pages": 36,
       "shards": 3
+    },
+    {
+      "seconds": 1.035932449000029,
+      "bytes": 5952411,
+      "pages": 30,
+      "shards": 2
+    },
+    {
+      "seconds": 0.7514245589991333,
+      "bytes": 257606,
+      "pages": 8,
+      "shards": 1
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790956870018829698-ff939fa5/train.jsonl

@@ -98,3 +98,4 @@
 {"step": 534800, "loss": 0.5913066267967224, "lr": 0.0003, "grad_norm": 0.16676529798441295, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 78277489012, "tokens_per_second": 152946.81677670413, "data_wait_s": 1.048099875333719e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 534900, "loss": 0.5911183953285217, "lr": 0.0003, "grad_norm": 0.15948313912046233, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 78292146458, "tokens_per_second": 152711.8416193403, "data_wait_s": 1.7186999684781767e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 535000, "loss": 0.6051273345947266, "lr": 0.0003, "grad_norm": 0.16704291770607504, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 78306858501, "tokens_per_second": 153400.40848717812, "data_wait_s": 1.5859999621170573e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 535100, "loss": 0.6086952090263367, "lr": 0.0003, "grad_norm": 0.16169766829009985, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 78321537261, "tokens_per_second": 146518.9366638648, "data_wait_s": 8.11400059319567e-06, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1790956870.abdf5fd7-18c6-4e42-b51e-c994b21d5ba1-dh6p5.3284.0.1790956870018829698-ff939fa5

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:68ca81794d7749e906e4381fd68641e87ee85f5a691b483164c727ae0b7595dd
-size 213531
+oid sha256:3a95905a4d62a4907e66a3ab9bd220a2946b18648db43d2cef995baae7777417
+size 215649