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": 75564.59082873639,
-      "required_pages_per_second_upper_estimate": 0.006549747216330345
+      "required_token_positions_per_second": 75266.23897071424,
+      "required_pages_per_second_upper_estimate": 0.006523886833442857
     },
     {
       "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": 9445.573853592048,
-      "required_pages_per_second_upper_estimate": 0.0037679806341120343
+      "required_token_positions_per_second": 9408.27987133928,
+      "required_pages_per_second_upper_estimate": 0.003753103506996681
     },
     {
       "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": 18891.147707184096,
-      "required_pages_per_second_upper_estimate": 0.004370134549212752
+      "required_token_positions_per_second": 18816.55974267856,
+      "required_pages_per_second_upper_estimate": 0.004352879936327737
     },
     {
       "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": 28336.721560776143,
-      "required_pages_per_second_upper_estimate": 0.0017255576717631602
+      "required_token_positions_per_second": 28224.839614017837,
+      "required_pages_per_second_upper_estimate": 0.0017187446482047884
     },
     {
       "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": 28336.721560776143,
-      "required_pages_per_second_upper_estimate": 0.0021814392668142793
+      "required_token_positions_per_second": 28224.839614017837,
+      "required_pages_per_second_upper_estimate": 0.002172826284843774
     },
     {
       "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": 28336.721560776143,
-      "required_pages_per_second_upper_estimate": 0.35184724487845515
+      "required_token_positions_per_second": 28224.839614017837,
+      "required_pages_per_second_upper_estimate": 0.3504580455445055
     }
   ],
   "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": 0.0929971171327395,
+    "producer_batch_seconds": 0.09741216246845852,
     "producer_p95_seconds": 0.28648294899903703,
-    "gpu_update_seconds": 0.9756950866987565,
-    "producer_batches_per_second": 10.753021500361665,
-    "gpu_batches_per_second": 1.0249103573776093,
-    "producer_headroom": 10.491670245068967,
+    "gpu_update_seconds": 0.9795626964791909,
+    "producer_batches_per_second": 10.265658565210417,
+    "gpu_batches_per_second": 1.0208637013171962,
+    "producer_headroom": 10.05585618527222,
     "sustainable": true,
-    "buffer_seconds": 60.4930953753229,
+    "buffer_seconds": 60.732887181709835,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 79,
+    "upload_pages_per_commit": 81,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 309344.6866023153,
-    "upload_bytes_per_second": 6001554.657417001,
+    "generation_bytes_per_second": 337541.16306008596,
+    "upload_bytes_per_second": 5912903.436221193,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 66.37816856910796,
+    "estimated_full_shards_per_hour": 72.42847603656706,
     "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": 1.1597189560015977,
-      "bytes": 13015122,
-      "pages": 33,
-      "shards": 4
-    },
     {
       "seconds": 1.0076978060023976,
       "bytes": 5410849,
@@ -562,6 +556,12 @@
       "bytes": 948458,
       "pages": 26,
       "shards": 1
+    },
+    {
+      "seconds": 1.2729225670009328,
+      "bytes": 7447736,
+      "pages": 37,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790485060735632541-7579de36/train.jsonl

@@ -164,3 +164,4 @@
 {"step": 101400, "loss": 0.597405731678009, "lr": 0.0003, "grad_norm": 0.16215826759214721, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 14838427351, "tokens_per_second": 147424.06715320845, "data_wait_s": 1.7465998098487034e-05, "gpu_peak_gb": 85.46991729736328}
 {"step": 101500, "loss": 0.6316697914153337, "lr": 0.0003, "grad_norm": 0.14569175651799, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 14853105249, "tokens_per_second": 148596.53276208186, "data_wait_s": 1.0170002497034147e-05, "gpu_peak_gb": 85.46991729736328}
 {"step": 101600, "loss": 0.5911474358290434, "lr": 0.0003, "grad_norm": 0.14319383396650515, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 14867664916, "tokens_per_second": 147244.51369991162, "data_wait_s": 1.5435998648172244e-05, "gpu_peak_gb": 85.46991729736328}
+{"step": 101700, "loss": 0.5779572483152151, "lr": 0.0003, "grad_norm": 0.14472449275624752, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 14882299425, "tokens_per_second": 147313.65997996804, "data_wait_s": 1.9014001736650243e-05, "gpu_peak_gb": 85.46991729736328}

runs/void-byte-v1/events.out.tfevents.1790485060.0b09bbb2-f535-42d2-bd00-cbae7fb94f51-r2qrq.28895.0.1790485060735632541-7579de36

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:9bf67b84ad071259188217e2389673226f37ba636a70dcbb13d1fc6579f36ed5
-size 354102
+oid sha256:3cb6e16ad161539a9ebd2303f440965c673b809aadce01f407c1ffd61e7b8b44
+size 356220