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": 77224.70418365693,
-      "required_pages_per_second_upper_estimate": 0.005342584221426106
+      "required_token_positions_per_second": 77075.23450449822,
+      "required_pages_per_second_upper_estimate": 0.005332243562204468
     },
     {
       "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": 9653.088022957116,
-      "required_pages_per_second_upper_estimate": 0.004041546375811867
+      "required_token_positions_per_second": 9634.404313062278,
+      "required_pages_per_second_upper_estimate": 0.004033723896639128
     },
     {
       "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": 19306.17604591423,
-      "required_pages_per_second_upper_estimate": 0.004020967332263549
+      "required_token_positions_per_second": 19268.808626124555,
+      "required_pages_per_second_upper_estimate": 0.004013184684166488
     },
     {
       "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": 28959.264068871344,
-      "required_pages_per_second_upper_estimate": 0.0017360636626929228
+      "required_token_positions_per_second": 28903.212939186833,
+      "required_pages_per_second_upper_estimate": 0.001732703483053456
     },
     {
       "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": 28959.264068871344,
-      "required_pages_per_second_upper_estimate": 0.002251455403013866
+      "required_token_positions_per_second": 28903.212939186833,
+      "required_pages_per_second_upper_estimate": 0.0022470976742237593
     },
     {
       "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": 28959.264068871344,
-      "required_pages_per_second_upper_estimate": 0.3714583454403015
+      "required_token_positions_per_second": 28903.212939186833,
+      "required_pages_per_second_upper_estimate": 0.3707393817317227
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 704566,
-    "producer_batch_seconds": 0.10802687855451154,
-    "producer_p95_seconds": 0.2329447269985394,
-    "gpu_update_seconds": 0.9547203939384344,
-    "producer_batches_per_second": 9.256955429804343,
-    "gpu_batches_per_second": 1.0474270858243397,
-    "producer_headroom": 8.837804134613332,
+    "producer_batch_seconds": 0.08003072656248378,
+    "producer_p95_seconds": 0.2263318920013262,
+    "gpu_update_seconds": 0.9565718544222804,
+    "producer_batches_per_second": 12.49520081789152,
+    "gpu_batches_per_second": 1.0453997735527645,
+    "producer_headroom": 11.952557417749288,
     "sustainable": true,
-    "buffer_seconds": 60.14738481812137,
+    "buffer_seconds": 60.26402682860367,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 287946.3412047003,
-    "upload_bytes_per_second": 5211475.68763986,
+    "generation_bytes_per_second": 558004.3119965938,
+    "upload_bytes_per_second": 5206698.564414214,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 61.78658177476651,
+    "estimated_full_shards_per_hour": 119.73473567889558,
     "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": 0.9222849990001123,
-      "bytes": 2428641,
-      "pages": 31,
-      "shards": 2
-    },
     {
       "seconds": 1.1589227389995358,
       "bytes": 11976819,
@@ -562,6 +556,12 @@
       "bytes": 10605987,
       "pages": 41,
       "shards": 3
+    },
+    {
+      "seconds": 0.9885955910031043,
+      "bytes": 2380823,
+      "pages": 28,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790989041640695387-36510d53/train.jsonl

@@ -202,3 +202,4 @@
 {"step": 577300, "loss": 0.5455593466758728, "lr": 0.0003, "grad_norm": 0.15737021817171568, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 84499154643, "tokens_per_second": 151778.9612393807, "data_wait_s": 9.257997589884326e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 577400, "loss": 0.5949903130531311, "lr": 0.0003, "grad_norm": 0.1855480196299965, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 84513850827, "tokens_per_second": 153547.8710361796, "data_wait_s": 1.4748999092262238e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 577500, "loss": 0.574500560760498, "lr": 0.0003, "grad_norm": 0.1523363075603015, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 84528550416, "tokens_per_second": 152968.316301744, "data_wait_s": 8.859002264216542e-06, "gpu_peak_gb": 89.37899398803711}
+{"step": 577600, "loss": 0.5894696116447449, "lr": 0.0003, "grad_norm": 0.1705478396303194, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 84543370450, "tokens_per_second": 154447.42466988543, "data_wait_s": 1.2088999937986955e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1790989041.fb9d4967-10f3-4fa9-9622-1d5a28f8589b-vh8db.427.0.1790989041640695387-36510d53

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:70f9f82978bdffdc90c38e6480dbe3905e3cae3ee5e92225b1a5de3501a74b86
-size 435369
+oid sha256:5af1a22752ddd32959a8038e65f0984a270c51484c07bd7d42a4f83410b9eee2
+size 437487