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": 77486.19008884332,
-      "required_pages_per_second_upper_estimate": 0.005360674423077791
+      "required_token_positions_per_second": 77136.47288546959,
+      "required_pages_per_second_upper_estimate": 0.0053364801754926874
     },
     {
       "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": 9685.773761105414,
-      "required_pages_per_second_upper_estimate": 0.004055231211818731
+      "required_token_positions_per_second": 9642.059110683698,
+      "required_pages_per_second_upper_estimate": 0.004036928800552865
     },
     {
       "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": 19371.54752221083,
-      "required_pages_per_second_upper_estimate": 0.004034582486814369
+      "required_token_positions_per_second": 19284.118221367396,
+      "required_pages_per_second_upper_estimate": 0.004016373269114395
     },
     {
       "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": 29057.32128331624,
-      "required_pages_per_second_upper_estimate": 0.0017419420429741971
+      "required_token_positions_per_second": 28926.177332051095,
+      "required_pages_per_second_upper_estimate": 0.0017340801633410707
     },
     {
       "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": 29057.32128331624,
-      "required_pages_per_second_upper_estimate": 0.0022590789201288523
+      "required_token_positions_per_second": 28926.177332051095,
+      "required_pages_per_second_upper_estimate": 0.0022488830547592667
     },
     {
       "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": 29057.32128331624,
-      "required_pages_per_second_upper_estimate": 0.3727161181015667
+      "required_token_positions_per_second": 28926.177332051095,
+      "required_pages_per_second_upper_estimate": 0.37103394430614145
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,36 +155,30 @@
   "shard_benchmark_scope": "Local packing/read measurements; upload capacity measured only on real data commits. Targets larger than available pages are sample-limited.",
   "estimated_fresh_generation_batch_seconds": 2.558590814491007,
   "selected": {
-    "prefetch_batches": 64,
+    "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.09133065037489985,
-    "producer_p95_seconds": 0.22345097900142719,
-    "gpu_update_seconds": 0.951498582075925,
-    "producer_batches_per_second": 10.949226748031867,
-    "gpu_batches_per_second": 1.0509737153977228,
-    "producer_headroom": 10.418173725580111,
+    "producer_batch_seconds": 0.078730627078059,
+    "producer_p95_seconds": 0.2391078960008599,
+    "gpu_update_seconds": 0.9558124353114977,
+    "producer_batches_per_second": 12.701537344654078,
+    "gpu_batches_per_second": 1.0462303722530055,
+    "producer_headroom": 12.14028734159375,
     "sustainable": true,
-    "buffer_seconds": 60.8959092528592,
+    "buffer_seconds": 60.216183424624354,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 80,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 362806.45697154,
-    "upload_bytes_per_second": 5307795.496295736,
+    "generation_bytes_per_second": 378287.7210756799,
+    "upload_bytes_per_second": 5297331.82625389,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 77.84981996402406,
+    "estimated_full_shards_per_hour": 81.17173885538861,
     "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": 1.7602326240003094,
-      "bytes": 10170709,
-      "pages": 35,
-      "shards": 3
-    },
     {
       "seconds": 1.23988030000055,
       "bytes": 8293432,
@@ -562,6 +556,12 @@
       "bytes": 11974676,
       "pages": 44,
       "shards": 4
+    },
+    {
+      "seconds": 1.1600040840003203,
+      "bytes": 6119662,
+      "pages": 36,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791399463488568880-054ba646/train.jsonl

@@ -145,3 +145,4 @@
 {"step": 899500, "loss": 0.5384294986724854, "lr": 3e-05, "grad_norm": 0.3202262936465282, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 131661317551, "tokens_per_second": 152490.0874122081, "data_wait_s": 1.6599000446149148e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 899600, "loss": 0.5393480062484741, "lr": 3e-05, "grad_norm": 0.354624739351512, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 131676024695, "tokens_per_second": 153852.00120140688, "data_wait_s": 1.560600139782764e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 899700, "loss": 0.5507715344429016, "lr": 3e-05, "grad_norm": 0.3216003760680233, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 131690786780, "tokens_per_second": 153708.08054490155, "data_wait_s": 1.6756999684730545e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 899800, "loss": 0.588070809841156, "lr": 3e-05, "grad_norm": 0.340170300486688, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 131705363330, "tokens_per_second": 151823.2719513825, "data_wait_s": 1.5900999642326497e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791399463.60f38c85-66c9-4fe6-8e58-7b3a56dc2cee-wtkw8.346.0.1791399463488568880-054ba646

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:608510b93ee0c435cda1403c57991c616a855ccddcccfebd4fe69be7da6eb349
-size 313077
+oid sha256:088c0f8821cf63bdf35e9f723d9e9035adb4345f38a9c567c24208fa1b1ffc42
+size 315195