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": 77156.97137770653,
-      "required_pages_per_second_upper_estimate": 0.005337898308748695
+      "required_token_positions_per_second": 76529.3376411368,
+      "required_pages_per_second_upper_estimate": 0.005294477150541891
     },
     {
       "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": 9644.621422213317,
-      "required_pages_per_second_upper_estimate": 0.004038001586883167
+      "required_token_positions_per_second": 9566.1672051421,
+      "required_pages_per_second_upper_estimate": 0.004005154444505799
     },
     {
       "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": 19289.242844426633,
-      "required_pages_per_second_upper_estimate": 0.004017440592952236
+      "required_token_positions_per_second": 19132.3344102842,
+      "required_pages_per_second_upper_estimate": 0.00398476070407404
     },
     {
       "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": 28933.86426663995,
-      "required_pages_per_second_upper_estimate": 0.001734540983332403
+      "required_token_positions_per_second": 28698.501615426296,
+      "required_pages_per_second_upper_estimate": 0.0017204313517701043
     },
     {
       "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": 28933.86426663995,
-      "required_pages_per_second_upper_estimate": 0.002249480680112298
+      "required_token_positions_per_second": 28698.501615426296,
+      "required_pages_per_second_upper_estimate": 0.0022311822692313303
     },
     {
       "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": 28933.86426663995,
-      "required_pages_per_second_upper_estimate": 0.37113254404946
+      "required_token_positions_per_second": 28698.501615426296,
+      "required_pages_per_second_upper_estimate": 0.3681135646724169
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 704566,
-    "producer_batch_seconds": 0.09142623818748064,
-    "producer_p95_seconds": 0.14498654900035035,
-    "gpu_update_seconds": 0.9555585021485526,
-    "producer_batches_per_second": 10.937779130203063,
-    "gpu_batches_per_second": 1.0465084008477992,
-    "producer_headroom": 10.451687842488537,
+    "producer_batch_seconds": 0.1306352941640796,
+    "producer_p95_seconds": 0.1467947890000687,
+    "gpu_update_seconds": 0.9633952451767858,
+    "producer_batches_per_second": 7.654899132725856,
+    "gpu_batches_per_second": 1.0379955734746202,
+    "producer_headroom": 7.374693426775991,
     "sustainable": true,
-    "buffer_seconds": 60.200185635358814,
+    "buffer_seconds": 60.693900446137505,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 78,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 500095.11111541954,
-    "upload_bytes_per_second": 4310434.583952965,
+    "generation_bytes_per_second": 477191.60455390014,
+    "upload_bytes_per_second": 4063098.714177061,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 107.3087692269987,
+    "estimated_full_shards_per_hour": 102.39420988524202,
     "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": [
@@ -298,6 +298,12 @@
       "bytes": 12214165,
       "pages": 34,
       "shards": 3
+    },
+    {
+      "seconds": 3.807216140000037,
+      "bytes": 7676406,
+      "pages": 38,
+      "shards": 4
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790868774355361490-de221e1d/train.jsonl

@@ -26,3 +26,4 @@
 {"step": 442800, "loss": 0.5626919269561768, "lr": 0.0003, "grad_norm": 0.16894435588564782, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 64812683395, "tokens_per_second": 153366.86244665916, "data_wait_s": 1.3889999991079094e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 442900, "loss": 0.5983905792236328, "lr": 0.0003, "grad_norm": 0.17057465081739728, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 64827354536, "tokens_per_second": 152355.02874252218, "data_wait_s": 1.4829000065219589e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 443000, "loss": 0.5869654417037964, "lr": 0.0003, "grad_norm": 0.15634538284797767, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 64842105954, "tokens_per_second": 153834.57817651666, "data_wait_s": 7.041999651846709e-06, "gpu_peak_gb": 89.37899398803711}
+{"step": 443100, "loss": 0.6199808120727539, "lr": 0.0003, "grad_norm": 0.14270231877020176, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 64856751537, "tokens_per_second": 153131.35662853392, "data_wait_s": 1.625300001251162e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1790868774.f92f6aaf-743a-4c24-a54a-850bd75954cb-8js8b.612.0.1790868774355361490-de221e1d

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:202a060a944134a81fca27b037b0835706f7931d496eebfd5228ccb844c3f0d5
-size 59469
+oid sha256:64d7a7f8f1437afb92936c9c2c1c6a08cebc1943b0a1b676a1b0cd12f5cdc2a2
+size 61587