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": 76244.45460514116,
-      "required_pages_per_second_upper_estimate": 0.006608676085370123
+      "required_token_positions_per_second": 75711.08150384364,
+      "required_pages_per_second_upper_estimate": 0.006562444656771421
     },
     {
       "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": 9530.556825642645,
-      "required_pages_per_second_upper_estimate": 0.003801881612271679
+      "required_token_positions_per_second": 9463.885187980455,
+      "required_pages_per_second_upper_estimate": 0.0037752852991784166
     },
     {
       "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": 19061.11365128529,
-      "required_pages_per_second_upper_estimate": 0.004409453179082113
+      "required_token_positions_per_second": 18927.77037596091,
+      "required_pages_per_second_upper_estimate": 0.004378606559097311
     },
     {
       "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": 28591.670476927935,
-      "required_pages_per_second_upper_estimate": 0.0017410827231432175
+      "required_token_positions_per_second": 28391.655563941367,
+      "required_pages_per_second_upper_estimate": 0.0017289028643394824
     },
     {
       "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": 28591.670476927935,
-      "required_pages_per_second_upper_estimate": 0.002201065940123414
+      "required_token_positions_per_second": 28391.655563941367,
+      "required_pages_per_second_upper_estimate": 0.0021856682384449916
     },
     {
       "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": 28591.670476927935,
-      "required_pages_per_second_upper_estimate": 0.35501285715792663
+      "required_token_positions_per_second": 28391.655563941367,
+      "required_pages_per_second_upper_estimate": 0.3525293413454855
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,27 +155,27 @@
   "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": 0.1434574083060457,
   "selected": {
-    "prefetch_batches": 63,
+    "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.06971516046093029,
-    "producer_p95_seconds": 0.18884669299950474,
-    "gpu_update_seconds": 0.9669949163099991,
-    "producer_batches_per_second": 14.344082311342008,
-    "gpu_batches_per_second": 1.0341316000046272,
-    "producer_headroom": 13.870654674199905,
+    "producer_batch_seconds": 0.10237993200780693,
+    "producer_p95_seconds": 0.20197022699994704,
+    "gpu_update_seconds": 0.9738072490254551,
+    "producer_batches_per_second": 9.767539208013398,
+    "gpu_batches_per_second": 1.026897264320796,
+    "producer_headroom": 9.5117004859038,
     "sustainable": true,
-    "buffer_seconds": 60.920679727529944,
+    "buffer_seconds": 60.37604943957822,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 87,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 385199.35774904164,
-    "upload_bytes_per_second": 5873996.86382255,
+    "generation_bytes_per_second": 348718.5939356792,
+    "upload_bytes_per_second": 6140663.194117641,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 82.65481519082486,
+    "estimated_full_shards_per_hour": 74.82689250519545,
     "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": [
@@ -256,6 +256,12 @@
       "bytes": 5773373,
       "pages": 27,
       "shards": 2
+    },
+    {
+      "seconds": 1.2743502019993684,
+      "bytes": 11373993,
+      "pages": 42,
+      "shards": 4
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790519810984249790-cb9c9513/train.jsonl

@@ -62,3 +62,4 @@
 {"step": 121200, "loss": 0.5928506460040808, "lr": 0.0003, "grad_norm": 0.1433698208387981, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 17736231330, "tokens_per_second": 150120.66039966655, "data_wait_s": 1.0407999980088789e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 121300, "loss": 0.6239339765161276, "lr": 0.0003, "grad_norm": 0.13210631957350794, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 17750932531, "tokens_per_second": 151137.59909397888, "data_wait_s": 9.966999641619623e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 121400, "loss": 0.6158111244440079, "lr": 0.0003, "grad_norm": 0.13957853017813, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 17765603387, "tokens_per_second": 148839.205693295, "data_wait_s": 9.33699993765913e-06, "gpu_peak_gb": 85.46427917480469}
+{"step": 121500, "loss": 0.5934399608522654, "lr": 0.0003, "grad_norm": 0.14158403337003114, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 17780151113, "tokens_per_second": 148631.56427067416, "data_wait_s": 1.730599979055114e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790519810.cad0b19a-59f2-43f0-8fb4-2a94097843e2-tjh7c.1336.0.1790519810984249790-cb9c9513

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:89d247c0e41baa794a2fb384e87d0ba0aab611f277983e6d0fb065c53ec641a3
-size 136500
+oid sha256:057bb3b9bc4cb1c558261da010980e92534aa2b7ae0f562494b639aff9c790b8
+size 138618