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": 75716.53229909552,
-      "required_pages_per_second_upper_estimate": 0.006562917117888939
+      "required_token_positions_per_second": 75421.10770076304,
+      "required_pages_per_second_upper_estimate": 0.006537310462452279
     },
     {
       "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": 9464.56653738694,
-      "required_pages_per_second_upper_estimate": 0.0037755570996437448
+      "required_token_positions_per_second": 9427.63846259538,
+      "required_pages_per_second_upper_estimate": 0.0037608259384854714
     },
     {
       "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": 18929.13307477388,
-      "required_pages_per_second_upper_estimate": 0.004378921795485011
+      "required_token_positions_per_second": 18855.27692519076,
+      "required_pages_per_second_upper_estimate": 0.004361836475102795
     },
     {
       "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": 28393.699612160817,
-      "required_pages_per_second_upper_estimate": 0.0017290273361517546
+      "required_token_positions_per_second": 28282.91538778614,
+      "required_pages_per_second_upper_estimate": 0.0017222811581272415
     },
     {
       "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": 28393.699612160817,
-      "required_pages_per_second_upper_estimate": 0.002185825594935213
+      "required_token_positions_per_second": 28282.91538778614,
+      "required_pages_per_second_upper_estimate": 0.002177297118672489
     },
     {
       "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": 28393.699612160817,
-      "required_pages_per_second_upper_estimate": 0.35255472158338175
+      "required_token_positions_per_second": 28282.91538778614,
+      "required_pages_per_second_upper_estimate": 0.3511791522875963
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.11017940758597433,
-    "producer_p95_seconds": 0.19445782999991934,
-    "gpu_update_seconds": 0.9737371451291454,
-    "producer_batches_per_second": 9.076106160941988,
-    "gpu_batches_per_second": 1.0269711954629925,
-    "producer_headroom": 8.837741702044699,
+    "producer_batch_seconds": 0.20023604722653232,
+    "producer_p95_seconds": 0.18046786400009296,
+    "gpu_update_seconds": 0.977551275069036,
+    "producer_batches_per_second": 4.994105775912933,
+    "gpu_batches_per_second": 1.0229642429031445,
+    "producer_headroom": 4.881994469073326,
     "sustainable": true,
-    "buffer_seconds": 60.37170299800702,
+    "buffer_seconds": 60.608179054280235,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 76,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 356339.77332233725,
-    "upload_bytes_per_second": 4152767.039122114,
+    "generation_bytes_per_second": 335083.66164090415,
+    "upload_bytes_per_second": 4204095.744142296,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 76.46222018959607,
+    "estimated_full_shards_per_hour": 71.9011534397158,
     "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": [
@@ -424,6 +424,12 @@
       "bytes": 8005286,
       "pages": 27,
       "shards": 3
+    },
+    {
+      "seconds": 0.9674303089996101,
+      "bytes": 7444249,
+      "pages": 41,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790390190216476996-4d8dc439/train.jsonl

@@ -44,3 +44,4 @@
 {"step": 4400, "loss": 0.8444898128509521, "lr": 0.0003, "grad_norm": 0.19591136392072275, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 643440986, "tokens_per_second": 149340.0880277449, "data_wait_s": 1.2088999937986955e-05, "gpu_peak_gb": 85.43514251708984}
 {"step": 4500, "loss": 0.7699046730995178, "lr": 0.0003, "grad_norm": 0.17142649689894135, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 658135815, "tokens_per_second": 147938.50445870945, "data_wait_s": 1.4480000572802965e-05, "gpu_peak_gb": 85.43514251708984}
 {"step": 4600, "loss": 0.7736689057201147, "lr": 0.0003, "grad_norm": 0.17025524734238284, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 672773231, "tokens_per_second": 149800.9268492934, "data_wait_s": 1.0876000487769488e-05, "gpu_peak_gb": 85.43514251708984}
+{"step": 4700, "loss": 0.803300105035305, "lr": 0.0003, "grad_norm": 0.18988430892715008, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 687479992, "tokens_per_second": 149042.21413900176, "data_wait_s": 1.7197000488522463e-05, "gpu_peak_gb": 85.43514251708984}

runs/void-byte-v1/events.out.tfevents.1790390190.c8125a87-1345-4bfd-b42f-0e32ed8b7bb7-6w295.451.0.1790390190216476996-4d8dc439

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:eafd67c9b5e550d7f1c94233a9a5d0381c08d62764112ba484d234a3fff31dc9
-size 95944
+oid sha256:91885045f89b933a0ffc445aca77bd8c0b9007c0a16694e08668ce785f5fb8bf
+size 98027