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": 76042.23837223394,
-      "required_pages_per_second_upper_estimate": 0.006591148494814083
+      "required_token_positions_per_second": 75618.43685435074,
+      "required_pages_per_second_upper_estimate": 0.006554414453359094
     },
     {
       "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": 9505.279796529243,
-      "required_pages_per_second_upper_estimate": 0.003791798227433079
+      "required_token_positions_per_second": 9452.304606793843,
+      "required_pages_per_second_upper_estimate": 0.0037706656321979588
     },
     {
       "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": 19010.559593058486,
-      "required_pages_per_second_upper_estimate": 0.004397758387432379
+      "required_token_positions_per_second": 18904.609213587686,
+      "required_pages_per_second_upper_estimate": 0.004373248631804801
     },
     {
       "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": 28515.83938958773,
-      "required_pages_per_second_upper_estimate": 0.0017364650077791666
+      "required_token_positions_per_second": 28356.913820381527,
+      "required_pages_per_second_upper_estimate": 0.0017267872744325295
     },
     {
       "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": 28515.83938958773,
-      "required_pages_per_second_upper_estimate": 0.00219522825310608
+      "required_token_positions_per_second": 28356.913820381527,
+      "required_pages_per_second_upper_estimate": 0.002182993722854454
     },
     {
       "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": 28515.83938958773,
-      "required_pages_per_second_upper_estimate": 0.35407128884348477
+      "required_token_positions_per_second": 28356.913820381527,
+      "required_pages_per_second_upper_estimate": 0.35209796516360836
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.13916789164812826,
-    "producer_p95_seconds": 0.3994541499996558,
-    "gpu_update_seconds": 0.9695664091198166,
-    "producer_batches_per_second": 7.185565493284884,
-    "gpu_batches_per_second": 1.0313888668108988,
-    "producer_headroom": 6.966882932819488,
+    "producer_batch_seconds": 0.11992004418715396,
+    "producer_p95_seconds": 0.4132774210011121,
+    "gpu_update_seconds": 0.9750003182690495,
+    "producer_batches_per_second": 8.33888952241665,
+    "gpu_batches_per_second": 1.0256406908413458,
+    "producer_headroom": 8.130419938366677,
     "sustainable": true,
-    "buffer_seconds": 60.11311736542863,
+    "buffer_seconds": 60.45001973268107,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 86,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 378497.24877336825,
-    "upload_bytes_per_second": 3613445.19881164,
+    "generation_bytes_per_second": 340124.9594615568,
+    "upload_bytes_per_second": 3645759.544042115,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 81.21669862175737,
+    "estimated_full_shards_per_hour": 72.98289859662083,
     "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": 1.0830463239981327,
-      "bytes": 6195752,
-      "pages": 36,
-      "shards": 2
-    },
     {
       "seconds": 0.7641328610043274,
       "bytes": 2288843,
@@ -562,6 +556,12 @@
       "bytes": 2603030,
       "pages": 38,
       "shards": 2
+    },
+    {
+      "seconds": 1.2956993760017212,
+      "bytes": 10664503,
+      "pages": 31,
+      "shards": 3
     }
   ],
   "cache_key": {

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

@@ -401,3 +401,4 @@
 {"step": 40100, "loss": 0.6127707604318857, "lr": 0.0003, "grad_norm": 0.14291243084088415, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 5868007171, "tokens_per_second": 134005.07927723238, "data_wait_s": 1.1392003216315061e-05, "gpu_peak_gb": 85.43514251708984}
 {"step": 40200, "loss": 0.6159282121807337, "lr": 0.0003, "grad_norm": 0.12916153736250993, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 5882491430, "tokens_per_second": 148199.0223179503, "data_wait_s": 1.5364996215794235e-05, "gpu_peak_gb": 85.43514251708984}
 {"step": 40300, "loss": 0.6500087641179562, "lr": 0.0003, "grad_norm": 0.14165207655343362, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 5897110808, "tokens_per_second": 149087.85499604337, "data_wait_s": 9.56900476012379e-06, "gpu_peak_gb": 85.43514251708984}
+{"step": 40400, "loss": 0.6147771067917347, "lr": 0.0003, "grad_norm": 0.1216657546652096, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 5911786558, "tokens_per_second": 140943.46984423554, "data_wait_s": 1.259699638467282e-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:b59b7e527928bc67146eb43194362b0e08f108d81eb1672949155aef91e088fc
-size 854209
+oid sha256:352dc9093d1eae5c09558d099f659b0f047faaa88de2cc8d47bdcc4a0e4c632b
+size 856327