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": 75629.04029084249,
-      "required_pages_per_second_upper_estimate": 0.006555333532360034
+      "required_token_positions_per_second": 74783.29209985494,
+      "required_pages_per_second_upper_estimate": 0.006482026222694426
     },
     {
       "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": 9453.630036355311,
-      "required_pages_per_second_upper_estimate": 0.0037711943658669663
+      "required_token_positions_per_second": 9347.911512481867,
+      "required_pages_per_second_upper_estimate": 0.003729021666061061
     },
     {
       "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": 18907.260072710622,
-      "required_pages_per_second_upper_estimate": 0.0043738618614093654
+      "required_token_positions_per_second": 18695.823024963734,
+      "required_pages_per_second_upper_estimate": 0.00432494962157807
     },
     {
       "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": 28360.89010906593,
-      "required_pages_per_second_upper_estimate": 0.0017270294095514352
+      "required_token_positions_per_second": 28043.7345374456,
+      "required_pages_per_second_upper_estimate": 0.0017077162992264418
     },
     {
       "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": 28360.89010906593,
-      "required_pages_per_second_upper_estimate": 0.0021832998285644506
+      "required_token_positions_per_second": 28043.7345374456,
+      "required_pages_per_second_upper_estimate": 0.0021588843147183046
     },
     {
       "source": "cortex",
@@ -91,9 +91,9 @@
       "read_seconds": 0.042780299000014566,
       "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": 28360.89010906593,
-      "required_pages_per_second_upper_estimate": 0.3521473373612865
+      "estimated_pages_for_buffer": 21,
+      "required_token_positions_per_second": 28043.7345374456,
+      "required_pages_per_second_upper_estimate": 0.34820932661317905
     }
   ],
   "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": 0.1434574083060457,
   "selected": {
-    "prefetch_batches": 62,
+    "prefetch_batches": 61,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 690670,
-    "producer_batch_seconds": 0.10504797805458566,
-    "producer_p95_seconds": 0.2003864770013024,
-    "gpu_update_seconds": 0.9748636200653643,
-    "producer_batches_per_second": 9.519459760380862,
-    "gpu_batches_per_second": 1.025784509153137,
-    "producer_headroom": 9.280175003071452,
+    "producer_batch_seconds": 0.16416728527347857,
+    "producer_p95_seconds": 0.1990985020001972,
+    "gpu_update_seconds": 0.9858886648310982,
+    "producer_batches_per_second": 6.091347605183012,
+    "gpu_batches_per_second": 1.0143133151564527,
+    "producer_headroom": 6.0053905574959865,
     "sustainable": true,
-    "buffer_seconds": 60.44154444405259,
+    "buffer_seconds": 60.139208554696985,
     "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": 412976.89478348073,
-    "upload_bytes_per_second": 3639552.470398413,
+    "generation_bytes_per_second": 406728.85896857415,
+    "upload_bytes_per_second": 3620660.788775344,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 88.61522801044767,
+    "estimated_full_shards_per_hour": 87.27454497139853,
     "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": 1.191119417000209,
-      "bytes": 6703440,
-      "pages": 36,
-      "shards": 3
-    },
     {
       "seconds": 1.0874789810000038,
       "bytes": 6434937,
@@ -562,6 +556,12 @@
       "bytes": 7496810,
       "pages": 38,
       "shards": 3
+    },
+    {
+      "seconds": 1.5263510319982743,
+      "bytes": 5728692,
+      "pages": 27,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790651964532391049-1e9227c4/train.jsonl

@@ -110,3 +110,4 @@
 {"step": 248100, "loss": 0.5804008096456528, "lr": 0.0003, "grad_norm": 0.14121031873210405, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 36315310179, "tokens_per_second": 149589.7096732957, "data_wait_s": 1.3102000593789853e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 248200, "loss": 0.6161615438759327, "lr": 0.0003, "grad_norm": 0.152028095916469, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 36330003488, "tokens_per_second": 149897.65148101526, "data_wait_s": 1.2304999472689815e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 248300, "loss": 0.6322708986699581, "lr": 0.0003, "grad_norm": 0.14915401360010824, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 36344678678, "tokens_per_second": 150042.549345468, "data_wait_s": 1.5436000467161648e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 248400, "loss": 0.6109188199043274, "lr": 0.0003, "grad_norm": 0.13899187421132125, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 36359395437, "tokens_per_second": 149853.73651365758, "data_wait_s": 1.3668999599758536e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790651964.96f3217e-af24-484e-a73b-abea9eb37bd2-c4mbg.442.0.1790651964532391049-1e9227c4

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:e292485c7532b9de84f24aea12e68dd776c546a98aa0b0af97f83964456ada7f
-size 238947
+oid sha256:c6c3b9ad0d92746f5cf926610b2e00464f407c5eab37806b9fb388dff2e3dc01
+size 241065