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": 77501.71887004354,
-      "required_pages_per_second_upper_estimate": 0.0053617487401929585
+      "required_token_positions_per_second": 77599.81567111977,
+      "required_pages_per_second_upper_estimate": 0.0053685353045073506
     },
     {
       "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": 9687.714858755442,
-      "required_pages_per_second_upper_estimate": 0.00405604390887007
+      "required_token_positions_per_second": 9699.97695888997,
+      "required_pages_per_second_upper_estimate": 0.004061177794134628
     },
     {
       "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": 19375.429717510884,
-      "required_pages_per_second_upper_estimate": 0.004035391045714996
+      "required_token_positions_per_second": 19399.95391777994,
+      "required_pages_per_second_upper_estimate": 0.004040498789884388
     },
     {
       "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": 29063.144576266328,
-      "required_pages_per_second_upper_estimate": 0.0017422911404949015
+      "required_token_positions_per_second": 29099.930876669914,
+      "required_pages_per_second_upper_estimate": 0.0017444964230346682
     },
     {
       "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": 29063.144576266328,
-      "required_pages_per_second_upper_estimate": 0.002259531655541763
+      "required_token_positions_per_second": 29099.930876669914,
+      "required_pages_per_second_upper_estimate": 0.0022623916285923066
     },
     {
       "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": 29063.144576266328,
-      "required_pages_per_second_upper_estimate": 0.37279081305096556
+      "required_token_positions_per_second": 29099.930876669914,
+      "required_pages_per_second_upper_estimate": 0.3732626682144908
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 64,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.13652390259376546,
-    "producer_p95_seconds": 0.21571927000150026,
-    "gpu_update_seconds": 0.9513079332295663,
-    "producer_batches_per_second": 7.3247246892403615,
-    "gpu_batches_per_second": 1.0511843379725958,
-    "producer_headroom": 6.968068705596825,
+    "producer_batch_seconds": 0.05921353935934803,
+    "producer_p95_seconds": 0.19166337600108818,
+    "gpu_update_seconds": 0.9501053496373093,
+    "producer_batches_per_second": 16.888029508442653,
+    "gpu_batches_per_second": 1.0525148609906652,
+    "producer_headroom": 16.0454071808041,
     "sustainable": true,
-    "buffer_seconds": 60.88370772669224,
+    "buffer_seconds": 60.806742376787795,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 82,
+    "upload_pages_per_commit": 84,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 456771.31705552887,
-    "upload_bytes_per_second": 5353847.441508419,
+    "generation_bytes_per_second": 457314.8301943182,
+    "upload_bytes_per_second": 5421550.82847409,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 98.01249154805565,
+    "estimated_full_shards_per_hour": 98.12911681530151,
     "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": 1.5475466269999743,
-      "bytes": 6964239,
-      "pages": 34,
-      "shards": 2
-    },
     {
       "seconds": 1.287959292999858,
       "bytes": 7677212,
@@ -562,6 +556,12 @@
       "bytes": 2669572,
       "pages": 36,
       "shards": 2
+    },
+    {
+      "seconds": 1.1629383399995277,
+      "bytes": 10339749,
+      "pages": 40,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791399463488568880-054ba646/train.jsonl

@@ -94,3 +94,4 @@
 {"step": 894400, "loss": 0.5623685717582703, "lr": 3e-05, "grad_norm": 0.31682533423268094, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 130914980459, "tokens_per_second": 153916.95287541757, "data_wait_s": 1.0222998753306456e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 894500, "loss": 0.5550358891487122, "lr": 3e-05, "grad_norm": 0.32708607573751536, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 130929634621, "tokens_per_second": 153431.50399084494, "data_wait_s": 1.5825999071239494e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 894600, "loss": 0.5219671726226807, "lr": 3e-05, "grad_norm": 0.32487118518920005, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 130944339359, "tokens_per_second": 154100.49231621806, "data_wait_s": 1.5692001397837885e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 894700, "loss": 0.510834276676178, "lr": 3e-05, "grad_norm": 0.3047739946325943, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 130959092199, "tokens_per_second": 153209.63930389134, "data_wait_s": 1.7791999198379926e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791399463.60f38c85-66c9-4fe6-8e58-7b3a56dc2cee-wtkw8.346.0.1791399463488568880-054ba646

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:d608bee8951498abd4bb338f032f3248f26e74c44218b3c09d39bc8a8bf20822
-size 204276
+oid sha256:d04c1527575d4ad6bb258ea3ee2b7f9b6e6869d2195d6c04bf1971ffd6b1ba6a
+size 206394