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": 74048.36666452323,
-      "required_pages_per_second_upper_estimate": 0.00641832474861136
+      "required_token_positions_per_second": 75606.79534556897,
+      "required_pages_per_second_upper_estimate": 0.006553405396883024
     },
     {
       "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": 9256.045833065404,
-      "required_pages_per_second_upper_estimate": 0.0036923750730275266
+      "required_token_positions_per_second": 9450.849418196121,
+      "required_pages_per_second_upper_estimate": 0.003770085135709319
     },
     {
       "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": 18512.091666130807,
-      "required_pages_per_second_upper_estimate": 0.004282446605273548
+      "required_token_positions_per_second": 18901.698836392243,
+      "required_pages_per_second_upper_estimate": 0.004372575367261525
     },
     {
       "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": 27768.137499196208,
-      "required_pages_per_second_upper_estimate": 0.0016909338855429648
+      "required_token_positions_per_second": 28352.548254588368,
+      "required_pages_per_second_upper_estimate": 0.0017265214343800796
     },
     {
       "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": 27768.137499196208,
-      "required_pages_per_second_upper_estimate": 0.0021376680918124367
+      "required_token_positions_per_second": 28352.548254588368,
+      "required_pages_per_second_upper_estimate": 0.002182657649515037
     },
     {
       "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": 21,
-      "required_token_positions_per_second": 27768.137499196208,
-      "required_pages_per_second_upper_estimate": 0.34478733376207465
+      "estimated_pages_for_buffer": 22,
+      "required_token_positions_per_second": 28352.548254588368,
+      "required_pages_per_second_upper_estimate": 0.3520437594470662
     }
   ],
   "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": 61,
+    "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.2289762241794051,
-    "producer_p95_seconds": 0.379161805001786,
-    "gpu_update_seconds": 0.9956735485338299,
-    "producer_batches_per_second": 4.367265656439903,
-    "gpu_batches_per_second": 1.0043452509836592,
-    "producer_headroom": 4.348370893537444,
+    "producer_batch_seconds": 0.10743396863267662,
+    "producer_p95_seconds": 0.3628431940014707,
+    "gpu_update_seconds": 0.9751504433301036,
+    "producer_batches_per_second": 9.308043002851935,
+    "gpu_batches_per_second": 1.0254827927730168,
+    "producer_headroom": 9.076742260766734,
     "sustainable": true,
-    "buffer_seconds": 60.73608646056363,
+    "buffer_seconds": 60.45932748646642,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 78,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 458331.2084130285,
-    "upload_bytes_per_second": 6305182.925547497,
+    "generation_bytes_per_second": 526660.7471500915,
+    "upload_bytes_per_second": 6214687.544045871,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 98.34720792096273,
+    "estimated_full_shards_per_hour": 113.00913630368288,
     "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": 1.0524133049984812,
-      "bytes": 12178249,
-      "pages": 40,
-      "shards": 3
-    },
     {
       "seconds": 1.109467931000836,
       "bytes": 5783909,
@@ -562,6 +556,12 @@
       "bytes": 12015881,
       "pages": 35,
       "shards": 4
+    },
+    {
+      "seconds": 1.0830463239981327,
+      "bytes": 6195752,
+      "pages": 36,
+      "shards": 2
     }
   ],
   "cache_key": {

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

@@ -323,3 +323,4 @@
 {"step": 32300, "loss": 0.6117614042013884, "lr": 0.0003, "grad_norm": 0.13833622898780742, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 4727005304, "tokens_per_second": 149149.09986046032, "data_wait_s": 1.377000444335863e-05, "gpu_peak_gb": 85.43514251708984}
 {"step": 32400, "loss": 0.6780863795429468, "lr": 0.0003, "grad_norm": 0.12556539759495675, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 4741489249, "tokens_per_second": 147126.84428959162, "data_wait_s": 9.2410045908764e-06, "gpu_peak_gb": 85.43514251708984}
 {"step": 32500, "loss": 0.6430358495563269, "lr": 0.0003, "grad_norm": 0.13856002775264428, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 4756103473, "tokens_per_second": 149169.28847649784, "data_wait_s": 1.808099477784708e-05, "gpu_peak_gb": 85.43514251708984}
+{"step": 32600, "loss": 0.6330348439514637, "lr": 0.0003, "grad_norm": 0.13050141602012488, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 4770742301, "tokens_per_second": 149874.27247702537, "data_wait_s": 9.467003110330552e-06, "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 sha256:0eacd1d660a73bcdadb651127c3b3945056c3246cd1a63cff8ddc9f7a490751e
-size 687439
+oid sha256:47c7b991e2763d72398ad1582dd0ef7f3699f36debc14a340b7cf2e772dbd693
+size 689557