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": 77063.78524909608,
-      "required_pages_per_second_upper_estimate": 0.005331451476149816
+      "required_token_positions_per_second": 77072.01517855343,
+      "required_pages_per_second_upper_estimate": 0.005332020841765742
     },
     {
       "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": 9632.97315613701,
-      "required_pages_per_second_upper_estimate": 0.004033124701120473
+      "required_token_positions_per_second": 9634.001897319178,
+      "required_pages_per_second_upper_estimate": 0.004033555413570888
     },
     {
       "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": 19265.94631227402,
-      "required_pages_per_second_upper_estimate": 0.004012588539675714
+      "required_token_positions_per_second": 19268.003794638356,
+      "required_pages_per_second_upper_estimate": 0.004013017058992746
     },
     {
       "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": 28898.919468411026,
-      "required_pages_per_second_upper_estimate": 0.0017324460960361965
+      "required_token_positions_per_second": 28902.005691957533,
+      "required_pages_per_second_upper_estimate": 0.0017326311104254191
     },
     {
       "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": 28898.919468411026,
-      "required_pages_per_second_upper_estimate": 0.0022467638757559224
+      "required_token_positions_per_second": 28902.005691957533,
+      "required_pages_per_second_upper_estimate": 0.0022470038160618006
     },
     {
       "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": 28898.919468411026,
-      "required_pages_per_second_upper_estimate": 0.3706843096985804
+      "required_token_positions_per_second": 28902.005691957533,
+      "required_pages_per_second_upper_estimate": 0.3707238964605063
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.49434742607823523,
-    "producer_p95_seconds": 0.20334100300169666,
-    "gpu_update_seconds": 0.9567139709227402,
-    "producer_batches_per_second": 2.0228688312048386,
-    "gpu_batches_per_second": 1.0452444830877832,
-    "producer_headroom": 1.9353068721578235,
+    "producer_batch_seconds": 0.07631615482817722,
+    "producer_p95_seconds": 0.19184492099884665,
+    "gpu_update_seconds": 0.9566118107745553,
+    "producer_batches_per_second": 13.10338554466561,
+    "gpu_batches_per_second": 1.0453561086500844,
+    "producer_headroom": 12.534853373159702,
     "sustainable": true,
-    "buffer_seconds": 60.272980168132634,
+    "buffer_seconds": 60.26654407879698,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 271635.0122302083,
-    "upload_bytes_per_second": 5494465.20072179,
+    "generation_bytes_per_second": 270952.11701898585,
+    "upload_bytes_per_second": 5554313.705685166,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 58.28655028514564,
+    "estimated_full_shards_per_hour": 58.14001686980421,
     "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": 0.9738086980005392,
-      "bytes": 6055828,
-      "pages": 36,
-      "shards": 2
-    },
     {
       "seconds": 1.397671879999507,
       "bytes": 6213114,
@@ -562,6 +556,12 @@
       "bytes": 5947734,
       "pages": 28,
       "shards": 2
+    },
+    {
+      "seconds": 1.2793344790006813,
+      "bytes": 12370460,
+      "pages": 40,
+      "shards": 4
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791039222312944374-63c10d34/train.jsonl

@@ -115,3 +115,4 @@
 {"step": 615900, "loss": 0.5587576627731323, "lr": 0.0001, "grad_norm": 0.15822508557198198, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 90149703377, "tokens_per_second": 152150.8785774947, "data_wait_s": 1.4863999240333214e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 616000, "loss": 0.5526002645492554, "lr": 0.0001, "grad_norm": 0.1583943054717085, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 90164240745, "tokens_per_second": 151031.268838026, "data_wait_s": 1.1617999916779809e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 616100, "loss": 0.5552738308906555, "lr": 0.0001, "grad_norm": 0.15385304505385142, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 90178856319, "tokens_per_second": 124564.46547175203, "data_wait_s": 1.091599915525876e-05, "gpu_peak_gb": 89.37861251831055}
+{"step": 616200, "loss": 0.5455089211463928, "lr": 0.0001, "grad_norm": 0.1775336915981058, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 90193515615, "tokens_per_second": 153203.51506671327, "data_wait_s": 9.033001333591528e-06, "gpu_peak_gb": 89.37861251831055}

runs/void-byte-v1/events.out.tfevents.1791039222.f2a393d4-e1e4-407b-a6aa-034aa2ada5ac-vwfgg.353.0.1791039222312944374-63c10d34

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:7229d609c89188b8f128fb2f5553effec59bbc42eee8c9b75795708b53d80c62
-size 250320
+oid sha256:365f2166644d3f1936f5c92ba6dd88b22da6c9dcb6440009074b05d5ea752519
+size 252438