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": 75779.02099599062,
-      "required_pages_per_second_upper_estimate": 0.006568333479759652
+      "required_token_positions_per_second": 75320.50188163084,
+      "required_pages_per_second_upper_estimate": 0.006528590205033017
     },
     {
       "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": 9472.377624498828,
-      "required_pages_per_second_upper_estimate": 0.0037786730590788366
+      "required_token_positions_per_second": 9415.062735203855,
+      "required_pages_per_second_upper_estimate": 0.0037558092928051123
     },
     {
       "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": 18944.755248997655,
-      "required_pages_per_second_upper_estimate": 0.004382535710550806
+      "required_token_positions_per_second": 18830.12547040771,
+      "required_pages_per_second_upper_estimate": 0.004356018128689222
     },
     {
       "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": 28417.132873496485,
-      "required_pages_per_second_upper_estimate": 0.0017304542988222764
+      "required_token_positions_per_second": 28245.188205611565,
+      "required_pages_per_second_upper_estimate": 0.0017199837706720373
     },
     {
       "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": 28417.132873496485,
-      "required_pages_per_second_upper_estimate": 0.0021876295522602514
+      "required_token_positions_per_second": 28245.188205611565,
+      "required_pages_per_second_upper_estimate": 0.002174392775753166
     },
     {
       "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": 28417.132873496485,
-      "required_pages_per_second_upper_estimate": 0.35284568426309004
+      "required_token_positions_per_second": 28245.188205611565,
+      "required_pages_per_second_upper_estimate": 0.3507107069497444
     }
   ],
   "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": 0.09406349160936145,
-    "producer_p95_seconds": 0.22384833799878834,
-    "gpu_update_seconds": 0.9729341845667399,
-    "producer_batches_per_second": 10.63111716236225,
-    "gpu_batches_per_second": 1.027818752658293,
-    "producer_headroom": 10.34337730739639,
+    "producer_batch_seconds": 0.08765065230464586,
+    "producer_p95_seconds": 0.2393573249992187,
+    "gpu_update_seconds": 0.9788569932243213,
+    "producer_batches_per_second": 11.408928213384165,
+    "gpu_batches_per_second": 1.0215996891497239,
+    "producer_headroom": 11.167709166865352,
     "sustainable": true,
-    "buffer_seconds": 60.32191944313787,
+    "buffer_seconds": 60.68913357990792,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 297259.3497845693,
-    "upload_bytes_per_second": 5917001.810635777,
+    "generation_bytes_per_second": 297492.0948468153,
+    "upload_bytes_per_second": 5957009.886437588,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 63.78493662026223,
+    "estimated_full_shards_per_hour": 63.834878292592464,
     "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": 0.9681096090007486,
-      "bytes": 2564629,
-      "pages": 35,
-      "shards": 2
-    },
     {
       "seconds": 1.0954296749987407,
       "bytes": 10402230,
@@ -562,6 +556,12 @@
       "bytes": 11876824,
       "pages": 39,
       "shards": 4
+    },
+    {
+      "seconds": 1.0809610530013742,
+      "bytes": 6077185,
+      "pages": 27,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790519810984249790-cb9c9513/train.jsonl

@@ -153,3 +153,4 @@
 {"step": 130300, "loss": 0.6134939007461071, "lr": 0.0003, "grad_norm": 0.14417946800087983, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 19068832999, "tokens_per_second": 138295.00993779086, "data_wait_s": 9.932000466505997e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 130400, "loss": 0.6093156430870295, "lr": 0.0003, "grad_norm": 0.12806106978875761, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 19083296183, "tokens_per_second": 147766.4906708138, "data_wait_s": 1.3663999197888188e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 130500, "loss": 0.5987209919840097, "lr": 0.0003, "grad_norm": 0.14051451020549627, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 19097971148, "tokens_per_second": 150032.96474808763, "data_wait_s": 1.0470999768585898e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 130600, "loss": 0.5643460601568222, "lr": 0.0003, "grad_norm": 0.13814585154836584, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 19112736787, "tokens_per_second": 150510.5215499491, "data_wait_s": 1.17090003186604e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790519810.cad0b19a-59f2-43f0-8fb4-2a94097843e2-tjh7c.1336.0.1790519810984249790-cb9c9513

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:eb0fc8cc151e8318e5b5d881cda55941a26198b693db73ce76acc31d7b86f212
-size 330804
+oid sha256:d8149d9d82fa2a2ca038d7eddfeb08ce481fbc0a396d6438c28bc25ef4f5f150
+size 332922